Part II - Driving Datafication

3 Data-Driven Platform as Service Classification, Market Access, and Digital Sovereignty

3.1 Introduction

At the platform level, our daily lives are increasingly transformed into structured data flows. Looking back to the negotiating history of the WTO, however, today’s platform economy was an unknown market phenomenon when the WTO was established in the early 1990s. At the time of the Uruguay Round negotiations, Mode 1 (cross-border) trade was considered insignificant or even “irrelevant” to most of the services sectors.¹ Such an understanding was reflected in the distribution of the commitments by mode of supply under the GATS. Some 30 percent of all members’ market access commitments on Mode 1 services trade are unqualified, that is, without limitations, as compared to only 15 percent of the commitments on Mode 3 (commercial presence).² Furthermore, the level of obligations for Mode 1 services trade does not differ in an obvious way between developed and developing countries.³ Evidently, the GATS negotiators could not have been aware of the future existence and influences of digital platforms. Those pro-liberalization market access commitments on Mode 1 were based on the brick-and-mortar business models in the pre-Google days.

Technological innovations have brought about exponential growth in data generation, analysis, and use. This raises the question of whether the GATS market access commitments, which were made decades ago, remain tenable in this datafied world.⁴ From geopolitical perspectives, as discussed in Chapter 2, issues surrounding national security now represent the most challenging battlefield facing international economic legal order. Digital transformation calls into question whether existing market access commitments, which are the result of “bargains” struck in a “previous technological era,” can be reasonably sustained,⁵ especially when considering the national security risks addressed above.⁶ One notable example is when the Trump administration in 2020 decided to bar people in the US from downloading TikTok and WeChat, two social media platforms owned by Chinese companies.⁷ China maintained that the services provided by the two platforms are covered by the US’ GATS Schedule of Commitments on advertising services, computer-related services, telecommunications, audio-video services, and entertainment services.⁸ The dispute over TikTok and WeChat demonstrates the key problems this chapter attempts to address: How are digital platforms covered by the GATS market access obligations? Are trade commitments “resilient” enough to outlive “technological generations?”

Even for services sectors that are less security-sensitive, classification nevertheless matters, as it decides to what extent GATS applies. To illustrate, there are two different groups of obligation under the GATS – “general obligations,” such as most-favored-nation treatment and transparency, which are applied across all sectors, and “specific commitments,” such as market access and national treatment, which are only applied to those sectors a member inscribes in its Schedule of Commitments. The flexible nature of the latter, as stressed by GATS drafters, was designed to allow each member to adjust the market entry conditions pursuant to its sector-specific policy concerns and objectives.⁹ By scheduling the specific commitments, which have been individually and uniquely exercised by each member “on their own pace,”¹⁰ a member is legally obliged to open a services market pursuant to its GATS Schedule.¹¹ In other words, market access applies under the GATS on a sector-by-sector basis. Sectoral classification, therefore, is the basis for identifying the scope of market access commitments under the GATS architecture.

Specific commitments have been scheduled based upon classification guidance W/120, prepared by the GATT Secretariat during the Uruguay Round negotiations. W/120 primarily relied upon the United Nations’ Provisional Central Product Classification (CPC), dating back to 1991.¹² It goes without saying that services sectors have undergone significant changes since then. Today’s commercial realities mean that W/120 classifications have become inadequate, and their correspondence with the CPC is out of date, leading to unreliable segmentations. More and more digital-related services identified in the W/120 and the CPC, such as facsimile services (7521**, 7529**), no longer make modern business sense. At the same time, more and more “new” services are not explicitly covered by the W/120 and the CPC with legal certainty.¹³ At the heart of the issue is the scheduling logic of the GATS architecture. The GATS was introduced as a positive-list agreement, in which there is no market access for services trade unless it has been positively inscribed in a member’s GATS Schedule.¹⁴ Such a positive-list architecture creates problems for any digitalized service that currently exists in the markets but was not explicitly “described” in the CPC.¹⁵ In this regard, much ink has been spilled surrounding whether search engines like Google are covered by GATS commitments, since China blocked Google.com and redirected traffic to local search engines. Are Google services already accommodated by the existing CPC? Commentators have argued both for and against with respect to the question of whether the W/120 and CPC subsectors “online information retrieval,” “value-added telecommunications service,” “data processing services,” “online hosting and publication services,” “advertising services,” and “database services” include search engines such as Google.¹⁶ In any event, Google cannot perfectly satisfy any definitions or conditions described in the existing CPC system, and as a result, classifying Google in either subsector seems illogical from a legal perspective. Likewise, how might chat applications such as WhatsApp best be defined? How might virtual meeting services such as Zoom be classified? Similar questions can be raised regarding the classification of endless lists of data-driven activities.

3.2 The “Renewal” of Trade Commitments

3.2.1 Services Digitalization: Technological Neutrality

Market access to data-related services has been the primary litigated area under the GATS. The first WTO ruling concerning e-commerce was on US restrictions over Internet gambling services. Antigua and Barbuda initiated a WTO dispute settlement case against the US,¹⁷ claiming that US Internet gambling restrictions at both the federal and state levels violated its market access commitments to “Entertainment Services” under the GATS. This dispute concerned various US domestic measures relating to gambling and betting services.¹⁸ The complaining party pointed out that the US measures in dispute constitute a “total prohibition on the cross-border supply of gambling and betting services.”¹⁹ It further claimed that the prohibition violated Article XVI:1 of the GATS because, despite having scheduled a “full market access” commitment for the cross-border supply (i.e., Mode 1) of gambling and betting services, the US “maintains and enforces measures prohibiting the cross-border supply” of those services.²⁰

In its submission, the US government was of the view that “new technologies, including high-speed telecommunications and the Internet” have brought about explosive growth in online gambling, and such a dramatic increase has raised serious regulatory and law enforcement concerns in the country.²¹ The US stressed that throughout its history, the country has consistently imposed tight regulations on the remote supply of gambling.²² According to the US submission, gambling has been regulated in the US back to the earliest years of the Colonial era,²³ and the country has expanded the relevant regulatory regime for the remote supply of gambling so that it addresses modern threats and, in particular, criminal activities on the Internet.²⁴ To illustrate the regulatory characteristics of Internet gambling, the US elaborated at length regarding how, compared with nonremote gambling services (i.e., traditional casinos), Internet gambling (i.e., virtual casinos) poses greater threats of organized crime, money laundering, fraud, and youth gambling.²⁵ The US therefore claimed that a proper interpretation of its GATS Schedule would show that it had never opened its market to online gambling.

WTO jurisprudence makes clear that “a member’s intent is not relevant” in determining whether the member has a commitment with respect to digital-enabled delivery.²⁶ Instead, under WTO law, the only relevant issue is whether the responding member in this particular dispute has explicitly excluded electronic means of service delivery from the market access commitments in its GATS Schedule.²⁷ The panel in U.S. – Gambling, by citing the “Progress Report of the Work Programme on Electronic Commerce,”²⁸ indicated that “[it] was the general view that the GATS is technologically neutral in the sense that it does not contain any provisions that distinguish between the different technological means through which a service may be supplied.”²⁹ Noting the principle of “technological neutrality,” which, according to the panel, “seems to be largely shared among WTO members,” the panel stressed that “where market access and national treatment commitments exist, they encompass the delivery of the service through electronic means.”³⁰ In short, by pointing out that “the GATS does not limit the various technologically possible means of delivery under Mode 1,”³¹ the panel considered transactions via electronic media one of the “inherent means of delivery” of Mode 1 trade. Thus, the panel concluded that a market access commitment for Mode 1 implies the right of foreign companies to supply services through all means of delivery, including the Internet.³²

3.2.2 Services Platformization: Evolutionary Interpretation

The case of China – Publications and Audiovisual Products is another compelling example of how digital technologies have disrupted the traditional understanding of market access commitments.³³ In its GATS Schedule, China opened its market to “sound recording distribution services”³⁴ and listed no national treatment limitations under Mode 3 (commercial presence) for Chinese-foreign contractual joint ventures. Given these commitments, Chinese-foreign contractual joint ventures, including the majority of foreign-owned joint ventures, should enjoy national treatment in terms of sound recording distribution. The Chinese domestic legal framework,³⁵ however, limits the ability of foreign-invested enterprises to engage in the distribution of sound recordings by prohibiting these enterprises from “electronically distributing” their music services via online platforms such as iTunes. The US therefore claimed that China’s measures were inconsistent with the GATS obligations.³⁶

China in turn asserted that online music platforms are not covered by China’s GATS market access commitments.³⁷ China argued that the music industry landscape has been undergoing major structural changes since its WTO accession negotiations. At the time of the negotiation of China’s GATS commitments, the legal framework governing the recorded music market exclusively addressed the distribution of sound recordings in their traditional, hard copy format.³⁸ China submitted several pieces of evidence to support its position that online music services did not constitute an established business operating within the Chinese legal framework during its GATS negotiations, and that China was not aware music was being electronically distributed when it joined the WTO.³⁹ According to China, the first online music service platforms in China were launched in the early 2000s. In other words, such digital platforms are a new phenomenon that did not exist at the time of China’s WTO accession.⁴⁰ China therefore claimed that online music platforms, which were not offered for liberalization at the time of its WTO accession, “cannot be committed post hoc through the dispute settlement process.”⁴¹

On this issue, the principle of technological neutrality was central to the US’ arguments against China.⁴² The US relied on the principle to point out that the differences between physical and digital distribution are not relevant to the interpretation of the scope of a GATS commitment unless specified in a member’s schedule.⁴³ In the view of the US, the electronic distribution of services merely constitutes “a new means of delivery for an existing service,” and the GATS is sufficiently dynamic to cover new technological innovations “affecting the delivery of services.”⁴⁴ The US rebutted that China’s position, if accepted, would result in “an unworkable outcome,”⁴⁵ simply because in that case, GATS commitments “must be renegotiated each time a new technology results in a new means of supplying a service.”⁴⁶ The US stressed that by this logic, WTO members could invoke such reasoning to “evade [market access] services commitments” whenever a new form of service delivery technology was developed.⁴⁷

The most important implication of this dispute is the Appellate Body’s ambiguous stance on the (in)significance of “technical possibility” and the (ir)relevance of “commercial reality” – namely, the state of technology and the market that existed at the time of the treaty negotiations. The Appellate Body emphasized that, at least in this dispute, the technical and commercial factual situations are not the central considerations.⁴⁸ In this regard, the Appellate Body stated that GATS Schedules constitute a part of the multilateral trade agreements, with “continuing obligations” entered into by WTO members “for an indefinite period of time.”⁴⁹ The Appellate Body drew attention to the treaty interpretation approach taken in U.S. – Shrimp,⁵⁰ where the term “exhaustible natural resources” in Article XX(g) of the GATT was read “in the light of contemporary concerns of the community of nations about the protection and conservation of the environment.”⁵¹ To summarize, under the concept of evolutionary interpretation, the Appellate Body in U.S. – Shrimp reasoned that the GATT was crafted more than fifty years ago, and the generic term “natural resources” in Article XX(g) is not “static” in terms of its content or reference, but is, rather, “evolutionary.”⁵² Likewise, the Applegate Body in China – Publications and Audiovisual Products concluded that the Chinese commitments in the dispute are “generic terms” whose content may “change over time,”⁵³ namely, from physical to digital.⁵⁴

3.3 Classifying Data-Driven Platforms

3.3.1 Rapidly Changing Markets: Technologically Future Proof?

Answering the question raised earlier – whether international trade commitments are sufficiently resilient to outlive technological generations, in particular, from physical to digital – the position of the GATS Council on Services is that the GATS applies even as technology changes a service’s delivery method.⁵⁵ Therefore, “much of e-commerce falls within the GATS’ scope,” and “GATS obligations cover measures affecting the electronic delivery of services.”⁵⁶ As discussed above, WTO case law has further confirmed that GATS disciplines and obligations extend to services supplied electronically. The reach of existing GATS commitments applies to digital means of service delivery that have emerged since the GATS was concluded in 1994. WTO members need not renegotiate their market access commitments “in the face of ever-changing technology.”⁵⁷

If we follow the logic of the Appellate Body in China – Publications and Audiovisual Products, when a term in a WTO member’s market access schedule is sufficiently “generic,” the commitment would be a “timeless” WTO obligation; on the other hand, when the term is sufficiently specific, the commitment would be a “time-bound” obligation that applies to a specific “technological status quo.” Under such an interpretive approach, virtually all GATS-specific commitments are made in a “sufficiently generic” way and are therefore open-ended and able to accommodate any possible future means of delivery. For example, Taiwan opened its “Cellular Mobile Phone Services” in its GATS Schedule when it joined the WTO in 2002. What the Taiwanese trade negotiators had in mind at the time of the trade negotiations was the state-of-the-art – namely, 2G – technology. The negotiators were of course unaware of how the powerful 5G technology of today would enable the IoT and AI, as well as how these 5G-enabled applications would change the way we live and work. However, since Taiwan did not specify “2G” in its GATS Schedule, the market access concessions of “Cellular Mobile Phone Services” would automatically evolve into the “new” context as time passes – from 2G to 3G, 4G, 5G … and xG. At the core of the issue are the “temporal variations” in human language. Our language meanings are under “constant flux” according to social, cultural, and technological contexts.⁵⁸ Should we interpret a GATS market access commitment using the “ordinary meaning” at the time of its treaty conclusion (i.e., historical language) or the “ordinary meaning” at the time of interpretation (i.e., modern language)?⁵⁹

In China – Publications and Audiovisual Products, China repeatedly stressed in its written submissions that the only viable way to address the issue of technological and social evolution should be via trade (re)negotiations, rather than via the “fast track” of WTO dispute settlement procedures.⁶⁰ If a member wants online services to be liberalized, then it must pursue such liberalization by undertaking new negotiations with other members.⁶¹ In China’s view, the WTO tribunals should refrain from anticipating the results of future trade negotiations.⁶² The Chinese government pointed out that unduly extending the scope of members’ existing GATS commitments would be antithetical to the principle of “progressive liberalization,” as reflected in both the GATS Preamble and Article XIX, which indicates that the GATS is aimed at establishing a multilateral framework for the expansion of services trade under conditions of transparency and progressive liberalization.⁶³ According to China, such a principle shapes the structure of the GATS, which allows members to undertake specific commitments through successive rounds of multilateral negotiations with a view toward incrementally liberalizing their services markets rather than doing so immediately and completely at the time of their WTO accession.⁶⁴ China therefore asserted that the same principle requires the panel to base its analysis of the relevant terms in a member’s GATS Schedule on their meaning at the time of trade negotiations, so as to prevent the extension of the scope of market access commitments based on “temporal variations in language.”⁶⁵

In light of the Appellate Body’s evolutionary approach to interpretation, key questions remain: Is the GATS “sufficiently dynamic,” as framed by the US in China – Publications and Audiovisual Products, to cover every new technological innovation affecting trade in services?⁶⁶ Digital technologies and data markets are changing rapidly and disruptively. Technically, how can we distinguish between “new service supply modalities” and “new services?” Does it make economic and legal sense that the former is already covered by the existing market access commitments, while market access for the latter is subject to future trade negotiations? Relying on concepts such as “technological neutrality” or “evolutionary interpretation” to expand the scope of market access commitments under the GATS, if applied to an extreme, may mean that international trade commitments are future-proof and are therefore “auto-renewed” across technological generations.

3.3.2 Innovative Business Models: Case-by-Case Approach?

Classification with definition matters in every legal instrument – it decides the scope of each legal obligation, and it defines the boundary of each regulation. To a large degree, technological uncertainty causes equal challenges to both international trade law and domestic regulation in this regard. After all, “treating like services alike” is a commonly shared regulatory principle. In the context of domestic regulation, classification issues arise along every step of digitalization and datafication. For example, should a “traditional service” and a “platform-based new service” be classified in the same manner for regulatory purposes? City authorities should enforce existing taxi regulations over Uber if it is classified as a transport service, like the traditional taxi. Similarly, Airbnb hosts should be required to obtain licenses if they are classified as real estate agents. For sectoral regulators, the decision on whether the two types of services – one in physical form and the other online – fall within the same regulatory category is dependent upon the characteristics of the services at issue. The central question always surrounds whether the digital platform is “similar to” or “different from” its off-line analogues.⁶⁷

The approaches taken by the European Court of Justice (ECJ) serve as an interesting reference here.⁶⁸ On the question of whether Uber should be classified as a “transport services supplier” and should therefore be required to seek prior regulatory approval, the ECJ decided that the services offered by Uber should not be regarded as an “Information Society Service” under the EU’s E-Commerce Directive. Rather, Uber should be legally equated with traditional transport services (i.e., taxi services) that must be regulated.⁶⁹ However, the ECJ ruled that Airbnb must be classified as an “Information Society Service” as defined in the E-Commerce Directive. In other words, the ECJ held that Airbnb and Uber should be treated differently.⁷⁰ According to the ECJ rulings, the commercial offering provided by Uber is in itself “more than an intermediary service.”⁷¹ Overall, Uber’s activities should be classified as “intermediation services forming an integral part of an overall service,” and “the main component of which is a transport service.”⁷² In this regard, Uber must comply with the domestic regulations of each EU member state pertaining to “transport services.”⁷³ However, the ECJ stressed that a similarly decisive influence of Airbnb, in terms of its power over the transactions of the accommodation services, could not be identified.⁷⁴ The ECJ thus classified the intermediation service offered by Airbnb as an “information society service” that is entitled to the benefits of free movement.⁷⁵ In short, different legal classifications trigger the implementation of different sets of rules.

In the above ECJ rulings, the different treatment of the two data-driven platforms is primarily attributable to two factors: first, the ability of users to successfully operate without the platform, and second, the ability of the platforms to control transactions. According to the ECJ’s assessment, Airbnb’s business model is not comparable to that of Uber, in the sense that “Airbnb’s intermediation service is in no way indispensable to the provision of accommodation services.”⁷⁶ In addition, Uber had exercised “decisive influence” over conditions that are “economically significant aspects of the service,” while Airbnb had not.⁷⁷ In brief, the technical features and business models of a digital platform are complex and involve a great number of variables.⁷⁸ The ECJ approaches to classifying the commercial nature of a platform are dynamic and pragmatic. Various conclusions can be drawn as to whether a specific platform qualifies as an “information service.” The nature of a platform is therefore legally unsettled. In the wake of the ECJ judgments, it would seem that the answer to the question of how to determine the exact nature of the services provided by platforms is dependent upon a case-by-case assessment of the characteristics of a particular platform. This, however, begs the question of relevant criteria, as well as their weighting.

The ECJ’s practice of platform classification demonstrates that the WTO is not alone in this “digital dilemma.” On the one hand, legal certainty and predictability are valuable attributes in any legal system, and consistency in the application of rules is an important source of legitimacy for any dispute resolution mechanism. It remains difficult to predict how a digital platform will be classified by the ECJ, thereby rendering such platforms subject to either a “light touch” or a “heavy hand” in terms of regulation. This case-by-case uncertainty may invite endless disputes. On the other hand, in light of the unpredictable nature and incredible pace of technology, judicial discretion is becoming a “necessary evil.” Judicial activism inheres in the incompleteness of rules. If the definition is sufficiently precise, there is less room for judicial judgments. However, in situations where a classification system does not provide clear-cut definitions, there are opportunities for the tribunals to operate in a reactive manner and engage in dynamic interpretation. Indeed, the classification systems under both the GATS Schedules and the EU E-Commerce Directive are outdated.⁷⁹ The WTO and the ECJ will need to continue to classify digital platforms in a judicially active way if the relevant definitions in the legal instruments are disconnected from rapidly changing technologies.

In this context, the EU’s Digital Services Act (DSA),⁸⁰ which aims to upgrade the more than twenty-year-old EU E-Commerce Directive, applies to a broad category of online players, including “mere conduit” services, “caching” services, hosting services, online platforms, very large online platforms (VLOPs), and very large online search engines.⁸¹ Among other terms, “online platform” is defined in Article 3 (i) of the DSA as follows:

… a hosting service that, at the request of a recipient of the service, stores and disseminates information to the public, unless that activity is a minor and purely ancillary feature of another service or a minor functionality of the principal service and, for objective and technical reasons, cannot be used without that other service, and the integration of the feature or functionality into the other service is not a means to circumvent the applicability of this Regulation…⁸²

It takes time to determine if such a broad definition can solve the practical issue of legal uncertainty present in the EU E-Commerce Directive. At first glance, the taxonomies offered by the DSA, to a certain extent, overlap and create uncertainty in determining if a regulation is applicable to a particular case. Admittedly, there is no “one-size-fits-all definition” of a digital platform.⁸³ It is necessary to create multiple taxonomies, as the DSA does, to classify the different kinds of data-driven activities depending on their primary functions, the actors involved, the ways in which they exploit data, the sources of revenue, and the level of control they exercise over users’ activities.⁸⁴ Logically, the same digital platform may also be classified into multiple categories simply because of the multiple functions it features.⁸⁵ In any event, the DSA intends to cover very broad and diversified sets of services offered on the Internet. The definitions therein are functionally constructed, allowing regulators to map the business practices to policy issues. In conclusion, will the approach taken by the DSA end the legal uncertainty? Probably not. However, the approach offers a high degree of generality, which is key in reducing the risk of regulatory disconnection.

3.4 Future Market Access for Data-Driven Platforms

3.4.1 The WTO: Reforms and Renegotiations Needed

Turning back to the international trading regime, what lessons and reflections can we draw from the EU’s regulatory experience? This book contends that trade negotiators in this age of datafication must be aware of the impossibility of drafting trade rules that speak directly to every situation and must also recognize the need for constant adaptation to technological advancements. By the same token, some sort of functional approach to market access commitments must be incorporated into trade negotiations. When opening a market to an innovative sector, international trade agreements should employ flexible language so as to allow reasonable room for technological changes, and to prevent market access commitments from becoming quickly overtaken by subsequent events. One example is the functionally simplistic approach to the telecom services sector classification proposed by the EU. Inscribing a simple entry, “any service consisting of the transmission and reception of signals by any electromagnetic means,” as suggested by the EU, would seem to end classification uncertainty in the telecom sector.⁸⁶ At the end of the day, it is not possible to have a precise and clear-cut definition for market access commitments in the data-driven sectors.

Looking toward the future, substantial reforms and renegotiation are needed in the WTO, assuming it is politically feasible someday. Although both U.S. – Gambling and China – Publications and Audiovisual Products raised the question of how a schedule of commitments is to be interpreted to include a new service not in existence at the time of negotiations,⁸⁷ WTO case law is not entirely clear regarding in what situation and to what extent the “state of technology” that existed at the time of the market access negotiations is relevant in determining the scope of the commitments. On the one hand, digital technology is organic and technologically innovative, and we must read the existing trade commitments in a dynamic and evolutionary way. On the other hand, judicial interpretation may not be the most appropriate method of clarifying whether the W/120 and the CPC cover certain digital trade-related services. What is covered and what is not covered by members’ GATS Schedules is an issue that should be (re)negotiated by WTO members, not by litigation.⁸⁸

In conclusion, there is an urgent need for the W/120 scheme to be replaced or revised.⁸⁹ Negotiating proposals, including technical discussions regarding how to classify platforms for cloud computing services,⁹⁰ have been submitted to address necessary adjustments, and to ensure that services classification and scheduling can accommodate modern commercial and technological developments. The GATS classification/scheduling system must be reformed in light of innovative advances and trends toward a datafied world.⁹¹ From iTunes and Google to TikTok and WeChat, the legal certainty of market access for digital trade cannot be ensured until classifications can be unambiguously defined. Moreover, renegotiations are needed to curb the notion that the GATS Schedules are “future proof” – covering all platform-based, data-driven e-commerce.⁹² At the time of this writing, it remains to be seen to what extent the WTO JSI on E-commerce will be able to clarify issues surrounding the market access of digital platforms.

3.4.2 The FTAs: Data Flows as both the Means and the End

Under the FTAs concluded in more recent years, the E-Commerce/Digital Trade Chapters can be seen as a flipped governing approach that creates “de facto market access” for cross-border services trade. Essentially, the concept underlying the E-Commerce/Digital Trade Chapters was that rather than the “services,” the “data flows used to supply the services” require legal protections under an international trade agreement.⁹³ Under the CPTPP’s E-Commerce Chapter and the USMCA’s Digital Trade Chapter, as examples, parties have agreed to allow the “cross-border transfer of information … by electronic means” when the activity is for the conduct of the business of a covered person under the respective FTA.⁹⁴ Therefore, by securing the free flow of data, the E-Commerce/Digital Trade Chapters themselves open the market for data-driven sectors.

At the crux of the matter is the relationship between Cross-Border Trade in Services (e.g., Chapter 10 of the CPTPP, Chapter 15 of the USMCA), on the one hand, and E-Commerce/Digital Trade (e.g., Chapter 14 of the CPTPP, Chapter 19 of the USMCA), on the other hand. To what extent does the latter expand market access for the former? Would the latter be construed as overriding the nonconforming market access measures reserved by parties under the former? Or is it the other way around? How might the regulatory disciplines surrounding data transfer provided in the latter affect the market access commitments of the former? Indeed, when read together, there is an overlap in the scope of application of the Cross-Border Trade in Services Chapter and the E-Commerce/Digital Trade Chapter. For example, a domestic measure restricting health-related data transfer directly involves the obligations on the free flow of data in the E-Commerce/Digital Trade Chapter, and at the same time amounts to a trade barrier to market access for cross-border health services. Therefore, the rules from both chapters simultaneously apply when a domestic measure restricts cross-border data flows. To summarize, when data flow is considered a “means” for the supply of a service, it falls within the scope of the Cross-Border Trade in Services Chapter; when data flow itself is an “end,” it is regulated under the E-Commerce/Digital Trade Chapter.⁹⁵

Taking the CPTPP as an example, Cross-Border Trade in Services under the CPTPP, as defined in Article 10.1, contains three modes: cross-border supply (the same as Mode 1 under the GATS), consumption abroad (the same as Mode 2 under the GATS), and the presence of natural persons (the same as Mode 4 under the GATS). Chapter 10 mirrors similar core obligations in the GATS, including market access. The language of Article 10.5 (Market Access) is to a large extent borrowed from GATS Article XVI, which provides that no party may adopt quantitative limitations on the supply of services. In general, the CPTPP parties take a broad approach to cross-border trade in services, with services covered unless specifically listed in a party’s schedule of nonconforming measures. Such a negative-list approach to services market access has in recent years been adopted by a large number of FTAs.⁹⁶ According to Article 10.7, the market access obligations under Article 10.5 shall not apply to any measure a party maintains with respect to sectors or activities, as excluded by that party in its Schedule to Annex II.⁹⁷ In other words, parties commit to opening the services market except in sectors where the “non-conforming measures” are listed in parties’ schedules. Such a modality allows CPTPP parties to reserve the right to maintain or adopt restrictions for a particular sector, namely, to apply any measures that are inconsistent with certain obligations.⁹⁸ In short, a party-specific list of reservations enables a party to have full discretion to maintain existing nonconforming domestic regulations or to adopt new regulations without any legal consequences under the CPTPP.

It is important to note that to the extent Chapters 10 (Cross-Border Trade in Services) and 14 (Electronic Commerce) overlap, the former trumps the latter. To illustrate, there are two provisions dealing with the Chapters 10/14 overlap. First, Article 14.2(4) of the CPTPP states that measures affecting the cross-border supply of a service that is delivered electronically are subject to the obligations, nonconforming measures, and exceptions applicable to Chapter 10.⁹⁹ Second, Articles 14.2(5) and 14.2(6) of the CPTPP further clarify that any market access-related obligations in Chapter 14 – in particular, Article 14.11 (Cross-Border Transfer of Information by Electronic Means) and Article 14.13 (Location of Computing Facilities) – are subject to the relevant provisions, exceptions, and nonconforming measures of Chapter 10.¹⁰⁰ For example, in its Annex II of the CPTPP, Australia listed nonconforming measures concerning the market access of distribution services. More specifically, it reserves the right “to adopt or maintain any measure with respect to wholesale and retail trade services of tobacco products, alcoholic beverages or firearms.”¹⁰¹ Assuming that Australia maintains a ban on online tobacco advertising, the marketing restrictions on digital platforms would fall within the Chapters 10/14 overlap. In that case, Article 14.11 (Cross-Border Transfer of Information by Electronic Means) would not apply because Australia did not forgo the right to introduce any market access-impairing tobacco measures in the future.

Much ink has been spilled by commentators regarding how the E-Commerce/Digital Trade Chapter can facilitate the free flow of data and thereby open the digital market. A strict focus on the E-Commerce/Digital Trade Chapter, however, might be too narrow to grasp the full picture.¹⁰² As illustrated above, the Chapter on Cross-Border Trade in Services covers the data flows required to supply the specified services. At the same time, the Chapter on E-Commerce/Digital Trade, although it primarily concerns the regulatory disciplines on data flows, contains market access-related provisions that aim to protect data movement against cross-border data transfer restrictions.¹⁰³ When the obligations of the two chapters are in conflict, the obligations in the Cross-Border Trade in Services Chapter prevail.

Turning back to data-driven platforms, the overlapping provisions in the two chapters of the FTAs complicate their global market access. The commitments on “cross-border data flows” would not apply when parties do not allow “cross-border trade in services.” In this regard, listing nonconforming measures under the Cross-Border Trade in Services Chapter would represent the most critical step in reserving the right to regulate the digital economy.¹⁰⁴ In the case of the CPTPP, several parties have carefully safeguarded the boundaries of the services markets. New Zealand, for example, has circumscribed the scope of its CPTPP market access commitments on cross-border trade in services to exclusively match its GATS commitments.¹⁰⁵ Japan, as another striking example, has listed nonconforming market access measures for “all unrecognized or technically unfeasible services” as follows:

Cross-Border Trade in Services:

Japan reserves the right to adopt or maintain any measure relating to services other than those recognized or other than those that should have been recognized by the Government of Japan owing to the circumstances at the date of entry into force of this Agreement. (emphasis added) … Japan reserves the right to adopt or maintain any measure relating to the supply of services in any mode of supply in which those services were not technically feasible at the date of entry into force of this Agreement (emphasis added).¹⁰⁶

Obviously, the Japanese inscription was inspired by the China – Publications and Audiovisual Products dispute. Considering that the negative-list approach adopted by the CPTPP would allow “new services” to be automatically included in the trade agreement and therefore entitle such services to be traded across borders in the CPTPP markets, Article 10.7 serves as a safety valve, allowing Japan to block not only “all unrecognized or technically unfeasible services,” but also the means to deliver such services. Put simply, the inscription ensures that Japan does not risk committing to future services and associated delivery means that do not yet exist.

3.5 The Relationship between Market Access and Domestic Regulation

3.5.1 The Right to Regulate: Digital Sovereignty

Of course, market access obligations under international trade agreements do not prevent countries from adopting or implementing domestic regulations in pursuit of legitimate national policies. The GATS explicitly recognizes the right of members to pursue policy objectives through regulation, even in sectors where they have undertaken full commitments to market access.¹⁰⁷ Once a state opens the international trade market of a specific sector, there might be a new need for domestic regulatory interference in the market to accomplish economic objectives such as market competition, or noneconomic social goals such as consumer protection.¹⁰⁸ In a nutshell, domestic regulation is essential to achieve legitimate policy objectives intertwined with trade in services, because the markets do not always work perfectly and cannot always bring about socially desired results by themselves.¹⁰⁹

A sovereign’s right to regulate in the age of datafication has been conceptualized into the notion of “digital sovereignty.”¹¹⁰ In Europe, as the most striking illustration, internal political supports have been overwhelming in shaping policy approaches to enhance Europe’s “strategic autonomy” in the digital age.¹¹¹ The concept of “digital sovereignty” is thus becoming more and more common in the policy and academic literature, although there are divergent interpretations of it.¹¹² Various meanings have been attached to the concept. It is often used to denote “the ability of nation states to control the digital infrastructure on their territory and the data of their citizens.”¹¹³ The terminology, however, is increasingly used in a much broader context and almost always refers to the digital dimensions of a nation’s autonomy,¹¹⁴ including the ability of states to take action, both proactively and offensively, in the construction of digital infrastructure systems themselves, as well as the production, storage, processing, and analysis of the data of their citizens.¹¹⁵ We can say that the unprecedented economic and social influence of big tech has served as the catalyst for the concept of “digital sovereignty,” which is rooted in the need to safeguard regulatory authority in a data-driven world.

To summarize, there are growing concerns that individuals are losing control over their data, and that regulators are losing power to shape the legal framework of the digital sphere. Against this context, digital sovereignty has become a common concept to describe a broader range of digital policies and justify related regulation designed to defend a state’s interests and values. Nonetheless, as the following chapters will continue to explore, domestic regulations are subject to disciplines under international economic law to safeguard the benefit of market access commitments, ensure that existing market entry is not nullified by unnecessary regulatory requirements, and reduce the likelihood of biased implementation of regulations.¹¹⁶ In the context of digital trade, governments are free to regulate platforms but must do so in a way that does not constitute unnecessary barriers to trade. After all, the ability of services suppliers to engage in international trade is particularly affected by domestic regulation, including measures related to licensing and qualification requirements and procedures, as well as technical standards. Regulatory protectionism may impose disadvantages on platform services in a manner that is not necessary for the fulfilment of a genuine public policy objective. Domestic regulation adopted without adequate prior notification may prevent foreign services suppliers from competing on a level playing field. Excessive standards may also significantly distort international trade. Therefore, international trade rules should be in place to decide when and under what conditions legitimate policy objectives prevail over digital trade interests. As argued by Delimatsis, market access of the services trade “is only useful when it goes hand in hand with the promotion of sound domestic regulation.”¹¹⁷

Domestic regulations that are not trade restrictions per se can still have unnecessary trade-restrictive effects. Article VI:4/5 of the GATS itself, however, is too weak to promote best regulatory approaches in an international context.¹¹⁸ For this reason, Article VI:4 of the GATS mandates the development of disciplines to ensure that domestic measures pertaining to licensing and qualification requirements and procedures and technical standards do not constitute unnecessary barriers to the services trade. The conclusion of the WTO JSI on Domestic Regulation, which sets out common rules requiring good regulatory practices, was a significant step in the multilateral trading regime to ensure that “existing liberalization commitments are not nullified by opaque and complex authorization procedures.”¹¹⁹ At the same time, more and more FTAs contain GATS-plus regulatory obligations, especially the trends to include a Regulatory Coherence/ Good Regulatory Practice Chapter in the FTA. This book will take a deep dive into this issue in Chapter 5.

3.5.2 The Quantity/Quality Dichotomy

Before we further address issues surrounding platform regulations in Chapters 4 and 5, a central, preliminary question lies in the relationship between market access and domestic regulation. Arguably, the structure of the GATS is based on a separation between domestic measures identified as “limitations on market access” (which fall under Article XVI)¹²⁰ and measures adopted to ensure the quality of the service (which fall under Article VI:4/5). Such a structural separation and distinction is based on the rationale that the regulatory intervention cannot be considered as “trade restrictiveness.” Pursuant to the Scheduling Guidelines (S/L/92),¹²¹ the market access column in a member’s schedule must be confined to specifying the “terms, limitations and conditions of market access” listed in Article XVI:2(a)-(f), which essentially refer to limitations on (a) the “number” of services suppliers; (b) the “value” of transactions or assets; (c) the “number” of service operations or quantity of output; (d) the “number” of natural persons; (e) the legal form of establishment, such as “joint venture” requirements; and (f) foreign “capital participation.” S/L/92 further clarifies that “approval procedures or licensing and qualification requirements” should not be scheduled under Article XVI “as long as they do not contain any of the limitations specified in Article XVI.”¹²² Regarding the distinction between Article XVI (market access) and Article VI:4/5 (domestic regulation), paragraph 11 of the S/L/92 states that the former refers to “maximum limitations,” while the latter applies to “minimum requirements.” A measure that does not conform to the latter “cannot be scheduled.”¹²³

In fact, during the Doha Round of negotiations, several members advocated for the “numerical target” as a complementary approach to the services market access negotiations that are “quantitative” in nature.¹²⁴ It is therefore reasonable to state that the negotiating modalities and the scheduling technology of GATS market access commitments are numerically oriented and quantity-driven. However, the dichotomy of quality vis-à-vis quantity may not always operate well in the real world. The quantity/quality distinction became blurred in WTO jurisprudence, and the problem of identifying the line between market access and domestic regulation is most acute in the broad interpretation of the concept of “zero quota.” In Mexico – Telecom,¹²⁵ Mexico submitted that the “permit requirement” establishes a “zero quota” for Mode 3 market access.¹²⁶ According to Mexico, the permit requirement is qualified by the paragraph indicating the following: “[t]he establishment and operation of commercial agencies is invariably subject to the relevant regulations.”¹²⁷ Mexico was of the view that nothing in its entry committed it to issuing the corresponding regulations, which is “equivalent to a zero quota.”¹²⁸ Mexico’s arguments point to the confusing boundary between Articles XVI (market access) and VI (domestic regulation) of the GATS. Unfortunately, the concept of “zero quota” since then has become rooted in WTO jurisprudence. The leading case in this issue is U.S. – Gambling. The US argued that regulations on online gambling services are not “quantitative limits” within the meaning of Article XVI:2, and that a commitment under XVI only prohibits “quantitative measures” included in XVI:2.¹²⁹ As claimed by the US, “there is no reason why a Member’s imposition of nationality-neutral limitations” could violate GATS Article XVI “so long as the particular measures in question do not take the form of numerical quotas or any other form prohibited by Article XVI:2.”¹³⁰ The Appellate Body in this particular issue ruled that the US measures in the dispute should be considered a limitation within the ambit of Article XVI:2(c) for the reason that it “totally prevents the services operations and output” through one or more or “all means of delivery that are included in Mode 1.”¹³¹ Such a ban, according to the panel and the Appellate Body, amounts to a “zero quota,” and this constitutes a quantitative limitation within the meaning of Article XVI:2(a) and (c).¹³²

As Pauwelyn pointed out,¹³³ the implication of this confusing ruling is the risk of equalizing the “zero quota effect” of a licensing or qualification requirement (that is a nonquantitative restriction) with a numerical quota (that is a real quantitative limitation). To be sure, the concept that market access and domestic regulation are “mutually exclusive” is overly simplified and at odds with administrative operations in the real world. The key, as commentators have suggested, is to make a distinction between the US regulations in U.S. – Gambling, that is, a blanket prohibition that bans gambling services suppliers to zero with respect to all cross-border trade, and other situations under which domestic regulation is of a “less drastic quantitative effect.”¹³⁴ To conclude, recognizing the “zero quota” interpretation will be potentially problematic for every domestic regulation that has the “effect” of restricting market access. Therefore, “zero quota” arguments should be used in very limited cases in which the quantitative impact of a licensing/qualification requirement is extreme.

3.5.3 “Market Access Commitments” Matter

Focusing on the rulings from the U.S. – Gambling case, to what extent can a WTO member regulate digital platforms assuming, arguendo, that such a service is covered by the member’s GATS Schedule of Commitments? If the standard of review set by U.S. – Gambling remains the precedent-setting case on the quantity/quality dichotomy,¹³⁵ how far can domestic regulation go without causing the “zero quota effect” and therefore falling within the scope of Article XVI of the GATS? The following two scenarios related to Teleland and Digiland – both of which are WTO members – demonstrate the relationship between market access and domestic regulation when it comes to platform regulations.

Scenario one: Teleland decided to prohibit transactions with Digiland-originated mobile apps, referring to them as significant threats to national security and privacy protection. Among the most popular of the banned apps was Let’sTalk, the messaging service owned by Digiland’s Superdigital Holding. The government overseeing Teleland cited national security concerns as the justification for the bans, flagging the data-collecting activities of Let’sTalk and the potential for the Digiland government to obtain Teleland citizens’ personal information. Let’sTalk was removed from the app stores and was no longer available for download in Teleland. In addition, the Ministry of Information Technology of Teleland issued a blocking order directing telecommunications companies, including the 5G operators and other ISPs, to block access to Let’sTalk.

Scenario two: Amid COVID-19, the considerable increase in demand for digital services, including e-education and e-health, led to an equivalent increase in data volumes in Teleland. At the same time, the social distancing policy dramatically boosted the demand for streaming entertainment content, which strained the telecommunications infrastructure in Teleland. Netfly, which is owned by Digiland’s Techmedia Holding, is the dominant subscription streaming TV service in Teleland. In order to alleviate network congestion, the telecom regulator of Teleland issued an order requiring ISPs to prioritize traffic associated with e-education and e-health over traffic from streaming TV services. The order effectively reduced Netfly’s network traffic in Teleland by 40 percent, which affected its quality of service and led to a substantial drop in subscribers.

Suppose that Teleland’s GATS Schedule of Specific Commitments includes full market access commitments in the relevant sectors on Modes 1 and 3. In these hypothetical cases, the critical question is whether the two measures in dispute – the blocking in scenario one and the throttling in scenario two – constitute market access limitations to cross-border supply pursuant to Article XVI:2 of the GATS. Following the approach of the Appellate Body, Digiland could argue that the ban on Let’sTalk, which amounts to a “zero quota,” constitutes a “limitation on the number of services suppliers in the form of numerical quotas” within the meaning of Article XVI:2(a). The issue in scenario two, on the other hand, is less straightforward. Although Teleland’s network management has reduced Netfly’s network traffic by 40 percent, it does not result in the “zero quota,” as the streaming TV service may still operate in Teleland.

To conclude, market access commitment matters. Once full market access commitments are made, the quantitative impact of a member’s domestic regulation, if it cannot be justified under general or security exceptions, should not be too extreme. Domestic regulation might be considered a market access limitation simply because it has a substantial “quantitative effect.” The broad reading of Article XVI of the GATS in WTO jurisprudence has to some extent intruded upon the regulatory autonomy of WTO members.

3.6 Conclusion

Digitalization, datafication, and platformization were unforeseen phenomena when WTO members made the GATS market access commitments. However, as evidenced by the U.S. – Gambling and China – Audiovisual Services disputes, these decades-old commitments have certainly played more than a marginal role in the story of the evolution of global datafication. Regardless of whether or not it is a “historical accident,” members have undertaken the obligations through the GATS market access commitments, effectively leaving the door open for big tech companies. In view of those market access commitments, international trade liberalization must be accompanied by the introduction of new domestic regulation to address the potential risks and harms. As explained by Cohen,¹³⁶ “the emergence of platform-based business models has reshaped … the consumption of goods and services.” Chapters 4 and 5 will continue to address how platformization may have caused problems – including information manipulation, data capitalism, and algorithm discrimination – that a state cannot easily police.

4 Data as Speech and Expression Trade Aspects of Media Content Regulation

4.1 Introduction

Digital media content regulation represents a striking aspect of the dilemmas facing governments in the age of datafication. For decades, the audiovisual sector has traditionally been heavily regulated.¹ Given the high social and cultural importance of the sector, sector-specific audiovisual measures usually include, among others, free speech safeguards, cultural diversification screen quotas, regulations concerning the protection of minors, content controls, must-carry rules, advertising restrictions, product placement rules, public service obligations, production subsidization rules, foreign ownership limitations, and taxation policies.² In conjunction with the trends toward datafication and platformization, debates surrounding media policy are now shifting to the ambit of digital content governance.

Given their power to datafy a social phenomenon, digital platforms have become important gateways – and the main gateway in most places of the world – through which the public accesses and disseminates media content. How, then, can we regulate the “media sector” in a digital world? At this moment, regulators all over the world are struggling with whether to impose more responsibilities on online platforms, so as to create a fairer and safer digital space where the “new media” on the Internet can fulfill its broader democratic and cultural functions. Moreover, a closely related technologically challenging issue is how to address the jurisdiction problem and ensure that “domestic” regulation of “global” digital platforms can be effectively enforced.

This chapter focuses on two angles – speech regulation and cultural policy – to explore the interplay between digital media content regulation and international economic law. First, when “data” is regarded as “speech,” the interface between domestic media regulation and international trade rules is transformed into a highly political or even sensational topic, as it involves the national constitutional discourse against disproportionately regulating “speech.” In many jurisdictions, speech must be narrowly regulated, and any speech regulation will inevitably involve complex policy questions surrounding to what extent it intrudes upon freedom of expression. That said, the proliferation of cyberbullying, hate speech, misinformation/disinformation,³ fake news, and other harmful or illegal content on social media has become a major societal concern, which has apparently been shown to impact public interests. To what extent should platforms be accountable for such “speech” generated by individuals? At the same time, how can we prevent excessive blocking of user-generated content (UGC) such as Twitter (now called X) tweets and Facebook posts? The central concern of this chapter is to explore how international trade agreements interact with domestic platform regulations that address these problems.

Second, digital media content is of cultural significance, and the “culture v. trade” debate reaches another level of controversy when “data” is regarded as “cultural expression.” Traditionally, the audiovisual sector – primarily television, film, and music – has been considered a venue reflecting the cultural and social characteristics of a nation and its people, as well as a means through which to present a nation’s unique identity to the rest of the world.⁴ Due to its cultural components,⁵ the audiovisual sector has long played an important role in shaping the identity of heritage, promoting linguistic diversity, and contributing to cultural diversity.⁶ Today, video streaming services (VSS) are fast becoming a substitute for traditional “television” and have brought about tremendous challenges to media regulators. Policy debates include whether and to what extent VSS such as Netflix should be subject to existing television regulation, and in particular, the screen quota requirement that has been the main tool in cultural policy to guarantee market share for local content output and thus protect local culture.⁷

From the aspects of international economic law, questions regarding how a state regulates speech and promotes culture also involve a broad set of issues that may be subject to trade negotiations or dispute settlement. When data is generated as the by-product of other services, it can be a mere digital footprint; but when data communicates to affect opinions and change others’ minds, it becomes speech and expression.⁸ On the path toward a datafied world, currently UGC and VSS have become targets of national regulations that intend to protect free speech and culturally diverse expression. At the same time, they are at the crux of international trade negotiations that aim to promote the free flow of data.

4.2 Regulating UGC: Trade Aspects of Speech Regulation

4.2.1 Social Media Platforms: “Digital Town Square?”

UGC refers to “media content that is produced by users of that medium.”⁹ Today, UGC is primarily associated with the text, comments, images, and videos users have created and directly uploaded to digital platforms. Given its focus on openness and decentralization, UGC is essentially a form of self-expression, social participation, and democracy engagement,¹⁰ which greatly differs from the “traditional” one-way media structure of broadcaster/publisher to audience/reader. Nonetheless, the risks associated with defamation, mis-/disinformation, and other harmful or illegal behavior are becoming more and more challenging, as social media is now a critical component of most people’s lives. The dynamic nature of UGC leads to the question of whether digital platforms, which act as “intermediaries” in facilitating user interaction, should be liable for UGC on their sites.

This question has been transformed into a matter of significant controversy in recent years, especially in the US, where social media was born and has been widely used in politics. The policy debate was fueled by Twitter’s suspension of former US President Trump’s account in response to his unsubstantiated tweet of election fraud. Social media’s decision to deny Trump a platform was certainly a move applauded by Trump’s political opponents but criticized by free speech advocates.¹¹ More recently, a Texas law prohibiting large social media companies from taking down UGC based on their political viewpoints went into effect after the 5th US Circuit Court of Appeals gave the Texas social media law a green light.¹² The legal battle, of course, is not over yet. The Supreme Court may still reject the Fifth Circuit’s decision and strike down the Texas social media law. In any event, what’s behind those sensational news headlines is the question of to what extent datafication-enabled content moderation on social media platforms counts as “censorship.”

Placing these issues in the political economy context, what do we want social media platforms – today’s global public square – to look like? Additionally, how have different approaches emerged in different countries? Should platform content moderation extend to “lawful but awful” UGC? If so, whose judgment holds on “awful content”? Elon Musk, in a statement about the acquisition of Twitter, described Twitter as “the digital town square, where matters vital to the future of humanity are debated.”¹³ Such an ambition to create a speech platform that allows unfettered free speech has been criticized as “a complete fantasy.”¹⁴ Social media platforms are important forums for speech, but an unregulated forum may discourage the free exchange of speech. As discussed below, Section 230 of the US Communications Decency Act (CDA 230) serves as the best example of such a dilemma.¹⁵

4.2.2 Speech-Platform Liability for UGC: The Dilemma of the US CDA 230

In most jurisdictions, “traditional media,” such as newspaper publishers and television stations, can be held liable for publishing or broadcasting harmful or illegal content. The emergence of UGC in the 1990s, however, disrupted the media landscape and brought about the question of whether digital services suppliers should be liable for third-party content on their services.¹⁶ In the US, where free speech enjoys relatively stronger constitutional protections in the world, one leading case in this regard is Stratton Oakmont v. Prodigy Services,¹⁷ in which the court ruled that Prodigy – a digital bulletin board services supplier – could be held liable for the “speech of their users.”¹⁸ In the view of the court, Prodigy had become a “publisher” because it voluntarily deleted some users’ offensive messages from its bulletin board. As a result, it was liable for other users’ defamatory postings on its service, which it had failed to take down.¹⁹ In this particular case, the fact that Prodigy moderated “some” offensive or indecent content made it a “publisher,” and therefore it could be held liable for “all” content, regardless of whether or not it moderated all of its content. In other words, those services that did not screen any content risked less liability than those that screened some or all content. Against this backdrop, the US Congress enacted the CDA 230 to protect “interactive computer services” (e.g., digital speech platforms) from defamation lawsuits over UGC.²⁰

The protection under the CDA 230 is twofold: Digital platforms cannot be held liable if they choose not to moderate UGC and, at the same time, digital platforms cannot be held liable if they choose to moderate UGC. To illustrate, the twofold protection can be summarized as follows: First, CDA 230 grants speech platforms such as Facebook (Meta) or Twitter immunity from liability that is based on their treatment, “as the publisher or speaker,” of any content posted by their users. As a result, when a user’s posts or tweets defame a third party, any defamation claim against the platforms would be barred by CDA 230.²¹ To a great extent, such an immunity “negates” the duty of platforms to actively monitor material on their sites and allows them to ignore hate speech, misinformation, and other dangerous or harmful content. Second, CDA 230 grants speech platforms immunity from filtering UGC that it finds “objectionable” without incurring liability for doing so. In other words, speech platforms have the freedom, through their “content moderation” process, to keep the content they see fit, as well as to remove the content they consider “unlawful conduct or harassment.”²² Users generally do not have any remedies if the platforms take down the content “in good faith.”²³

CDA 230 has incited immense controversy over the past quarter century. Many scholarly writings have explained how the US courts have stretched CDA 230’s plain language and granted (over)broad, if not absolute, immunity, which is not supported by the CDA’s congressional intent of 1996.²⁴ By pointing out that the “benefit” of CDA 230 has been extended to mobile applications through expansive judicial interpretation, commentators have criticized the US courts for failing to consider the “outer limits” of CDA 230.²⁵ On the politics front, not surprisingly, CDA 230 has been a hot potato within and beyond the US, and its future at the time of this writing remains in flux. The dilemma facing policymakers is evident, as documented below.

On the one hand, opponents of CDA 230 argue that the Act allows platforms to turn a blind eye to hate speech, mis-/disinformation, and harassment on their sites. In particular, those platforms that lack the incentive to combat such activities may become breeding grounds for harmful UGC and thus threaten public interests.²⁶ At the same time, for those platforms that actively engage in content moderation, the Act’s blanket immunity means that such platforms have broad discretion in determining whether certain speech is socially acceptable. As a result, some UGC may be prejudicially censored, including some controversial political speech that is constitutionally protected from government censorship. Critics of CDA 230 have pointed out that UGC is routinely subjected to arbitrary censorship by dominant platforms. These platforms may, when they wish, suppress users’ expressions or even completely oust a user²⁷ – literally playing the role of “judges of truth.” In fact, the scale of speech restrictions by large platforms is evident. Twitter, as an example, reported that from January to June of 2021, it removed 5,913,337 posts, among which 1,606,979 posts were deemed to constitute “hateful conduct.”²⁸ Meta, as another example, reported that in the first quarter of 2022, it took action on 15.1 million Facebook posts and 3.4 million Instagram posts considered to constitute “hate speech” – one of its many content moderation standards.²⁹ Specifically, Meta alone removed 205,556 “hate speech” posts per day.³⁰

On the other end, supporters of CDA 230 have labeled it “the law that built the Internet.”³¹ They have credited the Act for playing “an indispensable role in facilitating the growth of the world’s largest online platforms.”³² In their view, without such a “catalyst,” the digital industry would never have become what it is today. They also believe that the immunity has incentivized social media to moderate UCG, which has significantly contributed to the protection of the Internet from objectionable content. Without such a safe harbor, platforms would have under-moderated to avoid being punished for policing content.³³ Moreover, the costs of moderating UGC do not deter the growth of platforms because they are not required to police all content.

In any event, it is worth underscoring that CDA 230, together with other factors, makes the US a haven for social media. It is apparent that the Act has given a distinctive boost to digital platform services for UGC that impact the daily lives of most people. It goes without saying that big tech has repeatedly and vigorously defended CDA 230. Any efforts to narrow the broad immunity it provides will prompt significant opposition from those tech giants,³⁴ as evidenced by their recent strong resistance to the Texas social media law. The controversy surrounding CDA 230 is now reaching a fevered pitch, which raises the question of whether digital platforms can be immunized when they algorithmically “recommend” UGC. As Chapter 5 will further explore, when a platform’s algorithmic system promotes specific UGC, is such an action of algorithmic amplification still protected by CDA 230? Indeed, the operation of CDA 230 has gone far beyond its legislative intent in creating such immunity.

4.2.3 Exporting Speech Regulation: Immunity Abroad?

Of course, the fire from the battle over CDA 230 is not limited to the US, as large platforms would grab any opportunity throughout the world to advance their agenda of immunity protection. It is no secret that big tech firms have attempted to influence trade negotiations. From CPTPP to USMCA to IPEF, big tech companies have been advocating for digital trade rules that “overwhelmingly favor their interests.”³⁵ To further pursue their “digital trade agenda,” big tech firms have been lobbying,³⁶ perhaps through “behind-the-scenes access to the trade officials,” to push for trade rules that protect their interests.³⁷ If the allegation of Warren and other US lawmakers is true, the Office of the US Trade Representative (USTR) has been granting “insider status” to Amazon and Google lobbyists throughout the stages of international trade negotiations.

The “footprint” of CDA 230 has been extended to US-led international trade agreements, and in particular, the US–Japan Digital Trade Agreement and the USMCA.³⁸ Moreover, CDA 230-like immunity can be found in the consolidated negotiating text of the ongoing WTO JSI on E-Commerce.³⁹ More specifically, for example, Article 18(3) of the US–Japan Digital Trade Agreement mimics CDA 230(c), which protects digital platforms from liability for voluntarily removing harmful or objectionable content in good faith. Additionally, similar to CDA 230, the bilateral trade deal contains exceptions for enforcing intellectual property rights, criminal law, or other lawful orders of a law enforcement authority.⁴⁰ It should be noted, however, that the US and Japan have agreed to a side letter, in which the parties recognize that there are differences between their “respective legal systems governing the liability of interactive computer services suppliers”⁴¹ and have therefore agreed that Japan need not “change its existing legal system, including laws, regulations, and judicial decisions, governing the liability of interactive computer services suppliers,”⁴² in order to comply with Article 18 of the bilateral trade deal. As a result, the side letter has literally negated Japan’s treaty obligations to amend its platform liability law.

Just months after the signing of the US–Japan Digital Trade Agreement, the USTR continuously extended CDA 230 to its neighbors through Article 19.17 of the USMCA. In particular, Article 19.17(3), which is virtually identical to Article 18(3) of the US–Japan Digital Trade Agreement, once again mirrors the language of CDA 230(c)(2)(a).⁴³ Considering that neither Canada nor Mexico have a comparable statute in effect that covers the same grounds or provides the same protections as those afforded under the CDA 203, USMCA Article 19.17 can be seen as a tool to broaden the landscape of the CDA – “exporting” the US policy to the North and South.⁴⁴ Questions have therefore been raised regarding whether the expansion of the CDA 203 can be seen as a form of “American imperialism.”⁴⁵

Nonetheless, it is too soon to say that Canada and Mexico are now obligated to enact a CDA 230-equivalent statute at home, and that the US has successfully transplanted its controversial social media immunity abroad.⁴⁶ As a matter of fact, the real ramifications of USMCA Article 19.17(3) might be less apparent when reading other provisions together. Footnote 7 of the Digital Trade Chapter of the USMCA clarifies that a party may comply with Article 19.17 “through its laws, regulations, or application of existing legal doctrines as applied through judicial decisions.” This raises the question of whether Canadian case law is sufficient for compliance with USMCA Article 19.17.⁴⁷ More importantly, it is not immediately apparent if the rulings of the Canadian courts will consistently result in broad CDA-style immunity for social media, as was the case among US judges.⁴⁸ After all, as discussed above, today’s broad immunity stemming from CDA 230 is actually the product of expansive judicial interpretations of US courts. The mere fact that the US has transplanted the legislative text of the CDA 230 to the USMCA does not necessarily mean that the judicial interpretations of the US courts will guide the rulings of the Canadian courts.⁴⁹

In conclusion, with respect to the question of whether Article 19.17 of the USMCA is a big win for big tech, the answer is: perhaps, if the seeds planted continue to increase in size and gain strength.⁵⁰ In view of this, the ongoing WTO e-commerce trade negotiations represent a key battle that could set the tone for the future of platform immunities. Currently, CDA 230-like immunity can be found in the negotiating text of the WTO JSI on E-Commerce. The US-proposed language for “interactive computer services” clearly tracks Article 18 of the US–Japan Digital Trade Agreement and Article 19.17 of the USMCA.⁵¹ Paradoxically, the USTR continues to embrace and propagate CDA 230-like platform immunities in international trade agreements, no matter how much political controversy the Act stirs up at home.

4.2.4 Platform Content Moderation: Regulatory Landscape and Trade Governance

4.2.4.1 UGC Moderation under the DSA: The Gold Standard?

The potential for the CDA 230 to morph into a global standard for platform governance is further diminished under the EU’s accelerating effort to regulate digital platforms’ content moderation.⁵² Aiming to establish a benchmark for platform regulation at the global level, the “rules-based” platform governance model led by the EU is now a strong power in balancing CDA-based social media self-regulation. In a nutshell, transatlantic digital fragmentation has been bolstered by the DSA. On the other side of the Atlantic, the EU’s DSA has introduced a new set of obligations for online platforms. By setting high standards for effective intervention, the DSA increases the obligations of platforms and the powers of regulators, and it also empowers users to report illegal content in an easier way and challenge platforms’ content moderation decisions.⁵³

In particular, depending on the function and size of a digital services supplier, the DSA provides “cumulative obligations” – the larger a supplier is, and the more “critical” the services it supplies, the more obligations apply. Specifically, the DSA distinguishes between four tiers – intermediary services, hosting services, online platforms, and VLOPs – and applies asymmetric obligations to each tier.⁵⁴ Such a classification features four risk categories, and the DSA imposes increasingly severe obligations pursuant to the principle of proportionality. Accordingly, VLOPs and very large online search engines (VLOSEs), defined by the DSA as online platforms and online search engines providing more than 45 million average monthly active users in the EU (which represents 10 percent of the 450 million users in the EU market), are subject to the most broad and stringent requirements.⁵⁵ The EU legislative documents stress that very large players are emerging as “quasi-public spaces” for the exchange of information in this digital age, which may “pose particular risks for users’ rights, information flows and public participation.”⁵⁶ As such, they should “take mitigating measures at the level of the overall organisation of their service” to facilitate public debate, and to influence how people obtain information online.⁵⁷

Regarding content moderation of UGC, the DSA obliges online platforms to remove illegal content upon the receipt of an order issued by relevant authorities.⁵⁸ Online platforms are also required to put user-friendly mechanisms into place to allow any individual to notify them of any illegal content on their sites.⁵⁹ They are required to include information on “any policies, procedures, measures and tools used for the purpose of content moderation, including algorithmic decision-making and human review” in their terms and conditions.⁶⁰ Moreover, the DSA imposes transparency reporting obligations on platforms, including the publication of “easily comprehensible and detailed reports on any content moderation they engaged in during the relevant period,”⁶¹ which includes “any use made of automatic means for the purpose of content moderation.”⁶² Furthermore, the DSA imposes additional obligations on VLOPs to manage risks.⁶³ They are required to carry out regular assessments of the systemic risks stemming from their services. When conducting risk assessments, they should take into account how their content moderation systems influence the dissemination of illegal content.⁶⁴ Further, they have an obligation to take measures to mitigate any systemic risks.⁶⁵

As a whole, the DSA represents a big move toward a rules-based cyberspace order. Contrary to the laissez-faire approach taken in the CDA 203, the DSA aims to turn the unfettered market for UGC into a more heavy-handed, regulated public square.⁶⁶ By requiring social media to carefully curb hate speech, disinformation, and other extreme and unsafe content on their sites or otherwise risk billions of dollars in fines, the DSA, to a large extent, ends the era of content moderation’s self-regulation, in which speech platforms decide what content will be retained or removed based on their own unilaterally imposed policies.

4.2.4.2 Moving toward a Rules-Based Digital Order? The Elephant(s) in the Room

It should be noted that digital platforms established outside of the EU that offer their services in the EU are also subject to the obligations under the DSA. In April 2023, the EC officially designated the first batch of VLOPs and VLOSEs.⁶⁷ These big tech and other powerful firms will have to meet new EU requirements, including submitting yearly audits of systemic risks linked to their services. Needless to say, big tech firms boosted their EU lobbying in 2021–2022, during which the DSA was under discussion by the European Parliament and Council. With combined lobbyist spending of over 27 million euros in one year,⁶⁸ big tech certainly tried to influence EU policymaking on platform governance. At the end of the day, however, the EU’s “unilateral global governance,” as framed by Krisch,⁶⁹ “targets U.S. companies.”⁷⁰ Brussels eventually put the brakes on big tech’s uncontrolled power over content moderation.

What is most concerning to international economic legal order is the question surrounding to what extent speech platform governance formulated by the DSA will serve as a landmark that drives regulatory efforts beyond Europe. Under the trends of datafication, there is worldwide concern that states are no longer capable of regulating data.⁷¹ The DSA is now serving to spur a rules-based Internet. Countries such as Singapore and Taiwan have attempted to follow the EU’s example in regulating platforms with respect to UGC moderation.⁷² The years to come will be critical to the global governance of digital platforms. Key indicators include whether more and more countries will adopt DSA-like regulations along the EU regulatory path, whether the implementation of the DSA will have the same global influence as the General Data Protection Regulation (GDPR) does, and whether the EU’s accelerating “strategic autonomy” may effectively suppress the spread of the CDA 230 and eventually change the distribution of power in the social media ecosystem.

Of course, platform governance is not only about the US’ CDA model and the EU’s DSA model. The dispute between Twitter and the Indian government serves as a typical case pointing toward the bumpy road ahead for the global governance of speech platforms. India’s Section 69 (A) of the Information Technology Act allows authorities to issue blocking orders to social media intermediaries.⁷³ Blocking orders can be issued for, among other reasons, the “public order” of India’s society. Such discretionary power has been a major tool for the government of India to place pressure on big tech, as a social media services supplier can face criminal action and risk losing access to the Indian market for not complying with an order to take down certain content.⁷⁴ In July 2022, Twitter initiated legal action in the Indian administrative court for a judicial review of such orders,⁷⁵ claiming that blocking orders from the Indian government are procedurally and substantively deficient under Section 69 (A) of the Information Technology Act. Twitter alleged that Indian officials have abused their power, failed to meet the procedural requirements, and fallen short in demonstrating how the content at issue is under the purview of Section 69 (A). Additionally, Twitter argued that in many cases, the orders were “disproportionate, overbroad and arbitrary.”⁷⁶

There are other, even larger elephants in the room. Russia recently fined Google millions of USD for not deleting banned content on YouTube,⁷⁷ not to mention China’s data governance regime, which has long been the key cause of the fragmentation of global data policies.⁷⁸ There are also many smaller elephants in the room, such as Vietnam and Myanmar, which have mirrored the Chinese authoritarian model of data governance.⁷⁹ Ultimately, these competing models of platform governance will use international trade instruments, most likely through e-commerce/digital trade provisions, to influence other countries’ digital policies.

4.3 Regulating VSS: Trade Aspects of Cultural Policy

4.3.1 Global Dominance of the Streaming Platforms: Cut the Cord!

Now turning to the other angle of digital media content regulation – cultural expression on VSS. There are different terms used to describe broad and unique types of streaming platforms, which are often based on their business models. For the purpose of this chapter, “video streaming services,” as defined by the Body of European Regulators for Electronic Communications (BEREC)⁸⁰ and the Australian Communications and Media Authority (ACMA), refers to the “streaming of video content over the public Internet,” which provides “professionally produced content – either by the in-house film studios or through licensing deals, including on-demand services and/or linear content.”⁸¹ This content is subscription-based⁸² and ad-supported.⁸³ Unlike traditional audiovisual services such as broadcasting or cable television stations, for which video transmission requires a television aerial, a satellite dish, a set-top box, or a coaxial cable to connect to the services, VSS, such as Netflix, Disney+, Amazon Prime, or iQIYI, are delivered over the public Internet. Simply put, VSS allow audiences to bypass traditional distribution and thus “cut the cord.”⁸⁴

Digital media content has changed the landscape of the television industry. The increasing availability of public Wi-Fi, faster broadband Internet speeds, and unlimited mobile data plans offered by telecommunications operators have bolstered the growth of VSS. Dominant VSS platforms such as Netflix are transforming the prospects of the audiovisual industry and are now commonly considered a ready substitute for traditional television.⁸⁵ By way of illustration, in the US, VSS have significantly cut into cable’s market share, reducing the number of cable subscriptions from 85 percent of households in 2015 to 71 percent in 2021.⁸⁶ More and more people are turning to VSS for their television consumption, which had reached a US household penetration rate of 80 percent by March 2022.⁸⁷ The television landscape continues to transform, and this transformation has gone international.⁸⁸ Cord-cutting is becoming the norm in the Asian-Pacific region. In South Korea, Netflix alone had more than 5 million subscribers by June 2022.⁸⁹ In Japan, the VSS market, primarily offered by Amazon Prime Video, Netflix, and Disney+, had a total of 44 million subscribers as of the end of August 2021.⁹⁰ In Australia, approximately 71 percent of Australian adults had at least one subscription streaming service in their household in 2020.⁹¹

The trend of cord-cutting brought about new regulatory challenges for national media regulators. Television has traditionally played a major role in forming public opinion, shaping cultural identities, and promoting linguistic diversity, thus contributing significantly to social inclusion. Within the domestic legal framework, the sector has been heavily regulated.⁹² The justification for such heavy regulation of television content, however, has been weakened, because viewers of “traditional” television services also watch streaming content, which generally is not subject to the same regulation.⁹³ To illustrate, in most jurisdictions, there is a major dichotomy between (traditional) regulated television and (modern) unregulated video streaming services. In other words, although facing a looming regulatory storm, to a large extent, VSS remain totally unregulated, existing in a “regulation-free zone.”⁹⁴ Because VSS are steadily becoming a substitute for traditional television, policymakers are struggling over whether and how to eliminate regulatory distinctions that have left traditional television services suppliers handicapped in the face of streaming services. Should existing media regulations be extended to manage streaming services? Here, an important consideration that further complicates this regulatory issue is the reality that for most countries – probably only except the US and China, where the dominant streaming services are run by domestic rather than foreign companies – such a regulatory dichotomy is literally about whether “foreign-owned” VSS should be subject to the same regulations as local broadcasters and cable channels. Admittedly, there has been unequal competition between “foreign” streaming platforms and “domestic” television legacies. To translate this phenomenon into WTO language, most WTO members are the “importing countries” of VSS. However, VSS fall outside the scope of these countries’ domestic media law. As a result, local broadcasters and cable channels are at a significant business disadvantage compared to foreign VSS suppliers in terms of local bureaucracy, legal compliance costs, and other content regulations. The tension is sharp.⁹⁵

4.3.2 Leveling the Playing Field: Regulating Up or Down?

Traditional media services suppliers have been calling for a “level playing field,” or fair conditions between competitors in the media industry. On the one hand, it is a commonly shared regulatory principle to “treat like services alike” and refrain from using regulations as a means of choosing technological winners. Media regulators should not discriminate against different business models or technological features. Unequal and differential regulatory treatment in favor of the streaming platforms may mean that the regulators inevitably choose winners in the media industry. The business environment is that VSS gain “unfair” advantages by bypassing existing media content regulations.⁹⁶ It can be said that the asymmetric regulation has impeded the ability of broadcasters and cable channels to compete with their online competitors. This raises the question of whether the same requirements should be imposed on all services suppliers competing in the audiovisual sector, no matter whether the services are supplied via a traditional or a platform-based method.⁹⁷

Trade associations such as the Cable and Satellite Broadcasting Association of Asia (CASBAA) advocate that VSS should be categorized as television services and thus be subject to traditional broadcasting and/or cable rules. The association stresses that national regulators should “reduce regulatory asymmetries and foster competition” by “extending traditional regulation to online video services.”⁹⁸ To promote the concept of a “level playing field,”⁹⁹ the British Broadcasting Corporation’s (BBC) director-general has publicly stated that “TV-like” streaming services should be regulated to the same extent as the UK’s traditional TV so as to promote British content and preserve cultural heritage.¹⁰⁰ Conversely, the VSS groups emphasize that traditional television and video streaming has “inherent technical and functional differences.”¹⁰¹ They argue that different viewer experiences are a key consideration in distinguishing streaming platforms from their offline analogues. Some believe that the even-handed approach is not only unnecessary, but also infeasible. Additionally, extending old regulations to new media platforms could potentially harm business innovation, which in the long run may cause the decline of “media pluralism.”¹⁰² The primary position is that regulators should not interfere with new services in order to protect “old” services.¹⁰³

The question of whether the two types of services at issue should be required to meet the same standards involves the assessments of sector-specific policy objectives and market conditions under which legacy suppliers and digital platforms compete. In this regard, the Canadian Radio-television and Telecommunications Commission, while recognizing the need for a level playing field, ultimately decided that the two types of services at issue are “fundamentally different and warrant different regulatory treatment.”¹⁰⁴ Similarly, the ACMA also decided not to expand legacy media regulations to online players. Alternatively, however, the government reduced several general regulatory obligations for the broadcasters so as to level the playing field.¹⁰⁵ In other words, instead of “regulating up,” the Australian government reflected on the traditional media regulations and pursued regulatory parity by “deregulating down” the requirements imposed on incumbents.

4.3.3 Local Content Requirements on Netflix: Cultural Expression

Nevertheless, there is a legislative trend across various jurisdictions to issue new regulations to impose local content and local investment requirements on video streaming platforms. In other words, although many countries have decided that existing media law should not be applied across the board to all, they have opted to impose local content requirements on digital platforms. In Europe, the amended Audiovisual Media Services Directive (AVMSD) requires member states to ensure that VSS maintain at least a 30 percent quota of “European works” in their catalogues, with an emphasis on “the prominence of those works.”¹⁰⁶ In addition, the amended AVMSD permits member states to adopt measures to financially support European works, including direct investment in local content and contributions to national funds. Such financial contributions can only be based on the revenues earned in that member state and must consider other financial contributions levied by other member states.¹⁰⁷ Put simply, VSS, such as Netflix, Disney+, and Amazon Prime Video, must ensure that at least 30 percent of their libraries are dedicated to local content in the EU, either through production or licensing.¹⁰⁸ Moreover, they may be required to invest, directly or indirectly, in local content. As stressed in the background papers of the Directive amendment,¹⁰⁹ video streaming platforms directly compete with broadcasters but have circumvented the local content requirement. By imposing a minimum 30 percent quota on VSS, “European expression” and cultural diversity will be promoted. At the same time, by increasing the investment in original European content, local employment in the media industry will likely surge.¹¹⁰

Under the mandate, some EU countries have introduced tailored legislation to implement AVMSD, while others are in the midst of transposing the AVMSD onto the domestic legal framework. For example, Spain requires VSS to reserve 30 percent of their catalogues for European works, and half of them must be spoken in the language of Spain.¹¹¹ The government of Portugal imposes an annual fee that levies 1 percent of VSS income to support Portuguese language film productions.¹¹² The French decree implementing the AVMSD obliges services suppliers of video streaming platforms to invest 20–25 percent of their revenue earned in France in local content. Similar requirements have been adopted in other EU member states, such as Italy and Switzerland.¹¹³

Outside of Europe, local content requirements for VSS have been carried out or are being considered by many countries, including Australia, Argentina, Canada, Mexico, and Taiwan, to name just a few.¹¹⁴ In its policy consultation paper, the Australian media regulator pointed out that the ecosystem which supports domestic content is fading. Australian programs represent only 1.7 percent of titles for Netflix’s entire catalogue. In spite of this, Australian viewers are increasingly ditching local broadcasters that rely on advertising revenue to survive. Without regulatory intervention, local broadcasters will soon become unable to produce Australian content to meet quota requirements.¹¹⁵ The government is therefore aware of the urgent need to require VSS to make or source “quality Australian stories” that help Australians to “understand each other.”¹¹⁶

In brief, more and more national media laws are deciding to impose local content requirements on streaming platforms, ranging from including a certain amount of local content in their catalogues, displaying a minimum percentage of content in the local language, producing a certain amount of media content locally, and making financial contributions to local content production.¹¹⁷ Such a legislative trend, however, has been labeled a potential “trade barrier” by the US – the dominant “exporting country” of VSS – as it may impose an unnecessary compliance burden on digital platforms and limit global market access of VSS. This raises the question of whether the local content requirements placed on digital platforms violate obligations under international trade agreements, especially in the FTA context.¹¹⁸

4.3.4 “Foreign” and “Domestic” Streaming Content: Definitions, Likeness, and Exceptions

4.3.4.1 Performance Requirement: “Domestic” Streaming Content

First, a preliminary issue is how to define “domestic content” on video streaming platforms. How can we determine the “origin” of a streaming TV show or film? Indeed, the definition of “domestic content” becomes even more complicated when considering the winds of coproduction and outsourcing. Netflix, among other streamers, has extensively grown its coproduction partnerships in recent years.¹¹⁹ The industry reality is that digital media content is comprised of sophisticated inputs from multiple sources. Should the Netflix TV show series “Emily in Paris” be considered European content due to that fact that it was filmed in Europe (i.e., Paris)? Or should it be deemed non-European content because it was produced by an American company (i.e., Netflix) and it starred a British and American actress (i.e., Lily Collins) as the eponymous Emily? Will “cultural identity” be a factor in deciding the “nationality” of a TV show? Will it be a consideration if many French critics condemned the show for stereotyping Parisians and misrepresenting French culture? All of these questions lead us to rethink issues of cultural diversity in the digital context.¹²⁰

In this regard, the nationality assessment under the AVMSD serves as a proper illustration. The question of whether a TV show is a “European work” under the meaning of Article 13 of the AVMSD is determined according to the criteria provided in Article 1. To be qualified as “European works,” the works must originate in EU member states or from European third-country states that are parties to the European Convention on Transfrontier Television (ECTT)¹²¹ and must also fulfill certain conditions set out in the AVMSD.¹²² Alternatively, the works must be coproduced within the framework of agreements related to the audiovisual sector concluded between the EU and third countries and must fulfill the conditions set out in those agreements.¹²³ Overall,¹²⁴ key considerations provided by the AVMSD include the place of residence of authors and employees of a work, the place of establishment of the producers, and the financial contributions of coproduction costs.¹²⁵ In other words, whether streaming content can be qualified as a European work must be assessed based on relevant information and evidence proving the percentage of European authors, workers, producers, and fiscal sponsors in a specific TV show.

Such an exercise, however, may constitute performance requirements for investment in services. Most FTAs lay out general rules for the treatment of investors and investments. Local content requirements – whether they involve hiring a certain proportion of local “TV people” or increasing investment in local production facilities – must be considered with respect to their potential violation of international economic law. For example, Article 14.10 of the USMCA prohibits the parties from imposing any requirement to “purchase a service from a person in its territory,”¹²⁶ or to “achieve a given level or percentage of domestic content.” ¹²⁷ Such a provision, on its face, serves as a protection against local content requirements on digital platforms.¹²⁸ Based on that, the US has been “actively monitoring” whether Mexico’s audiovisual reform bill, which calls for local content requirements on VSS, is consistent with the USMCA obligations.¹²⁹

4.3.4.2 Nondiscrimination: “Like Digital Products”

Another important issue is discrimination. Let us continue the analysis in the FTA context using the USMCA as an example. Article 19.4 of the USMCA requires parties to ensure “non-discriminatory treatment of digital products” by according no “less favorable treatment” to “like digital products.”¹³⁰ A USMCA party may not discriminate against digital products originating in another party, either because they were “produced in another party” or because they were “produced by a person of another party.”¹³¹ According to Article 19.1, “digital products” include a video that is “digitally encoded, produced for commercial sale or distribution, and that can be transmitted electronically.”¹³² It is worth mentioning that a similar provision can be found in Article 14.4 of the CPTPP, with an additional subparagraph clarifying that the provision shall not apply to broadcasting.¹³³ In any event, taking the example of Mexico’ s audiovisual reform bill from above, a Mexican law imposing local content quotas on VSS may be considered as treating US TV shows less favorably than Mexican TV shows on digital platforms. Here, a violation could be found by comparing US and Mexican TV shows. If they are “like TV shows,” the adverse treatment of the US TV shows under the Mexican measure may be considered discrimination.

At the core of the question is how to determine “likeness” when the products at issue are those directly associated with cultural expression. The tension between the liberalization of the audiovisual trade and the preservation of cultural identity is an old “trade v. culture” debate. The production and distribution of television programs typically fall under the definition of “core cultural products” within the meaning of the United Nations Educational, Scientific and Cultural Organization’s (UNESCO) Cultural Diversity Convention.¹³⁴ Now, the streaming technologies put the old wine into new bottles: Can the difference in cultural perspectives be taken into account when assessing the likeness of digital media content?

The Canada – Periodicals dispute, in which the confrontation between trade and culture is best demonstrated, can be used as an apt source in answering this question.¹³⁵ Aiming to protect the Canadian publications industry, the Canadian Parliament amended the Excise Tax Act and imposed an advertising tax on split-run editions of foreign periodicals, targeting popular US magazines such as Time.¹³⁶ According to the official statement from Canada, the measures were intended to “foster conditions in which indigenous magazines can be published, distributed and sold in Canada on a commercial basis,”¹³⁷ thus, to “maintain an environment in which Canadian magazines can exist in Canada alongside imported magazines.”¹³⁸ The US brought the case to the WTO, claiming that Canada’s discriminatory measures violated GATT Article III on national treatment. The panel basically agreed with the US position that the “cultural products” at issue, namely the US split-run magazine and the Canadian magazine, were “like products” in terms of relevant factors, including the magazines’ end uses in the Canadian market, consumers’ tastes and habits, and the magazines’ properties, natures, and qualities.¹³⁹ The Appellate Body took a different approach and found that the imported US split-run periodicals were “directly competitive or substitutable” with domestic Canadian non-split-run periodicals in the Canadian market.¹⁴⁰

The arguments put forth in Canada – Periodicals, now more than two decades old, still echo today’s cultural concerns. Throughout the litigation, Canada argued that the two products in dispute were not “like products” within the meaning of Article III:2 of the GATT. Canada stressed that the split-runs substantially reproduced foreign editorial material, whereas the local magazines were developed for the Canadian market.¹⁴¹ Canada further pointed out that the nature of the magazines was for intellectual consumption, which differed from normal goods in terms of physical use and physical consumption.¹⁴² It followed that the “prime characteristic” of cultural goods should be their “intellectual content” rather than the criteria for noncultural goods, such as a bicycle or canned tuna fish.¹⁴³ Nevertheless, both the panel and the Appellate Body rejected Canada’s “cultural claims” because “cultural identity was not at issue in the present case.”¹⁴⁴ The former USTR Mickey Kantor even called the Canadian “cultural identity claim” an “excuse to protect the economic viability of the Canadian (cultural) industry.”¹⁴⁵

The Canada – Periodicals dispute demonstrates the status of “culture” in the determination of “likeness.” Turning back to the scenario above, if we follow WTO jurisprudence, there is little room for a trade tribunal to rule that US TV shows and Mexican TV shows on streaming platforms contain different cultural elements and are therefore “unlike digital products.” Local content requirements for VSS may therefore violate the obligations of nondiscriminatory treatment of digital products under the Digital Trade/E-Commerce Chapter of the FTAs, simply because the domestic regulation treats “like TV shows” unlike.

4.3.4.3 “Digital Cultural Exemption” and Side Letter

The ultimate question is therefore whether the local content requirements, if found to be inconsistent with international trade or investment rules, can be justified under the exceptions. Looking back to the WTO negotiating history, “cultural exceptions” had been proposed during the Uruguay Round Negotiations but were eventually dropped from the negotiating texts.¹⁴⁶ As a result, neither GATT nor GATS explicitly provides for any “cultural exception” to WTO law.¹⁴⁷ Although the media landscape has gone through a digital transformation since then, little progress has been made within the WTO in this regard.¹⁴⁸ The question of how to adequately preserve cultural policy space for members has not been properly addressed in the WTO since the conclusion of the Uruguay Round. The debate, to a large extent, has been shifted to the arena of the FTAs. For example, the screen quota issue was at the heart of the controversy of US–South Korea FTA (KOURS) negotiations in the late 2000s, but South Korea decided not to seek a cultural exception under the KOURS.¹⁴⁹

The concept of “cultural exceptions” has been at least partially realized in recent FTAs through mechanisms such as exemptions or side letters. In particular, Chapter 32 of the USMCA¹⁵⁰ – Exceptions and General Provisions – contains a far-reaching exemption to protect Canada’s cultural industry, including the cultural content of digital platforms. To further clarify, Canada’s cultural industry exemption from NAFTA has been continued in Article 32.6 of the USMCA to allow Canada to protect its “cultural industries,” in particular, the production and distribution of television, video recordings, and film.¹⁵¹ It should be noted, however, that additional text on retaliation in Article 32.6 has been added to allow the US and Mexico to take measures toward “equivalent commercial effect” in response to any action taken by Canada to protect its cultural industry.¹⁵² Nevertheless, Article 32.6 of the USMCA effectively safeguards Canada’s regulatory autonomy over cultural industries. Accordingly, cultural measures on VSS, including local content requirements, are subject to the “cultural carve-out.”

A less comprehensive and somehow more straightforward approach has been taken in the CPTPP. Canada has concluded bilateral agreements with all other CPTPP parties via the exchange of side letters to “ensure Canada’s ability to adopt programs and policies that support its cultural sector, including in the digital environment.”¹⁵³ In the side letters, CPTPP parties agree that Canada has the discretion to define “Canadian content,” to protect its online content, and to maintain full policy flexibility to protect its cultural sector.¹⁵⁴ Although its civil society criticized the side letters, contending they do not constitute a broad cultural exemption to ensure the production of high-quality Canadian content in the digital environment, the CPTPP side instruments leave the door open for the Canadian government to regulate VSS in line with its cultural policy. In fact, at the time of this writing, Canada’s parliament had just passed the Online Streaming Act to impose local content requirements on streaming platforms.¹⁵⁵ By requiring foreign streaming giants like Netflix and Spotify to feature a certain amount of Canadian content, the Act aims to “ensure Canadian stories … are widely available on streaming platforms,” and to “reinvest in future generations of artists and creators in Canada.”¹⁵⁶

To conclude, the “trade v. culture” history has a new chapter: the protection of local culture in global digital platforms. For decades, trade policy debates surrounding the audiovisual sector have been prominent, given that the sector is closely concerned with cultural activities. The same holds true in the twenty-first century, where certain digital media content can be seen as the manifestation of a culture. Some TV shows on streaming platforms can be described as a way to reflect the social characteristics of a country and its people, as well as a venue by which to present a country’s cultural identity to the rest of the world.¹⁵⁷ When VSS stream such content, they convey and expand certain cultural expressions to their audiences. Notwithstanding, the audiovisual sector is also an economically significant sector that shares commercial characteristics common to other services sectors. The argument that a certain portion of digital media content is more entertaining than cultural characters and they are therefore in no way different from any other commercial product is convincing and hard to rebut.¹⁵⁸ From Hollywood to Silicon Valley, international trade agreements have been struggling to play a more prominent role in mitigating the collision between commercial and cultural interests. As Burri pointed out, there are many paths open to improve the approaches to cultural diversity in the trade context.¹⁵⁹ Amid the hype over the Metaverse, the media landscape is now developing into a collective space created by VR technologies that allow users to interact with one another through avatars.¹⁶⁰ Cultural diversity in the VR space will be even more challenging and multifaceted to policymakers when it comes to avatars. The “trade v. culture” clash in international economic law will continue, and it will also become much more complex in a datafied world.

4.4 Media Platformization and Local Presence

4.4.1 Jurisdiction and Law Enforcement Problems

In any event, media platformization brings about issues surrounding jurisdiction. No matter how carefully crafted, both angles of this chapter – speech regulation on UGC and cultural policy on VSS – must face enforceable reality. Given the global access of the digital platform services, the platforms regulations discussed above – whether content moderation obligations or local content requirements – should ideally be evenly enforced, both onshore and offshore. In this regard, a survey conducted by the EU when the DSA was being introduced revealed that 83 percent of the stakeholders believed that the territorial scope of the Act should be expanded to “digital services established outside the EU when they provide services to the EU users.”¹⁶¹ However, technically speaking, digital platforms that house their servers outside a country’s territory are largely out of regulatory reach in terms of law enforcement.¹⁶² How, then, can we govern services suppliers based in other jurisdictions?

In the age of the platform economy, if offshore services suppliers can easily escape the same level of regulation, such a law enforcement problem will create a misalignment of incentives for their media market to be served from overseas. For example, Singapore’s Content Code for Over-The-Top (OTT) Services (the Content Code) applies to both local and offshore OTT TV services suppliers.¹⁶³ Even so, although “the law in the books” makes no distinction between onshore and offshore services suppliers, “the law in action” might be quite different, simply because it is much more difficult for its media regulator – the Infocomm Media Development Authority (IMDA) – to compel offshore OTT TV services suppliers to comply with the Content Code.¹⁶⁴ Such an uneven enforcement of law can effectively force local suppliers to relocate and hollow out the local industry.

Placing this problem in WTO language, the legal question here is how to evenly enforce domestic regulation on services suppliers of cross-border services trade (Mode 1) and commercial presence (Mode 3). In this regard, a notable example is Turkey’s Regulation on the Transmission of Radio, Television, and On-Demand Services on the Internet (Turkey’s Internet Regulation), which requires streaming services suppliers to comply with a comprehensive licensing scheme and to establish a commercial presence in Turkey¹⁶⁵ – virtually restricting digital platforms from supplying services merely on a cross-border basis without establishing a physical presence in Turkey.

In a similar vein, a relatively soft requirement has been adopted in the DSA. Article 13 of the DSA requires digital platforms that offer services in any of the EU member states but do not have an establishment therein to designate “a legal or natural person as their legal representative” within the EU. Such a legal or natural person should be mandated by the platform companies to cooperate with the relevant authorities and comply with the DSA. The same Article further states that “the designated legal representative can be held liable for non-compliance with obligations” under the DSA, without prejudice to the liability and legal actions that could be initiated against the platform company.¹⁶⁶ It should be noted, however, that unlike the “commercial presence requirements” under Turkey’s Internet Regulation above, the DSA explicitly states that the designation of a legal representative within the meaning of Article 13 shall not amount to an establishment in the EU.¹⁶⁷ It may be presumed from the preparatory documents of the DSA that Article 13 is a dedicated drafted provision that is designed, on the one hand, to ensure similar supervision regardless of the place of establishment of the digital platforms, and on the other hand, to comply with the EU’s market access commitments undertaken in the GATS.¹⁶⁸

4.4.2 Local Presence as a Condition for Digital Platform Services

From the above, and in the light of media platformization, jurisdiction and law enforcement problems, among others, can be addressed through requirements for commercial presence (as stipulated in Turkey’s Internet Regulation) or legal representatives (as required by the DSA). At any rate, local presence requirements have long been identified by the US as a trade barrier for cross-border services trade.¹⁶⁹ The US efforts to prevent the spread of local presence requirements can be traced back to the US-led trade deals in the early 2010s.¹⁷⁰ A typical provision can also be found in the Annex on E-commerce of the Trade in Services Agreement (TiSA) negotiating text.¹⁷¹ Such a provision has become a “template” for the services chapter in several subsequent FTAs.¹⁷² For example, Article 10.6 of the CPTPP states the following:

Article 10.6: Local Presence

No Party shall require a service supplier of another Party to establish or maintain a representative office or any form of enterprise, or to be resident, in its territory as a condition for the cross-border supply of a service (emphasis added).¹⁷³

Nevertheless, this obligation may be subject to a reservation in a party’s schedule. For example, the nonconforming measures maintained by Australia, as set out in its Schedule of the CPTPP Annex II, reserve the country’s right to adopt or maintain any local presence measures with respect to audiovisual services transmitted electronically.¹⁷⁴ It is surprising to note, however, that the reservation of a local presence on digital media services is, to date, rarely used in international trade agreements.

All in all, international trade rules that ban local presence requirements enable platform companies to supply services without establishing a local presence, which may have significant implications for the ability of governments to regulate and enforce platform regulations. From the perspective of law enforcement, governments have many reasons to introduce local presence requirements as a condition for foreign digital platforms to supply services on a cross-border basis, as it is virtually a prerequisite for a government to ensure that domestic regulation can be effectively applied and thus enforced. In countries where there is no requirement to maintain a local presence, digital platform operations may circumvent regulatory obligations.¹⁷⁵ As advocated by civil society, “without a local presence of companies, there is no entity to sue and the ability of domestic courts to enforce [local] standards … is fundamentally challenged.”¹⁷⁶ Said another way entirely, a platform company with a locally registered entity would render the authority’s law enforcement much easier, in that the companies can be legally compelled by the local authority to engage with the administrative process, submit information, and thus comply with domestic regulation.¹⁷⁷ In the case of digital media content regulation, a local presence requirement helps to ensure that the regulators maintain their ability to enforce UGC content moderation as appropriate. Digital speech platforms with a local presence can be sued for breaches of domestic regulations, brought to the local court, and held accountable for a legal remedy to address illegal content. Similarly, in the case of a local content requirement, local presence also means that cultural policies and court judgments against VSS can be more effectively enforced.

The challenges of holding digital platform companies accountable without a local presence are evident. In view of this, international trade rules prohibiting local presence requirements constrain a state’s policy space to address free speech and cultural expression. To conclude, international trade agreements in the age of datafication now face this dilemma. At one end of the spectrum, a state’s sovereignty to enforce regulations against platforms might be compromised without their local presence. At the other end of the spectrum, cross-border digital trade without a local presence in other countries may be the most efficient business model for many platforms, especially SMEs. Requiring a digital platform supplying services from offshore to have a local presence in the country may inevitably add operational costs and thus constitute a market access barrier to cross-border services trade.

4.5 Conclusion

In this chapter, we have examined the interplay between digital media content regulation and international economic law. To better explain their interactions, this chapter spotlights two policy areas – UGC moderation and VSS screen quota – as windows for exploration. When data becomes speech and expression, data governance requires perspectives that extend well beyond technological and economic factors. Both the UGC and VSS regulations analyzed in this chapter are prime examples demonstrating the need to regulate content moderation, to alter the power distribution in the Internet ecosystem, and to protect noneconomic values such as cultural diversity in global digital platforms. The findings from this chapter extend beyond issue-specific contexts to identify the contradictions between, on the one hand, the free trade and global expansion path offered by the international trade regime and, on the other hand, the public policy objectives pursued by national media law and policy. It goes without saying that the excessive market concentration of social media platforms, acting as “gatekeepers” of the flow of digital media content, may systematically discriminate against certain “speech” in datafication-enabled content moderation or promote extremist speech through algorithmic amplification. It is equally obvious that the overwhelming global market share of the dominant streaming platforms may work against the goals of media pluralism and cultural diversity. This “competition” angle is our focus in Chapter 5.

5 Data as Capital and Algorithmic Input Competition, Transparency, and Trade Rules

5.1 Introduction

The most important feature of platformization is scale. A platform can only provide value to users if it grows to a significant size.¹ When a digital platform reaches a certain scale, it gains access to more and more of its users’ data. Of course, this feature is not even remotely novel, because larger factories have always been more efficient than smaller ones, even in the “old economy.” However, datafication forces this economic logic to the extreme.² In this regard, companies in smaller countries are often disadvantaged vis-à-vis companies in larger countries, simply because of the constraints of smaller markets in terms of efficiencies of scale and volume of data.³ The leading platforms – including Google, Facebook, Amazon, Baidu, and Alibaba – were launched in the US or China, where they could operate and reach the necessary scale in a large domestic market before they went global.⁴

More importantly, big tech companies have the ability to commoditize our data, which is the key ingredient of many digital services, including AI.⁵ Many commentators have rightly pointed out that data is the single-largest lasting asset of these globally dominant companies.⁶ Indeed, data is now becoming a form of capital.⁷ The ability to collect, use, and apply data is a competitive parameter whose relevance is quickly increasing.⁸ The data held by these leading platforms is particularly valuable due to the scale and scope of user data collected, which further provides these big players with strong competitive advantages, allowing them to dominate in the relevant market, create entrance barriers to potential competitors, attract more and more users, build richer and richer data sets, and reinforce their market power.⁹ In short, control over data may effectively act as a market entry barrier.¹⁰ Big platforms’ control over our data strengthens their market position and makes it easier for them to enter new markets. In the case of Meta, as an example, access to data not only enables Facebook to tailor services, but also to use data to benefit other business lines, such as Messenger, WhatsApp, and Instagram.¹¹

The reality of this battle is clear: Big platforms’ business practices interlock a combination of forces to dominate the data market – presenting a new form of modern monopoly.¹² To some extent, the emerging phenomenon of leading platforms that appropriate and extract data for profit can be conceptualized as “data colonialism.”¹³ Overwhelming “economies of scope” empower these large incumbent platforms, giving them a strong competitive advantage.¹⁴ Platformized transactions further enable the expansion of data capitalism, which works both domestically on a home country’s populations, and also on a global scale. In this twenty-first century version of colonialism, big tech companies benefit from colonization all over the world. From this aspect, the North–South divide does not seem to matter as much as it usually does.¹⁵ The US and East Asia account for 90 percent of the market for large-scale digital platforms, whereas Africa and Latin America’s combined share comprises only 1 percent of the market.¹⁶ These uneven, if not one-way, transnational data flows indicate that “data” – the input for AI and other technologies – has largely originated abroad for the benefit of big tech’s data analysis.¹⁷ Given the inordinate concentration of digital technologies in developed economies and a few Asian countries, most developing countries are becoming “net data exporters” that consistently supply valuable data without fairly benefiting from the digital economy.¹⁸

Against this backdrop,¹⁹ this chapter explores two notable features of the platform economy: First, data has become capital. At the crux of the matter are questions regarding how to “decolonize” data, how public policy should evolve to promote competition in the digital market, and how data capitalism as a whole should be confronted.²⁰ More importantly, what role can international law assume in promoting competition in the digital market? Second, data has become input or feeds for platforms’ algorithmic recommendations, which are at the heart of the datafied economy’s business models. Problems arising from this phenomenon are potential unfair competition and the lack of algorithmic transparency and accountability. Big tech’s datafication-enabled advertising, or, more generally, its overall business model, calls for deep thinking about how to appropriately regulate associated datafication practices. Recognizing that our behavioral data has increasingly become a commodity, the surveillance power of the digital platforms, left without proper constraints, will deepen the negative effect of datafication. What are the necessary steps toward holding digital platforms accountable due to their crucial roles in this datafied society? Are existing regulatory tools sufficient to provide domestic regulators with the information they need to ensure the adequate transparency of algorithmic systems in order to supervise how digital platforms moderate, rank, and recommend content to their users?²¹ In particular, what is the interplay between international trade agreements and national algorithmic transparency requirements? These are the key issues this chapter seeks to address.

5.2 Trade and Competition Policy for a Platform Economy

5.2.1 When Winners Act Globally and Take All

Much like the persistently unequal distributions in the broadband networks, the upper layer of the Internet architecture – the platforms – is now facing threats posed by data capitalism. The digital economy is gradually being shaped by increases in market concentration on a global scale, the proliferation of anti-competitive practices by digital platforms, and the abuse of dominant market positions by platform monopolies.²² Taken as a whole, “winner takes all” is a predictable phenomenon of the digital economy,²³ in which big tech companies do not “compete in the market,” but, rather, “compete for the market” to displace each other.²⁴ In this context, competition policy must be tailor-made in order to ensure its effectiveness vis-à-vis dominant digital players, thereby safeguarding competition in the markets.²⁵

In fact, big tech firms have increasingly been investigated for abuse of dominance and anti-competitive behavior all over the world, which offers plenty of food for thought about data capitalism. Many are in the midst of investigations by competition authorities, some are facing the scrutiny of the administrative courts, and several cases have been concluded.²⁶ For example, in its Apple Dating app case,²⁷ the Netherlands Authority for Consumers and Markets (ACM) found that Apple abused its dominant market position by imposing unreasonable terms and conditions on the services suppliers of the dating app in the App Store. In the view of the ACM, Apple enjoys a dominant position in the relevant market of the mobile operating system for dating app providers. “Having a dominant position is not illegal in and of itself. Abusing one, however, is,” explained the ACM.²⁸ Due to the fact that Apple restricts dating app providers’ freedom of choice, including banning them from referring within their own dating apps to alternative payment systems outside the app, the ACM came to the conclusion that Apple violated the Dutch Competition Act.²⁹ Similarly, in the Google Shopping case,³⁰ the European Commission concluded that Google, by favoring its own comparison shopping service, had abused its dominant position in the relevant market for online general search services. According to the commission’s decision,³¹ Google breached the EU competition rules because it had “systematically given prominent placement to its own comparison shopping service.”³² Google appealed this decision, and the European General Court upheld the Commission’s decision,³³ finding that Google’s self-preferential practices constitute an abuse of dominance and have potential anticompetitive effects.³⁴

Overall, these cases have led to emerging regulatory efforts with regard to digital platforms. In particular, these decisions underscore the difficulties in correcting ex-post negative effects on competition and the need to have ex-ante regulations in place to constrain big tech’s behaviors and require platform transparency. The trend of rising inequality between those who provide the data and those who control the use of such data challenges our existing legal approaches to the problem of anticompetition. There is an urgent need to revisit the fundamental goals of competition law in the light of digital trade.³⁵

5.2.2 Emerging Competition Law for Data Markets

5.2.2.1 Big Tech and Competition Policy

Today, competition authorities all over the world are considering the benefits associated with digital platform obligations.³⁶ One potential mechanism, among others, is to require leading digital platforms to share data with other services operated by their potential rivals, which may “enhance data access, resolve data bottlenecks, and contribute to a fuller realization of the innovative potential inherent in data.”³⁷ In any event, all of the approaches require greater cross-border collaboration. Big tech companies act globally, and dominant platforms are global in scope. The impact of their market power can be more meaningfully addressed through competition rules at the international level.³⁸ Compared with national regulations, competition disciplines at the international level would be more effective in defining the relevant (global) market, identifying the abusive market power (globally), addressing (cross-border) collusive practices and digital cartels, and reviewing mergers of (global) platforms. After all, the leading platforms operate on a global scale, and as such, efforts at the international level would be more commensurate with the scale of impact of digital platforms.³⁹

In practice, the competition assessment will necessarily depend on the extent and type of data to be shared, the precise form of the data-sharing arrangement, the degree of transparency requirements, and the definition of the relevant market.⁴⁰ However, the gap in competition policies and enforcement among jurisdictions will likely leave any competition authority ill-equipped to effectively address the anticompetitive practices of the big tech companies, simply because data flows do not stop at borders.⁴¹ In short, more and more countries are now focusing their regulatory attention on big tech. Recent regulatory progress toward defining “dominant platforms” and establishing ground rules to promote competition reflect a growing understanding that the platform economy requires tailored intervention. The dynamics of global data flows, however, make it legally challenging to enforce data competition policies without global regulatory harmonization. The lack of consistency among national competition laws demonstrates the need for a more consistent, streamlined system among competition regimes – either through greater international collaboration or the creation of additional cross-border disciplines for competition policy.

A number of new regulations and regulatory recommendations have been floated at both the national and the regional level. Among these, the Organization for Economic Co-operation and Development (OECD) policy papers point to the high concentration of data-driven markets, express caution regarding the absorption of new entrants through acquisitions by dominant incumbents, and call for competition rules that seek to promote the efficient use and exchange of data.⁴² The European Parliament, in addition to the adoption of the two landmark pieces of legislation on digital platforms – the DSA and the Digital Markets Act (DMA) – has also passed or proposed regulations including the Data Act, the Data Governance Act, and the Artificial Intelligence Act, which seek to provide an overall framework for data governance.⁴³ At the same time, examples in the Asia Pacific region include Australia’s News Media and Digital Platforms Mandatory Bargaining Code, which was developed by the Australian Competition and Consumer Commission (ACCC) to address “bargaining power imbalances between the digital platforms and Australian news media.”⁴⁴ Moreover, the ACCC has already conducted public consultations on policies that would provide for greater regulatory oversight of digital platforms with strong market positions, such as Google and Facebook.⁴⁵

5.2.2.2 EU as Global Norm-Setter?

While various regulatory initiatives are still subject to policy debates, and certainly there are divergent views on how competition law should be restored to account for specific concerns brought about by datafication and data capitalism, the legal approaches share common elements:

• Relevant Markets and Market Power: All of the proposals address the need to clarify what constitutes the “relevant market” of a digital platform. To summarize, identifying relevant markets inside the ecosystem of “data” can prove particularly challenging, because big tech companies always assume multiple roles.⁴⁶ Competition authorities must identify a multi-side market and consider relevant data flows in the market.⁴⁷

• Dominant Position and Anti-Competitive Practices: A closely related issue is market power assessment in the context of data access and data control, which requires, among other things, specific criteria to assess the impact of a dominant market position. It is a generally shared view among competition authorities that when data is overly concentrated in the hands of big tech companies, it may provide these firms with a substantial competitive advantage against new entrants.⁴⁸ The misuse of data to maintain market power should be considered an anticompetitive practice that requires governmental intervention.⁴⁹ Most policy papers attempt to identify the types of anticompetitive conduct that are enabled through the control of data, including collusive practices and digital cartels.⁵⁰ As emphasized in several policy papers, the incentive for digital platforms to use data to collude with each other is enormous.⁵¹ As a result, the need for competition authorities to adapt tools to address digital cartels is overwhelmingly strong.

• Mergers and Acquisitions: Another closely linked dimension is “data-driven mergers and acquisitions.”⁵² As evidenced by the Facebook/WhatsApp merger,⁵³ it is not uncommon for digital platforms to acquire other digital companies and start-ups, which increases the risk of monopolization of data.⁵⁴ Policymakers increasingly understand the need to examine the impact of mergers on data, the overall competitive implications of mergers and acquisitions involving digital platforms, and the new threshold for merger control in competition law.⁵⁵

Against this backdrop, the EU’s DMA sets a high global benchmark for regulating digital platforms.⁵⁶ Overall, the DMA addresses digital market “imbalances” in the EU, imposes tailored asymmetric ex-ante rules on “gatekeeper platforms,”⁵⁷ provides a legal mechanism based on market investigations, and establishes harmonized rules prohibiting certain unfair practices among gatekeeper platforms.⁵⁸ In particular, the DMA establishes a set of defined criteria for qualifying a large digital platform as a “gatekeeper,”⁵⁹ with the underlying principle that digital platforms with a certain degree of market power should be subject to more stringent obligations than smaller players.⁶⁰ In September 2023, the EC has designated six gatekeepers: Alphabet, Amazon, Apple, ByteDance, Meta and Microsoft.⁶¹ Within six months of being designated gatekeepers, digital platforms are obliged to comply with a number of “special obligations.” More specifically, the DMA imposes obligations and prohibitions to limit gatekeepers’ abilities to process and use personal data,⁶² to negotiate certain conditions with business users,⁶³ and to restrict end users from switching between digital applications.⁶⁴ Note that the European Commission, based on market investigations, has the discretion to “update” obligations for gatekeepers to ensure the obligations are “up to date.”⁶⁵

Such an ambitious agenda reveals the EU’s aim to be a global norm-setter in digital markets.⁶⁶ Expectedly, the DMA will assert significant regulatory control over digital platforms, both within Europe and beyond. The rules will bind global platforms, rendering them de facto global standards – more commonly known as the “Brussels Effect.” As for big tech, the stakes are particularly high, because the EU is one of the world’s largest consumer markets. They must accept the EU’s “terms of business” as the price of admission. To conclude, driven by economic and strategic rationales, the EU has been leveraging its economic muscle and vying for a leadership role in shaping the global rulebook governing digital platforms. The EU’s intensified efforts to set international standards for the digital economy could be part of the solution tool kit to curb data capitalism.

5.2.3 Calling for a WTO “Data Reference Paper?”

5.2.3.1 Trade and Competition

At the multilateral level, competition was one of the so-called new issues under the WTO framework two decades ago, at which time members attempted to address how domestic and international competition policies interact with international trade.⁶⁷ To a certain extent, international economic law and competition policy are grounded in complementary principles.⁶⁸ While the former emphasizes free trade and prohibits discrimination between trading partners, the latter aims to preserve the freedom of business activities and control the anticompetitive conduct of private enterprises.⁶⁹ Much discussion has been carried out in relevant legal literature regarding the importance of competition policy to trade liberalization, which generally describes how international cartels affect international trade, how transnational abuses of a dominant position constitute trade barriers to goods or services,⁷⁰ and how anticompetitive vertical market concentrations exclude foreign suppliers from a market.⁷¹ Indeed, the rationales underlying both trade and competition are arguably closely related. Nonetheless, to date no significant consensus on the convergence of the two areas has emerged.⁷²

To illustrate, historically, the interaction between trade and competition policy has been an important element of both multilateral and regional trade negotiations.⁷³ The issue of competition policy, however, was dropped from the Doha Round of the WTO trade negotiations.⁷⁴ In brief, negotiating efforts to create a general agreement on competition policy under the WTO have been unsuccessful, and the Working Group on the Interaction between Trade and Competition Policy has been inactive since 2004.⁷⁵ At the same time, competition policy has been addressed in FTAs,⁷⁶ with an evident trend toward a dedicated chapter in recent years,⁷⁷ which to a certain degree implies a growing perceived need for common competition disciplines among countries.⁷⁸ Given such a trend,⁷⁹ the interface between international trade and competition policy is now primarily manifested by the incorporation of “basic competition principles” in the FTAs.⁸⁰ Most specifically use the terms “anti-competitive agreements” and “abuses of market power,”⁸¹ while few mention “anti-competitive mergers,” “merger control,” and “merger review.”⁸² It should also be noted that approximately half of the FTAs contain competition provisions pertaining to cooperation and technical assistance.⁸³

Notwithstanding those “WTO-plus” obligations at the regional level, key problems remain. “General competition policies,” which prohibit or require broad categories of business behaviors defined in rather general terms, are not sufficient to address digital cartels and data monopolization. The digital sector needs specifically tailored regulatory disciplines. How can international economic law help to ensure that additional pro-competitive regulations are put into place? The real question is this: How can we restore the relevance of international economic law in the digital economy? Could such a restoration be launched with the modernization of the WTO Telecommunications Reference Paper for the data-driven economy?

5.2.3.2 Telecom Reference Paper as a Model

While WTO members have to date failed to agree on competition rules, in the context of the post-Uruguay Round WTO negotiations on Basic Telecommunications Services, most WTO members have committed to the regulatory principles spelled out in the Telecom Reference Paper under the GATS,⁸⁴ which sets out specific obligations for competition.⁸⁵ In the absence of general competition rules under the WTO regime, the Telecom Reference Paper serves as a sector-specific competition agreement, through which anticompetitive practices can be challenged using the WTO dispute settlement system.⁸⁶

The Telecom Reference Paper requires members to adopt and maintain competitive safeguarding rules to prevent abusive restrictions on bottleneck facilities, which may result in a de facto limitation on market access to basic telecommunications services.⁸⁷ It also prohibits discriminatory competition conditions within the markets and prevents anticompetitive practices among dominant suppliers.⁸⁸ Key provisions include the following:

• Relevant Market and Dominant Supplier: The Telecom Reference Paper defines “major supplier” as a supplier that has the ability to materially affect the terms of participation surrounding price and supply in the relevant market for basic telecommunications services as a result of: (a) control over essential facilities; or (b) use of its position in the market.⁸⁹

• Anticompetitive Practices: The Telecom Reference Paper imposes obligations on WTO members to maintain measures for the purpose of preventing suppliers – which alone or together are major suppliers – from engaging in or continuing anticompetitive practices.⁹⁰

• Interconnection Arrangement and Transparency: There is a clear stipulation that interconnection with a major supplier should be provided under nondiscriminatory terms, conditions, and rates, and should be of a quality no less favorable than that provided for its own like services or its subsidiaries.⁹¹

In brief, the Telecom Reference Paper requires WTO members to ensure that dominant companies do not abuse their market position. As discussed in Chapter 1 of this book, the case of Mexico – Telecom represents a concrete application of competition policy within the framework of the Telecom Reference Paper.⁹² In this case, the US claimed that the interconnection rates negotiated by Telmex, the incumbent supplier in Mexico, were not cost-oriented. The panel found that Mexico had failed to fulfill its commitments under Section 2.2(b) of the Telecom Reference Paper, in that it did not ensure a major local supplier to provide interconnection at cost-oriented rates to other members’ suppliers for the cross-border supply of telecommunications services.⁹³ The panel also found that Mexico had not met its GATS commitments under Section 1 of the Telecom Reference Paper to maintain “appropriate measures” to prevent anticompetitive practices.⁹⁴

By pointing to the model in the Telecom Reference Paper, this book raises the following questions: To what extent is a set of sector-specific competition disciplines for the data industry possible? Further, what should comprise the “Data Reference Paper?” Turning back to the common elements of the regulatory recommendations by the OECD, the EU, and the Australian competition authority, the proposed Data Reference Paper should be a binding set of commitments, perhaps the lowest common denominator, which would serve to guide WTO members to better regulate data, to discipline dominant players, and to thereby help smaller tech companies enter these markets. Much like the regulatory disciplines for the telecommunications market, the concept of “essential facilities” might be applied to big tech companies to prevent the abuse of market dominance by platforms. Similarly, based on the model of the Telecom Reference Paper, a Data Reference Paper would impose obligations on WTO members to maintain measures for the purpose of preventing dominant services suppliers from engaging in or continuing anticompetitive practices. Appropriate mechanisms to prevent collusive practices and review mergers should also be put into place. In addition, a similar focus on pro-competitive effects could include the principles of nondiscrimination and transparency, which would prohibit big platforms from engaging in self-preferential practices that favor their own services.⁹⁵

By imposing cross-border disciplines for competition policy and thus curbing the power of big digital platforms, the proposed Data Reference Paper may well be an effective instrument to address data colonization, which, as discussed in Chapter 3, is an unforeseen phenomenon that interacts with GATS digital trade market access. Moreover, if a set of international competition rules that frame competition concerns in a policy context can be established, there would be less need for ex-post enforcement of competition law by competition authorities in developing countries and LDCs, which have relatively limited resources to tackle issues pertaining to digital cartels and data monopolization.⁹⁶ To conclude, the increasing inequality in datafication, and in particular the dissymmetric power between those who are the sources of the data and those who accumulate the data, calls for a set of WTO data-specific competition rules to appropriately address market power in the data sector. There is a renewed need for a WTO Reference Paper 2.0, which migrates the competition disciplines from the context of the telecommunications industry to that of the data industry.

5.2.3.3 The Inherent Complexity: Sufficient Momentum Needed

Will the need for international competition disciplines for the data sector find an outlet along the path of international economic law, as it did in the telecommunications sector two decades ago? Serious challenges lie ahead. The idea of creating a WTO “Data Reference Paper” may prove difficult in gaining sufficient negotiating momentum to bring it to fruition, primarily because of two structural problems. The first main obstacle is the highly complex, legally technical nature of regulating the digital market. To the extent that the regulatory principles spelled out in the Telecom Reference Paper can inform the development of the data regulatory framework, some adaptations are needed due to the specific characteristics of the data market. As pointed out by the Ofcom, the UK’s communications regulator, the regulatory principles of telecommunications services cannot simply be “read-across” and “applied as they exist” to digital services.⁹⁷ There are significant similarities between the telecommunications and digital markets. Nevertheless, substantial differences remain.

More specifically, in terms of assessing the “relevant market,” far more factors must be taken into consideration when defining the relevant market for digital platforms. All of the “big tech” firms are characterized as multi-sided, which renders the scope of the relevant market even more difficult to define. What constitutes the relevant market of a digital platform inside the big data ecosystem, when various players are involved and have assumed multiple roles? For example, Apple – as a digital platform through the Apple Store and iTunes – also plays an important role in cloud computing services using the iCloud. At the same time, Apple closely interacts with other key social media businesses, including Facebook and LinkedIn. Should each side of the above be defined as a separate market?⁹⁸ The multi-sided platform structure poses new difficulties for competition regulations.

Moreover, in terms of assessing abuses of market power, determining “market power” is less straightforward in digital markets. In the case of telecommunications services, dominant market position and significant power are closely related to natural monopolies in physical infrastructure, that is, network facilities. Digital services, however, are not necessarily natural monopolies, because their market powers are primarily derived from their access to large datasets created from their users.⁹⁹ In practice, market shares in telecommunications markets (i.e., the 25 percent threshold) typically provide useful indications of market importance.¹⁰⁰ In most jurisdictions, a telecom operator is presumed to have significant market power when it holds more than a 25 percent share of a market in a particular geographical area.¹⁰¹ On the other hand, the possession of data can be used as a barrier to entry, thus becoming the primary source of market power in digital services. The relationship between “market share” and “control over data,” however, would prove difficult for competition authorities to investigate.¹⁰² In this regard, the “qualitative criteria” set out by the DMA have been criticized as “arbitrary thresholds”¹⁰³ that lack “methodological considerations.”¹⁰⁴ It is particularly problematic that the designation of gatekeepers is based on the presumption that the “qualitative criteria”¹⁰⁵ will be met by a firm if it meets the “quantitative criteria.”¹⁰⁶ In summary, traditional measuring tools, such as market shares, must be adapted in a digital platform context. All of this highly technical complexity will lead to endless technical discussions and will become an obstacle toward the goal of creating a set of international competition principles for digital services.

Another equally or even more important consideration that may impede the creation of such international disciplines is the inherent complexity of the political economy surrounding digital capitalism. Looking back at its history, the telecommunications industry began to rapidly develop in the late 1990s. As a result, the political momentum toward telecommunications liberalization rendered market access and regulatory discipline under WTO negotiations possible. In other words, adoption of the Telecom Reference Paper was seen by “key” delegations – notably, the US, the EU, Canada, Australia, and Japan – as necessary,¹⁰⁷ given the risk that competition in foreign countries’ infrastructure markets may be restricted by incumbent operators’ abuses of market power. To illustrate, the telecommunications market, especially decades ago, has exhibited specific features that enable incumbents to maintain a certain degree of market power over the competition.¹⁰⁸ Major incumbent suppliers have strong incentives and ample opportunities to delay the provision of interconnection to new entrants, and such delays can significantly inhibit competition.¹⁰⁹ The incumbents could also, for example, impose anticompetitive interconnection conditions on their competitors.¹¹⁰ National measures might be needed to prevent incumbent operators from using their market power to distort competition. From the perspective of international trade, market access commitments alone cannot guarantee that a market will become truly liberalized. To be able to effectively compete, telecommunications companies of developed countries must be ensured a level playing field in foreign markets.

However, such political economy momentum, which led to the conclusion of the Telecom Reference Paper, cannot be found in the context of digital services. Unlike the negotiation background of the telecommunications services industry, the economic interests (as well as the regulatory approaches) of the data services industry are quite divergent among key players. Generally speaking, US digital platforms have been persistently dominant in the world, including the European market. At the same time, China – by establishing its own self-sufficient platform economy through Chinese digital giants Baidu, Alibaba, and Tencent¹¹¹ – has largely escaped US domination. Nonetheless, three different models for data governance are emerging: The US generally favors an ex-post approach, which broadly seeks punitive action for past infractions. Such an innovation-friendly approach is primarily driven by the concept of self-regulation.¹¹² The EU model, as discussed above, is holding the normative high ground. The ex-ante regulations would result in sweeping supervisory actions that impact Silicon Valley’s future. China’s main concern, however, is to ensure its political stability and security. It is conceivable that China will continue to rely on domestic protectionist regulations to restrict cross-border data flows.¹¹³ When these three models interface in an international organization, it is less likely to result in a compromise given the associated concerns.¹¹⁴

5.2.3.4 Digital Trade and Competition Disciplines

How can international economic law contribute to overcoming these impediments? The first direction is the soft law mechanism, which leaves sufficient space for national regulators. Here again, the FTAs provide some inspiration. Although to date none of the Digital Trade/E-Commerce Chapters of the FTAs have incorporated competition rules for the data market, the lesson we have learned from their general approaches pertains to the soft legal nature of key provisions.¹¹⁵ The USMCA parties, for example, merely “recognize the importance” and “endeavor to” comply with certain rules under the Digital Trade Chapter.¹¹⁶ It might therefore be criticized as a weak instrument. Nevertheless, it can always be argued that without such vague provisions, the Digital Trade Chapter would never have been finalized by the parties. In future trade negotiations on data governance, the “softness” of a treaty requires that substantive rules remain somewhat general. For example, it might be necessary to leave key concepts such as “anticompetitive practices” undefined to allow for policy alternatives. The lack of specificity in the treaty language would allow parties to cater to differences in local needs and maximize the likelihood that the rules will be effectively implemented by regulators. Given the variations that exist in the digital markets of different countries, the strategic use of hard and soft law is of practical significance in introducing a set of data rules into the WTO regime.

The second approach concerns flexible negotiating modality, which helps to reach a critical mass of trade negotiation results. Against this contentious political and economic backdrop, the probability of reaching a consensus under the “single-undertaking” system seems slight.¹¹⁷ Balancing the interests of 164 WTO members across diverse issues surrounding data governance has made it difficult, if not impossible, to conclude negotiations that “bind all WTO Members equally.” In this regard, negotiating on the basis of a critical mass approach, which involves arrangements between a number of parties that do not represent the entire membership but account for a very high proportion of international trade in data services, seems to be a more realistic direction. In this context, despite strong opposition from several members,¹¹⁸ the ongoing JSI on E-commerce offers a pathway for the WTO to remain responsive and relevant in the digital economy.¹¹⁹

5.3 Trade Rules and Algorithmic Transparency

5.3.1 Algorithms and Platform Competition

A closely related but distinct aspect is algorithmic transparency, which can serve as a regulatory starting point for global platform governance. In this data-driven world, one important legal tool to promote platform competition is to ensure that algorithms are fair and transparent.¹²⁰ As illustrated above, platform capitalism would not ever have existed without big data analytics algorithms. In summary, the market dominance of big tech can be simplified in the following pattern. By taking advantage of their vast user base, big tech companies provide datafication-enabled advertising services to businesses, which in turn bring big tech advertising revenues. Big tech companies then invest more in their data analytics tools to enhance the accuracy of target advertising, which allows big tech to attract more users and collect more data. The more data that is collected, the more the algorithms can learn.¹²¹ Under this pattern, with more and more data amassed, big tech companies are able to attract even more advertisers. The cycle continues.

To put it another way, each time we Google a product and then click on the sponsored ads, Yahoo not only loses a search query to Google, but also loses our behavioral data, which represents an opportunity for its algorithms to learn, as well as potential advertising revenues. This results in a gap between Google and Yahoo in their ability to train their AI, to personalize search results, and to enhance the accuracy of targeted advertising. The more people Google, the wider such gaps grow.¹²² In this context, concerns have been raised in relation to the abusive use of algorithms. After all, it may amount to the unfair treatment of competitors if algorithms are programed – wittingly or unwittingly – to prioritize a platform’s own services over those of competitors.

Accordingly, increasing consideration is being given to why an algorithm of a search engine makes certain recommendations, as well as how a consumer can achieve different search results or access different ranking systems. In addition, turning back to the social media moderation practices discussed in Chapter 4, AI plays a material role in content moderation, including some context-specific content such as hate speech. When our behavioral data becomes algorithmic input, it is important to empower consumers to understand how a social media algorithm makes content moderation decisions, whether it is appropriate to use that content-moderating algorithm, and how consumers can receive objective algorithmic decisions.

More and more governments and civil societies are addressing commensurate algorithmic practices and calling for algorithmic transparency. They express caution over the unfair outcomes due to the lack of algorithmic transparency and advocate for algorithmic accountability, primarily to require platforms to disclose how their algorithms generate certain outputs.¹²³ In this regard, disclosing the algorithms or the source code that drives a platform’s system may reveal whether the platform values one type of content over another. Alternatively, a lower degree of transparency that informs users of the main parameters that determine ranking, using plain language, may also be useful for the public in understanding how search engines work. Conversely, as discussed in greater detail below, some experts have pointed out that such transparency requirements may prove neither meaningful nor feasible due to the protection of trade secrets and the nature of machine learning.¹²⁴ The design of algorithmic systems is the product of substantial investment and may easily be duplicated by competitors if disclosed.

At this moment, regulators all over the world are contemplating various forms of regulation that will improve platform accountability, AI ethics, and algorithmic transparency and explainability for automated decision-making.¹²⁵ Algorithms can be governed by various regulatory choices, including who and what will be regulated, to whom information will be disclosed, and how disclosure will occur.¹²⁶ In the context of international economic law, the interplay between AI regulation and international trade agreements centers on questions relating to how to ensure that algorithmic accountability is appropriately distributed, as well as how to safeguard public access and oversight over algorithms. More profoundly, how can we balance competing interests in particular trade secrets and prevent AI regulations from being overly trade restrictive?

5.3.2 Platform Transparency Requirements

To answer the above questions, five “representative” digital/data regulations – including the DSA, the DMA, the GDPR, the Montréal Declaration for a Responsible Development of Artificial Intelligence (Montréal AI Declaration),¹²⁷ and the US bill on the Algorithmic Justice and Online Platform Transparency Act (Algorithmic Justice Act)¹²⁸ – are selected for an in-depth investigation. Transparency requirements in the five digital/data regulations, including both hard law and soft law, are reviewed to reveal the variables and their relationships. To clarify, this approach provides examples and is not meant to be comprehensive, as there are plenty of AI-related regulations or draft bills floating around,¹²⁹ but it aims to serve as an illustration to better understand the key dimensions of platform transparency requirements in a comparative context.

5.3.2.1 Who: The Scope of the Regulatory Targets

As noted in Chapter 4, obligations under the DSA are cumulative depending on the function and size of a services supplier. The larger a services supplier is, and the more extensive the services it provides, the greater the number of obligations that apply. In terms of transparency, the DSA imposes a baseline transparency requirement, which applies to all intermediary services suppliers, and it then adds obligations for hosting services and online platforms, and then once again adds a further set of transparency obligations for VLOPs. In other words, VLOPS must comply with all of the transparency requirements imposed by the DSA, including transparency reporting,¹³⁰ transparency of recommender systems, transparency of online advertising,¹³¹ and data sharing with authorities and researchers.¹³²

Similarly, as stressed in the DMA’s preamble, in most cases, gatekeepers provide online advertising services to business users in a nontransparent manner, which results in higher costs for online advertising services. To address this problem, it is important to ensure that the platform environment, and in particular, the conditions that apply to advertising services and rankings, is generally fair, transparent, and contestable. Therefore, the transparency obligations primarily fall to the gatekeepers.¹³³

Unlike the asymmetric regulatory models of the DSA and the DMA, the scope of the regulatory targets of the GDPR, the Algorithmic Justice Act, and the Montréal AI Declaration is broader and more general. This is because concerns surrounding the lack of platform transparency extend to all kinds of platforms, regardless of their size.¹³⁴ In this regard, under the GDPR, provisions on platform transparency and accountability apply to all data controllers and processors, as defined in Article 4.¹³⁵ Likewise, the principles in the Montréal AI Declaration are general and abstract and aim to apply to the digital and artificial intelligence field in a broad sense.¹³⁶ As broad as it may sound, however, the nonbinding nature of the declaration means that implementation is on a voluntary basis. Finally, the Algorithmic Justice Act imposes transparency obligations on an “online platform,” which is defined as “any public-facing website, online service, online application, or mobile application which is operated for commercial purposes and provides a community forum for user generated content”¹³⁷ – a rather broad net.

5.3.2.2 What and to Whom: The Degree of Transparency

The five regulations vary in terms of what type of information should be transparent, ranging from algorithm-based advertising, algorithmic content moderation, algorithmic reporting, factors and parameters shaping algorithmic decisions, auditable records of the algorithmic process, and source code. Indeed, the degree of disclosure is controversial, and the policy options involve trade-offs. Unfettered government access to source code and algorithms may help to ensure that algorithms are fair but may also lead to unnecessary intrusion into privacy and trade secrets.¹³⁸ Taken to the extreme, platforms may be required to disclose source code.¹³⁹ However, would it be more proportionate to require them to make their algorithms transparent? Or is it more reasonable for platforms to merely publish the general factors and logic involved in algorithmic decisions? These questions must be analyzed in conjunction with the question of who receives transparency. As discussed below, the question of “disclosure to whom” dictates “what to disclose.”

▪ Authorities and Experts

As shown in Figure 5.1, platforms’ disclosure obligations to regulatory bodies generally rank high on the scale when measuring transparency. To illustrate, transparency can be restricted to governmental authorities. The rationale is that authorities need to use these records to verify compliance with regulatory requirements. On this matter, the GDPR focuses on supervisory authorities’ investigative powers to obtain access to any premises of the data controller and the processor, including “any data processing equipment and means.”¹⁴⁰ The Montréal AI Declaration, although nonbinding in nature, stresses the need for the code for algorithms to be accessible to relevant public authorities for verification and control purposes.¹⁴¹ The Algorithmic Justice Act also calls for platforms to maintain records of algorithmic processes, and to make available to the supervisory agency complete records upon request.¹⁴² After all, the algorithm itself may only be part of the story. Algorithmic decision-making cannot be accurately audited by reviewing the algorithms alone. It is important to ensure that the data on which an algorithm is trained is equally accessible.

Figure 5.1 Transparency requirements in context

Notes:

–Degree of Disclosure Obligations (Rows 1–3)

H: high (all data including algorithms and source code)

M: medium (algorithm parameters, algorithm auditing)

L: low (transparency reports, standard contracts)

N: none (no explicit rules)

–Degree of Explainability (Row 4)

H: high (inspections and interviews)

M: medium (machine-readable formats)

L: low (meaningful and comprehensible descriptions)

N: none (no explicit rules)

In practice, however, the disclosed information, whether algorithms or source code, may not be understood by regulators, let alone the general public. Government agencies in most circumstances do not have the technical expertise to keep pace with and thus oversee the industry’s sophisticated algorithms, which represents an impediment that would negate many benefits of disclosure.¹⁴³ Disclosure requirements may therefore need to involve specialized experts in carrying out necessary analyses.¹⁴⁴ In other words, authorities most likely rely on experts who are in a position to make assessments of the disclosed technical details. In this regard, the DSA explicitly stipulates that auditors and experts appointed by the EC during administrative inspections may require VLOPs to provide necessary information, including their information technology systems and algorithms.¹⁴⁵ Similarly, the DMA states that the Commission may require access to any data and algorithms of undertakings and information in order to carry out its duties.¹⁴⁶ Persons working under the supervision of the authorities, including auditors and experts appointed pursuant to the DMA, shall not disclose information acquired during inspections.¹⁴⁷ Limiting transparency to regulators and researchers, but not to the general public, appears to be an important mechanism by which to protect potentially sensitive information and data that would not be considered releasable to the public.

▪ Interested Parties

Variations can be found in platforms’ disclosure obligations to interested parties. In Figure 5.1, the degree of the obligations ranks low in the DSA and the DMA, medium in the GDPR, high in the Montréal AI Declaration, and is nonexistent in the Algorithmic Justice Act. To illustrate, both the DSA and the DMA have transparency provisions that would be available to interested parties, but the obligations are limited to relatively general information, without associated technical details. For example, regarding content moderation, a platform should inform the recipient at the time of the removal of the decision through “a clear and specific statement of reasons” when it removes specific content.¹⁴⁸ Likewise, regarding advertising and ranking, a gatekeeper shall provide each advertiser to which it supplies advertising services with information regarding prices, fees, and remunerations. The gatekeeper shall also provide advertisers and publishers with access to “the performance measuring tools of the gatekeeper and the data necessary for advertisers and publishers to carry out their own independent verification of the advertisements inventory, including aggregated and non-aggregated data.”¹⁴⁹

On the contrary, the Montréal AI Declaration empowers “stakeholders and those affected by the situation” the same opportunity to access “the code for algorithms” for verification and control purposes.¹⁵⁰ The GDPR, in this regard, requires the data controller to provide individuals with the information necessary to ensure fair and transparent data processing. Where automated decision-making is involved, the disclosure obligations include the provision of “meaningful information about the logic involved” and “the envisaged consequences of such processing” to individuals.¹⁵¹ A great deal of literature has explored what constitutes “meaningful information.”¹⁵² In any event, it can be presumed by reading in context that the information required under Articles 13 and 14 of the GDPR, although it need not be source code or complex algorithms, must be more than a general overview of the decision-making system – sufficiently concrete for the data subject to understand the reasons for a specific decision.¹⁵³ It therefore ranks medium in terms of the degree of transparency.

▪ The Public

Finally, in connection with disclosure obligations to the public, among the five selected regulations, three of them rank medium, while two rank low. As revealed in Figure 5.1, the transparency requirements toward the public under the DSA, the Montréal AI Declaration, and the Algorithmic Justice Act extend to the level of algorithms’ parameters and auditing, whereas those in the DMA and the GDPR remain at the level of general transparency reporting and standard contracts. More specifically, under the DSA, VLOPs are required to publish transparency reports at least every two months, and to release, in their terms and conditions, “the main parameters used in their recommender systems, as well as any options for the recipients of the service to modify or influence those main parameters.”¹⁵⁴ The “main parameters” include but are not limited to “the criteria which are most significant in determining the information suggested to the recipient of the service” and “the reasons for the relative importance of those parameters.”¹⁵⁵ In a similar manner, the Montréal AI Declaration calls for public transparency of “the social parameters of the artificial intelligence systems,”¹⁵⁶ while the Algorithmic Justice Act advocates that the online platform should disclose, among other matters, “the method by which the type of algorithmic process prioritizes, assigns weight to, or ranks different categories of personal information to withhold, amplify, recommend, or promote content to a user.”¹⁵⁷

In terms of a textual comparison, public transparency obligations under the DMA and the GDPR are less specific and are therefore weaker. Under the DMA, gatekeepers are required to publish and update nonconfidential summary of transparency reports, as well as make publicly available and update an overview of the audited descriptions.¹⁵⁸ The obligation is even less clear in the GDPR, under which the EC has the power to “lay down standard contractual clauses” for data processing. In addition, associations representing platforms may prepare codes of conduct addressing “fair and transparent processing.”¹⁵⁹ On the whole, the degree of transparency to the broad public is relatively low in these two instruments.

5.3.2.3 How: The Degree of Explainability

Some skeptics emphasize that algorithmic transparency requirements may lend themselves to legal problems, because the algorithms themselves cannot explain “how they have arrived at a particular output.”¹⁶⁰ For example, releasing algorithms does not necessarily explain how they work and may not provide much meaningful information about, for example, the content moderation or ranking systems. Any degree of meaningful algorithmic transparency must therefore be accompanied by the requirement of explainability, that is, providing insights into the functioning of the algorithms “in a format that makes sense to the reader.”¹⁶¹

The question is how. First of all, accurately explaining the functioning of algorithmic systems – particularly those developed using AI – is not always technically feasible. Algorithmic decision-making may involve complex and dynamic processes, as well as multiple profiling elements and data sources.¹⁶² It may be commercially impossible to carry out an effective review.¹⁶³ Moreover, disclosure of incomprehensible technical details, even when available in machine-readable format, does not necessarily empower either the regulators or the interested parties to understand how they actually work.¹⁶⁴ Accordingly, as a subset of transparency requirements, the degree of explainability strongly correlates with to “whom to explain.” As for the regulators, the power to acquire insights into how and why the algorithms work in certain ways by conducting inspections and carrying out input and output auditing is useful for supervisory purposes. Conversely, as for the general public, digital platforms’ descriptions of algorithms’ objectives, processes, and functioning, in plain language that is intelligible to lay people, represent meaningful information responsive to the right to know.¹⁶⁵

Both the DSA and the DMA require different levels of explainability depending on “whom to explain to” and “why explain.” The two instruments impose upon platforms the obligation to publish annual transparency reports “in clear, easily comprehensible language” and “in a machine-readable format.”¹⁶⁶ Moreover, the EC has the power to conduct all necessary inspections at the premises of the VLOPs/gatekeepers, including requiring the services suppliers to provide access to and explanations of their “organization, functioning, IT system, algorithms, data-handling and business practices,” and allowing regulators to “address questions to their key personnel,” and “record or document the explanations given.”¹⁶⁷ The EC can also require access to any data and algorithms and request explanations of them.¹⁶⁸ Furthermore, the EC is mandated to carry out interviews and take statements during the investigation and is entitled to “record such interviews by any technical means.”¹⁶⁹ As shown in Figure 5.1, the VLOPs/gatekeepers’ cumulative obligations for explanations rank high in the given context. In contrast, requirements that algorithmic decision-making is explainable are relatively light in the Montréal AI Declaration and the Algorithmic Justice Act, which basically obligate services suppliers to provide a description of how the algorithmic process operates,¹⁷⁰ or to justify AI-based decisions in plain language that is understood by individuals whose quality of life and reputation are affected by the consequences of the algorithmic process.¹⁷¹

In this regard, it is worth noting that the text of the GDPR has provoked debate over whether it imposes upon data controllers an obligation of explainability when algorithmic decision-making is involved. Article 22 states that individuals “have the right not to be subject to a decision based solely on automated processing, including profiling” if such a decision has “legal effects or similarly significant effects” on concerned individuals. The same article requires a services supplier to “implement suitable measures” to safeguard individuals’ “rights and freedoms and legitimate interests.”¹⁷² This requirement has been a source of controversy over whether algorithmics should be explainable. Note that suitable safeguards, according to Article 22, must include “at least the right to obtain human intervention on the part of the controller, to express his or her point of view and to contest the decision.” The term “at least” indicates that this is not an exhaustive list of rights. Namely, it is an open list of suitable safeguards, and a services supplier is expected to do more than the basic requirements. Additionally, Recital 71 has added that the suitable safeguards are to include a right to an explanation of an automated decision. This textual interpretation thus creates a sui generis due process for algorithmic practices – a right to be heard, to be given an explanation, and to challenge an automated decision.¹⁷³

In any event, just as the DSA and the DMA empower authorities to demand a high level of explainability for supervisory purposes, Article 58 of the GDPR states that authorities’ investigative powers include access to any premise of the services supplier, any data processing equipment and means, as well as the ability to order the services supplier to “provide any information it requires for the performance of its tasks” – an outright obligation of explainability.¹⁷⁴ For this reason, the GDPR ranks high in terms of explainability, as shown in Figure 5.1, which denotes the cumulative degree of a given regulation.¹⁷⁵

5.3.2.4 Variables and Comparison

To summarize, the five regulations selected implement various transparency measures, depending on the regulatory objectives and their context. Generally speaking, of the four dimensions assessed in Figure 5.1, the transparency obligation to authorities and experts appears in all five regulations and is the most significant requirement in terms of what must be disclosed. It is worth noting that the most evident variations are seen in transparency obligations to the interested parties and to the public. Moreover, the degree of technical detail required for explanation also varies greatly, from detailed auditable trails of algorithmic processing to general, plain-text descriptions. However, consistency has been found in the DSA, the DMA, and the GDPR, as they contain identical provisions requiring a high degree of explanation. To conclude, this case study demonstrates that the emerging regulations bring about greater transparency and oversight of the algorithms. The legal-technical details, however, differ in important ways. Although all of the five instruments place pressure on digital platforms to be more transparent, what kind of information should be provided and how it should be provided are dependent upon context, as well as to whom the disclosure is made.

Figure 5.1 measures the key elements of platform transparency and explainability reflected in the selected regulations, which form the basis of the mapping exercise in Figure 5.2. Before this mapping exercise, one caveat is in order. Issues arise when comparing values across these axes, because each has a unique measuring scale. Note especially that the variables on the explanation obligations (row 4) represent different measuring scales. In other words, variables on the axis of explanation obligations are independent. With all of this in mind, Figure 5.2 shows the four variables mapped onto axes. The central axis is defined as “none,” the second central axis as “low,” the inner axis as “medium,” and the outer axis as “high.” As such, the areas of the polygons represent the overall degree of transparency and explainability requirements. Evidently, the five polygons overlap in some areas but are completely independent in other areas. Additionally, the shapes of individual polygons vary greatly. This case study thus concludes that the fragmentation of platform governance within and across jurisdictions is growing. Section 5.4 will continue to address how international trade agreements can help curb such fragmentation. However, let us first dive into the relevant provisions on nondisclosure of source code and algorithms in the FTAs. What are the relevant provisions in international trade agreements, and how, if at all, do they affect a state’s ability to regulate platform?

Figure 5.2 Regulatory fragmentation of transparency requirements

Notes:

–Central axis: none

–Second central axis: low

–Inner axis: medium

–Outer axis: high

5.3.3 Trade Rules on Source Code Disclosure

5.3.3.1 Source Code Nondisclosure Provisions under the FTAs

A growing number of FTAs contain provisions to restrict access to source code. Article 14.17 of the CPTPP, for example, prohibits “the transfer of, or access to, source code” as a condition of the “import, distribution, sale or use” of software.¹⁷⁶ Essentially, source code nondisclosure provisions aim to prevent states from requiring technology transfer in exchange for international trade. In the context of digital platforms, services suppliers generally have invested resources in the development of source code. If they are required to disclose the source code as a condition of market access, they risk exposing their technologies to competitors and losing their competitive advantage. Source code nondisclosure provisions therefore safeguard software owners’ rights over their intellectual property against mandatory disclosure.¹⁷⁷

Traditionally, trade secret law has been the most widely used legal mechanism to protect software source code. Nonetheless, the specific expression of innovative software ideas can also be protected through other intellectual property rights, such as copyrights and patents. In this regard, the WTO Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS) has for some time been the most relevant legal tool in terms of source code protection under the realm of international economic law.¹⁷⁸ Source code nondisclosure protections under the FTAs, however, go beyond the TRIPS by directly barring governments from demanding the disclosure of source code as a prerequisite for market access, subject to exceptions. Arguably, the trend to include the ban on the mandatory disclosure of source code provisions in the FTAs stems from the emerging need for greater source code protections, and in particular, in response to some state actions, such as China’s policy of forcing foreign firms to disclose their source code and/or “other parts of their intellectual property” as a condition for doing business in China.¹⁷⁹

In addition to source code, several FTAs extend the general prohibition on disclosure requirements to algorithms. For example, Article 19.16 (Source Code) of the USMCA bans mandatory transfers and access to both the source code and the algorithm expressed in it, stating the following:

1. 1. No Party shall require the transfer of, or access to, a source code of software owned by a person of another Party, or to an algorithm expressed in that source code, as a condition for the import, distribution, sale or use of that software, or of products containing that software, in its territory (emphasis added).

2. 2. This Article does not preclude a regulatory body or judicial authority of a Party from requiring a person of another Party to preserve and make available the source code of software, or an algorithm expressed in that source code, to the regulatory body for a specific investigation, inspection, examination, enforcement action, or judicial proceeding, subject to safeguards against unauthorized disclosure¹⁸⁰ (emphasis added).

Here, an algorithm is broadly defined in these agreements as “a defined sequence of steps, taken to solve a problem or obtain a result.”¹⁸¹ At the time of this writing, similar text could be found in Section C.3 (1) – Source Code of the WTO JSI on E-commerce.¹⁸² The inclusion of algorithms substantially broadens the scope of the source code nondisclosure provisions. By preventing the mandatory disclosure of algorithms in addition to that of source code as a condition for market access, the USMCA-type of source code nondisclosure provision makes it clear that it also prohibits requirements to disclose algorithms. Unless justified by exceptions, a USMCA party seeking access to the algorithms used by digital platforms as a condition for digital trade market access would be inconsistent with Article 19.16 of the USMCA.

The expanded scope of the prohibition on disclosure requirements thus raises concerns about its impact on the ability of governments to regulate the use of algorithms at the domestic level. Civil societies such as the Australian Council of Trade Unions stressed that preventing relevant authorities from accessing source code and algorithms altogether would unduly restrict governments from supervising platforms’ compliance with domestic regulations and ensuring algorithmic accountability.¹⁸³ They believed that keeping source code and algorithms secret also makes it difficult for the public to understand how the algorithms make decisions, and whether there is bias or discrimination within the process.¹⁸⁴ Indeed, compared to FTA provisions that only preclude the accessibility of source code, FTA provisions that additionally preclude the mandatory accessibility of algorithms place further restrictions on the power of authorities to protect consumers.¹⁸⁵ It is notable that the key source codes and algorithms that drive datafication are primarily owned by the digital platforms, and in particular the big tech companies. Some critics have therefore asserted that the consequences of restrictions to algorithm access would jeopardize the ability of governments to “develop regulatory measures that could ensure transparency of algorithmic governance tools.”¹⁸⁶ In their view, the expanded scope of source code nondisclosure provisions in international economic law conceivably protects private capital – the source code and algorithms owned by the digital platforms – at the expense of governments’ full regulatory autonomy to supervise legal compliance.¹⁸⁷

5.3.3.2 The Supervision of Algorithms

To the contrary, some believe that the USMCA-type source code nondisclosure provisions, given their broad scope to cover “vital digital assets,” can better help to enhance business trust and protect foreign firms from unauthorized disclosure.¹⁸⁸ In the view of Mitchell and Mishra, a diligent balance has been achieved in USMCA Article 19.16 by restricting governments from forcing the disclosure of both source code and algorithms on the one end, and ensuring that governments can place demands on foreign firms to disclose their source code and algorithms for accountability and regulatory purposes, on the other end.¹⁸⁹ Specifically, attention should be given to the second paragraph of USMCA Article 19.16, which explicitly states that the provision does not preclude a regulatory body or judicial authority from requiring access to source codes or algorithms “for a specific investigation, inspection, examination, enforcement action, or judicial proceeding.”

In fact, such a similarly worded regulatory/judicial preclusion can be found in the source code nondisclosure provisions of several international trade agreements, such as the US–Japan Digital Trade Agreement, the Australia-Singapore Digital Economy Agreement (DEA), and the Korea-Singapore Digital Partnership Agreement (KSDPA).¹⁹⁰ Moreover, as a subset of regulatory/judicial preclusion, the EU-Japan Economic Partnership Agreement and EU-New Zealand FTA, for example, also allow a court, an administrative tribunal, or a competition authority to require access to source code owned by foreign firms to remedy a violation of competition law.¹⁹¹ Altogether, source code nondisclosure provisions do not seem to compromise the regulatory autonomy on digital platforms, because governments still retain the power to investigate and audit automated content-moderating systems, biased and discriminatory algorithms, and algorithmic recommendation and ranking processes.

It should also be emphasized that the source code nondisclosure provisions in the FTAs target the market access of software businesses, namely, “import, distribution, sale or use of software,” for supplying international trade. Essentially, these provisions are centered on “market access” and are not designed to constrain governments from requiring information to investigate compliance. It can be argued that a country’s right to regulate digital platforms would not be adversely affected by these provisions as long as the regulatory actions do not block the market access of foreign services. Turning back to the first row of Figure 5.1, platforms’ disclosure obligations to the regulatory body generally rank high on the scale when measuring transparency. For instance, Article 69 of the DSA mandates that regulators conduct inspections, including requiring access to the algorithms of the very large online search engines. In a similar vein, Articles 21 and 23 of the DMA empower regulators to require access to all necessary information for the inspection, including any data and algorithms. One could contend that such inspection power under the DSA and the DMA does not necessarily constitute a condition for market access and therefore is consistent with the FTAs’ source code nondisclosure provisions.

Nonetheless, as discussed in Section 3.5, there are situations in which the boundary between market access prerequisites and non-market access regulations is blurred. To some extent, any transparency requirement, no matter how “pure” its design for regulatory purposes, would potentially affect the “distribution, sale, or use” of software in the given market. Therefore, it is disputable how far the FTAs’ nondisclosure provisions can go in this regard. In any event, assuming that any regulatory action had been broadly construed as a market access prerequisite, the government would have to rely on the carve-out and exceptions to justify the mandatory disclosure of the source code and algorithms.

First turning to the carve-outs for critical infrastructure services: Some source code nondisclosure provisions allow for the forced disclosure of information relating to critical infrastructure.¹⁹² The term “critical infrastructure,” however, was left undefined in the Digital Trade/E-Commerce Chapter. Accordingly, what the concept of “critical infrastructure” might cover would be interpreted pursuant to the previous discussion in Chapter 2.

Second, the general exceptions and the security exceptions are available in all of these provisions.¹⁹³ In my view, even if measures that are necessary to enhance platform accountability and maintain digital legal order were found to fall within the scope of market access-related restrictions, they could be justifiable under exceptions, and in particular, the public morals general exceptions. As previously discussed in Chapters 1 and 2, the country implementing the disputed transparency requirements would have to prove that the measure at issue satisfies the necessity test and the Chapeau test under the public morals exceptions. It is worth reiterating that according to WTO jurisprudence, a country has the regulatory autonomy to choose its level of protection in relation to the objective being pursued, which is, in this context, fair and transparent algorithmic decision-making in a datafied business environment. Any less trade-restrictive alternatives proposed by the governments of foreign platforms must qualify as reasonably available alternatives under the necessity test.

At the end of the day, the ambit within which source codes and algorithms can be kept secret under international economic law relies on the exceptions, entailing a delicate balance between the competing interests involving platforms’ trade secrets, digital innovation, algorithmic accountability, and other legitimate public objectives. With that in mind, this book argues that international regulatory coherence and cooperation surrounding platform regulations should be an important direction for global governance in a platform economy. Good regulatory practices (GRP) with ex-ante regulatory impact assessment (RIA), when appropriately implemented, can help identify and address unnecessary trade restrictions, forestall potential trade disputes at earlier stages, and reduce the likelihood of lengthy and costly dispute settlement procedures. Section 5.4 will continue to discuss the opportunities and challenges of such an approach to global platform governance.

5.4 Good Regulatory Practices for Platform Governance

5.4.1 Potential Fragmentation and Overreaching

Drawing from the analysis above, the fragmentation of platform regulation – whether it relates to content moderation, competition policy, or algorithmic transparency – is growing. The proliferation of platform regulations and algorithmic disciplines means that services suppliers are confronting more and more difficulties when attempting to comply with diverse national “behind-the-border measures.” The lack of regulatory coherence among states may create costly and burdensome tasks for services suppliers, because when operating at international scale, they must meet different degrees of obligation and satisfy divergent legal standards. Such regulatory fragmentation may direct resources away from more effective business management. This is particularly true for non-big tech, small and medium-size enterprises (SMEs) that lack the resources to manage unique legal requirements in different locations. Taking transparency obligations as an example, as shown in Figure 5.2, complexity surrounding various legal approaches may become a barrier to digital trade, especially for SMEs when those rules apply to them. Truly, as Trachtman pointed out, platform regulations of different states have varying concerns and varying exceptions, and thus may prove contradictory in a complex fashion.¹⁹⁴ The emerging fragmentation and possible contradiction of data governance call for cross-border regulatory coherence and cooperation mechanisms to harmonize the divergent regulatory approaches stemming from disparate public policy objectives and digital trade interests.

At the same time, and relatedly, there are more and more examples of potential regulatory overreaching governmental intervention, which render global platform governance even more difficult. Going beyond the selected five regulations in Figure 5.2, several national and local regulations on algorithmic transparency have raised concerns about policymakers’ emerging overreaching practices in this innovation sector. Taking China’s “Qinglang – 2022 Algorithm Comprehensive Governance Special Action” as an example, which aims to keep digital platforms on tight leashes, under this special action, Chinese tech giants – including Alibaba, Tencent and TikTok (ByteDance) – have submitted algorithms used in their services to China’s Internet watchdog.¹⁹⁵ Indeed, digital platform regulations reflect different priorities in different countries, which further complicates the question of how to manage fragmentation in global platform governance.

At the crux of the matter is the desire to promote the development of a more harmonized regime that encourages innovation while protecting other public objectives. Conceptually, as denoted by the Manila Principles, platform regulations should comply with the tests of necessity and proportionality to ensure that platform obligations are linked to a specific public objective, the degree of obligations are proportionate to that objective, and regulators’ powers extend only to the necessary information.¹⁹⁶ This raises the question of whether international economic law can serve as an effective tool to promote regulatory coherence and prevent regulatory overreach. As the global economy goes digital, how can we apply GRP to the domestic regulation given the complex issues involved in a datafied world? Can the regulatory disciplines in the GATS or the FTAs help reduce the likelihood of disguised or unnecessarily restrictive impediments to the platform economy and ease potential regulatory fragmentation and overreaching?

5.4.2 Good Governance Obligations

There has been substantial progress in the development of regulatory discipline through international trade instruments at both the multilateral and regional levels. The ambition of the international GRP agenda is aimed at fostering regulatory coherence by offering a systemic response to the inherent potential problems of regulation that may impose costs but add little overall benefits.¹⁹⁷ At the multilateral level, the WTO JSI on Domestic Regulation of Services (the “DR JSI,” or “DR Reference Paper”) was concluded in December 2021.¹⁹⁸ The participating members, which represent more than 90 percent of the services trade, have agreed on a set of GRP, which will be implemented by incorporating the DR Reference Paper into their GATS Schedules of Commitments. While the DR JSI was negotiated plurilaterally, the outcome will be implemented on a most-favored-nation basis. In other words, the benefits will apply to all WTO members, not just the participants.¹⁹⁹ The DR JSI contains three core principles:²⁰⁰

• Regulatory transparency: measures aimed at promoting prompt publication and availability of information and stakeholder engagement in regulatory processes;

• Regulatory predictability: measures aimed at ensuring a reasonable time frame and procedural due process;

• Regulatory quality: measures aimed at disseminating GRP to facilitate the services trade, including ensuring that regulators develop technical standards through open processes and reach their decisions in a manner independent of services suppliers.

In the context of digital platform regulation, these regulatory disciplines establish legally binding guidelines, within which participating countries have the flexibility to regulate the platforms and pursue their policy objectives. Participating countries retain the right to impose digital/data regulations including algorithmic transparency, which arguably is a “technical standard” within the meaning of Article VI:4 of the GATS, on services suppliers.²⁰¹ However, it must be done in accordance with the regulatory disciplines laid out in the DR Reference Paper – that is, to improve the good governance landscape through regulatory transparency, predictability, and nondiscrimination.²⁰²

At the regional level, the development of GRP under the FTAs can be seen as a response to the issues of regulatory fragmentation and overreaching. Within the last decade or so, good governance obligations have been progressively negotiated, and in some cases, incorporated into bilateral and regional trade agreements.²⁰³ The reality that market access alone cannot sufficiently safeguard services suppliers to operate effectively in foreign markets has led international trade negotiations to address the issue of GRP.²⁰⁴ Notable developments in this regard can be found in the more recent FTAs that include a comprehensive set of good governance obligations in a separate chapter. Parties to these FTAs agreed to adopt “GATS plus” regulatory obligations, which are largely equivalent to the DR Reference Paper or even extend beyond it with more advanced disciplines. More recent examples include Chapter 22 (Good Regulatory Practices and Regulatory Cooperation) of the EU–New Zealand FTA, concluded in 2022, and Chapter 26 (Good Regulatory Practice) of the UK–Australia FTA, concluded in 2021. These new regulatory disciplines are set to stipulate minimum standards that must be observed by the FTA parties when adopting and applying domestic regulations, which, if implemented effectively, might in turn offer promising venues for overcoming regulatory fragmentation and potential overreaching. The assumption is that domestic administrative practices in different jurisdictions would become more compatible and reasonable if countries observe, throughout the regulatory cycle, a set of “minimum common quality standards.”²⁰⁵

Taking CPTPP and USMCA as representative instruments, the Regulatory Coherence Chapter under the CPTPP and the Good Regulatory Practices Chapter under the USMCA set forth specific obligations with respect to GRP, such as promoting information quality, procedural transparency, clear and plain regulatory language, early planning and retrospective review of regulations, central and internal coordination, and engagement with interested persons.²⁰⁶ In particular, parties have committed to “minimize unnecessary regulatory differences and to facilitate trade or investment,”²⁰⁷ without compromising each state’s regulatory autonomy to pursue its public policy objectives. Possible mechanisms for promoting regulatory coherence include exchanging technical information and data and exploring common approaches to tackling the risks posed by the use of emerging technologies.²⁰⁸ In addition, under both the CPTPP and the USMCA, the core of GRP is the need to conduct an impact assessment when developing regulations. More specifically, the second paragraph of Article 28.11 of the USMCA delineates the procedures and considerations under which an RIA should be conducted, which highlights the procedures of the regulatory assessment as follows:

1. (a) the need for a proposed regulation, including a description of the nature and significance of the problem the regulation is intended to address;

2. (b) feasible and appropriate regulatory and non-regulatory alternatives that would address the need identified in subparagraph (a), including the alternative of not regulating;

3. (c) benefits and costs of the selected and other feasible alternatives, including the relevant impacts … as well as risks and distributional effects over time, recognizing that some costs and benefits are difficult to quantify or monetize; and

4. (d) the grounds for concluding that the selected alternative is preferable.²⁰⁹

Amid the Biden administration’s agenda to increase economic cooperation through the IPEF, it is not surprising to see that GRP is one of the key initiatives under the framework. The USTR has declared that it will seek to “advance the benefits of good regulatory practices in supporting good governance”²¹⁰ and “build on the outcome reached in the WTO Joint Initiative on Services Domestic Regulation, as appropriate.”²¹¹ In fact, when the USTR of the Obama administration led the TPP negotiations, the USTR repeatedly stressed the importance of the inclusion of a chapter on regulatory coherence in the TPP, as it reflects the growing relevance of regulatory issues to international trade and investment.²¹² In light of the fact that the legal and regulatory environment in TPP parties is diverse, the USTR had advocated for the incorporation of regulatory coherence into the TPP to eliminate the problem of “overlapping and inconsistent regulatory requirements or regulations being developed unfairly and without a sound basis.”²¹³ Now, six years after the US withdrew from the CPTPP, the IPEF seems to have the potential to reassert US economic engagement to counter China’s growing influence and promote US regulatory approaches and standards in the region.

The crux of the matter, however, is to what extent good governance obligations in international trade agreements can tackle regulatory fragmentation and combat unreasonable administrative practices. Will these regulatory principles and methodological tools bring about greater policy coherence in the governance of the digital economy? When domestic regulators are faced with questions of how to regulate digital platforms, in what way can these GRP converge the fragmented legal approaches to issues such as content moderation, cultural diversity, platform competition, and algorithmic transparency? For the reasons explained below, this book contends that far more political will is needed to ensure that the GRPs/RIAs do the trick.

5.4.3 Marching toward Policy Coherence in Platform Governance?

5.4.3.1 The Breadth and Strength of GRP/RIA

First of all, although relatively ambitious GRP provisions can be found in some FTAs,²¹⁴ other FTAs contain only symbolic GRP provisions. For example, the chapter on the “General Provisions and Exceptions” of the RCEP establishes obligations on minimum standards of regulatory transparency and due process in the administrative proceedings. However, the RCEP is silent on many higher standards of GRPs to prevent undue delay in authorization or arbitrary administration. In particular, the absence of RIA in the RCEP implies fewer safeguards to ensure regulatory quality than if such provisions were included, leaving the RCEP behind in the developments of GRPs compared to other recently negotiated FTAs.²¹⁵ Arguably, divergences in the regulatory frameworks of Asian countries, especially China, have made it challenging to conclude “harder GRP obligations.”

Second, even for those countries that are parties to FTAs with more ambitious GRPs, one fundamental question surrounding the possible contributions of the GRPs to the quality of platform regulation is the breadth of their application. Normatively speaking, the applicable coverage of the GRPs is rather uncertain. Here, a typical example is Article 25.1 of the CPTPP, which stipulates that the “covered regulatory measure” under its Regulatory Coherence Chapter refers to “the regulatory measure determined by each Party.”²¹⁶ According to Article 25.3, each party shall determine and make publicly available the scope of its covered regulatory measures. Article 25.3 further requires that “in determining the scope of covered regulatory measures, each Party should aim to achieve significant coverage.”²¹⁷ In other words, although the definition of “regulatory measure” is rather broad, referring to “any measure of general application … adopted by regulatory agencies with which compliance is mandatory,”²¹⁸ the exact scope of the “covered regulatory measures” is flexible. Each party has the discretion to decide to what extent its domestic regulation should be subject to GRPs. Empirically speaking, some notifications from the CPTPP parties in this regard are encouraging. In particular, developed countries such as Japan and Australia have adopted “covered regulatory measures” that are general and extensive.²¹⁹ However, considering that the application of the GRPs generally requires administrative resources to complete, it is likely that some governments may tend to establish a minimum threshold in terms of the obligation to perform GRPs.²²⁰ After all, GRPs not only require changes in the institutional setting, but also demand adjustments in the behavior and mindset of civil servants.²²¹ Therefore, platform regulations are not necessarily subject to GRPs/RIAs.

Third, even assuming that a country has made legally binding commitments to GRPs in FTAs, the real impact may not be significant due to the narrow range of domestic regulatory instruments that are subject to GRPs. For example, Article 28.1 of the USMCA explicitly limits the scope of “regulation” under the Good Regulatory Practices Chapter to the “measure of general application adopted, issued, or maintained by a regulatory authority with which compliance is mandatory.”²²² The definition of “regulatory authority” under the chapter, however, is limited to “an administrative authority or agency at the Party’s central level of government that develops, proposes or adopts a regulation.”²²³ The definition explicitly excludes “legislatures or courts.”²²⁴ In other words, the GRPs/RIAs only apply to “few” but not “all” regulatory instruments that impact platform activities. In terms of platform regulation, the fact that the GRPs only apply to the executive branch of governments but not to legislatures or parliaments means that the RIAs, in some countries, are at most carried out by agency-made digital rules.

Finally, assuming arguendo that a platform regulation falls within the range of regulatory instruments subject to GRPs, the strength of the application of RIAs might be weakened by divergent methodologies. As previously indicated, RIA has been defined as a systemic approach to critically identifying and assessing the regulatory effects – including the existing regulations and proposed nonregulatory alternatives.²²⁵ In the context of digital platform regulations, policymakers should identify problems arising from platform monopolies and data capitalism, determine the need for intervention, propose alternative regulatory options ranging from competition law to algorithmic transparency, and then assess and decide upon the preferred policy option. Throughout the process, the assessment – whether it is qualitative or quantitative – can be accomplished by employing various techniques. According to the OECD, the adoption of RIA is now becoming widespread, but it operates differently within and across countries.²²⁶ Methodologically speaking, various RIA procedures can be designed and used. As a result, policymakers’ understanding of the procedure of RIA varies greatly. In practice, RIA has been uniquely practiced by different countries or even different governmental bodies.²²⁷

This raises the question of how the good governance obligations under the FTAs help ensure that policymakers “choose the best regulatory option” when designing and implementing platform regulations, and how the GRPs/RIAs help converge the fragmented legal approaches to the platform economy. At their heart, GRPs under the FTAs were intentionally framed in flexible terms to reserve room for an individual FTA party to decide how extensively it will commit. It is clear that the RIA provisions under the FTAs do not provide much guidance as to how to operate the process. To place these questions into a more concrete setting, both the DSA and the DMA “passed” the RIA test conducted by the EC before implementation. The European Commission’s Impact Assessment Report of the DSA (the RIA Report) is over 200 pages long, containing a detailed analysis of what the problem drivers are, why the EU should act, what the policy options are, what the impacts of these options are, how the options compare, and which option is preferred.²²⁸ The RIA Report concludes that the DSA is in full compliance with the EU’s international obligations in the WTO and FTAs.²²⁹ The RIA Report underscores that the DSA is in line with the nondiscrimination provisions of the GATS, as it establishes objective criteria, including the definition of VLOPs, regardless of the origin of the services supplier.²³⁰ Moreover, the RIA Report states that the asymmetric regulation, which imposes additional transparency requirements on the VLOPs, does not jeopardize the protection of trade secrets of digital platforms because the DSA entails a secrecy obligation on the authorities and experts with regard to trade secrets.²³¹

Would the same conclusion of the RIA Report be drawn if a different RIA methodology is introduced? Probably not. As explained by Alemanno, a cost–benefit analysis in many cases can be “extra-territorially blind.”²³² Both the costs of regulatory fragmentation and the benefits of regulatory coherence might be underestimated. Measurements in the RIA procedure can also be problematic when quantification is not possible, and in that case, the RIA might become a “rubber stamp” to endorse administrative decisions without carefully scrutinizing them,²³³ effectively serving as a mere justification for policymakers’ choices. At the end of the day, the cost–benefit analysis is more about local value preferences. When local preferences and policy priorities are involved, the process has a more subjective character.²³⁴ As Gari observes, this world might be increasingly globalized and interconnected, but deep differences among countries remain in terms of regulatory needs.²³⁵

5.4.3.2 Political Support Needed for Future Governance

All in all, in light of the above, the trend of developing GRP obligations under the international trade agreements has the potential to serve as an important tool to address regulatory fragmentation, but much will depend on the political will of the governments involved. Given that the “obligations” of regulatory coherence are predominately phrased in a soft and loose way, the effectiveness of the GRPs, especially the successful incorporation of RIAs into regulatory policy, is dependent upon strong administrative determination. To put it another way, regulatory divergence can also be seen as an outcome of democratic legitimacy.²³⁶ The potential impact of the good governance obligations under the WTO or FTAs in tackling regulatory fragmentation is therefore constrained.²³⁷ Pushing the goals of GRPs/RIAs too far may generate concerns related to an undue degree of intrusion upon a country’s right to regulate.²³⁸ For this reason, the breadth and strength of the GRPs/RIAs fundamentally rest on political support for good-faith implementation. However, despite the above considerations, there are reasons for being optimistic about better governance through international trade agreements.

First, in terms of the applicable coverage of the GRPs, there seems to be a growing political willingness to endorse the relevant initiatives. At the WTO, additional members have joined the DR JSI to “make their regulatory environment more conducive to business,” and to “lower trade costs for services suppliers seeking to access foreign markets.”²³⁹ At the regional level, more and more new-generation FTAs contain the GRP obligations. Although they are primarily drafted in flexible language, they nevertheless provide a baseline for parties to undertake regulatory reform. Over time, such soft obligations may help the parties to align with trends leaning toward more transparent and predictable regulatory frameworks – which are vital to services trade across borders.²⁴⁰ Platform suppliers, especially SMEs, will eventually benefit from improved regulatory quality and facilitation.

Second, in terms of a greater political willingness to implement GRPs, one promising direction is to establish a central oversight body to perform checks of draft RIA reports.²⁴¹ Historically, it has not been unusual for regulators to conduct an “incomplete” assessment of the economic costs and benefits of regulatory alternatives.²⁴² Policymakers may fail to generate a comprehensive evaluation of the possible impacts of regulation, resulting in “stove-piped” internal decisions that lead to conflicting regulatory approaches in the administrative system.²⁴³ However, according to OECD studies, more and more jurisdictions, including both OECD and non-OECD countries, have dedicated a single governmental body to be responsible for reviewing regulatory quality in the national administration from a “whole-of-government” perspective.²⁴⁴ Alternatively, an increasing number of countries are establishing a specific parliamentary committee with responsibilities for monitoring the quality of the RIA system as a whole.²⁴⁵ As previously discussed, in the context of platform regulation, the primary institutional challenge in the introduction of RIAs is to coordinate diverse considerations – for example, competition policy, consumer protection, freedom of speech, trade secrets, industrial innovation, and other public objectives – in an integrated manner. The fact that many countries are now creating a centralized coordinating body within their jurisdictions as a quality assurance mechanism is a sign of governments’ willingness to embed GRPs and RIA scrutiny in their policymaking systems.

Finally, in terms of the methodology of RIAs, some innovative approaches have been introduced into the realm of platform regulation, which represent a significant step in promoting coherence in digital policy. For example, the “Digital Checklists” contained in the EU’s “Better Regulation toolbox” are specifically designed to identify the digital issues and impacts surrounding new policies. Any proposed regulation with digital dimensions should go through the Toolbox when defining the problem, assessing impacts, developing policy options, and choosing digital policy solutions.²⁴⁶ Similar approaches have been adopted in the UK’s RIA guidance, which effectively integrates competition policy with RIAs by requiring the assessment to “incorporate an explicit consideration of the competition impacts of regulatory proposals.”²⁴⁷ More importantly, there is a general tendency to not only embrace domestic market dynamics, but to also address international aspects such as market openness.²⁴⁸ The increasing degree of attention accorded data-driven factors in the RIA context indicates a promising venue to advance the more coherent governance of platform activities.²⁴⁹

5.5 Conclusion

We have investigated the driving forces of platformization. When data becomes capital and the algorithmic input of the digital platforms, the interplay between national regulation and international economic law is complex and multilayered. The specific aspects considered in Chapters 3–5 help formulate views and outlooks regarding the role of international economic law in the politics of datafication. The discussions in Part II also lead us to reflect on what international trade agreements have done and can do in global platform governance. Let us now shift gears to the broad digital ecosystem and continue to explore the phenomenon of datafication from a cross-layer perspective: data flows. Chapter 6 will bring a sharper focus to privacy and cybersecurity.