Introduction
Following a successful track at the Data for Policy Conference 2024, we are issuing an open call for contributions to a special collection (virtual special issue) on "Generative AI for Sound Decision-Making: Challenges and Opportunities” to published in Data & Policy, an open-access journal at Cambridge University Press.
This collection will investigate the impacts of LLMs and other Generative AI Models (GAIs) on our daily life and how regulators can respond to ensure that GAIs make sound recommendations/decisions, for the benefits of humanity and societies.
ChatGPT, the popular chatbot of OpenAI, has been accessed by more than 100 million users monthly the first two months after launching, making itself the fastest-growing consumer application. Its phenomenal success attracts substantial business and public interests. LLMs continues to break boundaries and be vastly applied in different fields, e.g. education, healthcare, law/public policy, etc., tremendously benefitting citizens of different communities and facilitates public decision-makings. However, results/decisions generated by LLMs/GAIs have demonstrated to be equally confusing and biased when the training datasets are distorted or unevenly distributed across different sub-populations. A deeper understanding of the potentials and limitations of LLMs/GAIs is critical for the future advancement of LLMs/GAIs, and the future sustainability of humanity and societies.
This special collection of papers aims to explore the technical and associated moral and ethical challenges posed by LLMs/GAIs, and what moral and ethical guidelines should be in place for addressing these challenges. We will explore the trustworthiness and explainability of existing LLMs/GAIs, and when LLMs/GAIs can be considered trustworthy and explainable. Are there any specific moral and ethical guidelines available to guide the future development of LLMs/GAIs?
Policy significance of this collection
Given that the application of LLMs/GAIs can transform the traditional practices across many fields, it is important for decision makers to properly anticipate the upcoming changes and provide regulatory frameworks to guide the research and development of GAIs for sound decision-makings.
Key themes
The key themes below cover the technical and the associated ethical and moral challenges of LLMs/GAIs, and the opportunities that these models presented for socially beneficial, moral and ethical decision-makings, covering, but not limited to, the following topics:
- Challenges of LLMs/GAIs in decision-makings, such as education, health care, regulatory decision-makings
- Regulatory frameworks and guiding principles of LLMs/GAIs for sound decision-makings
- Methods for determining if LLMs/GAIs are making socially beneficial, moral and ethical decisions
- Methods for evaluating trustworthy and explainable LLMs/GAIs
- Methods to reduce data/model biases and improve fairness
- Methods to reduce disinformation/misinformation and improve trustworthiness
- Methods to improve data privacy and security
- Case studies covering socially beneficial, ethically and morally sound LLMs/GAIs development
Timetable
Authors should submit to Data & Policy by the final deadline of November 29, 2024.
Articles will be published as soon as possible after acceptance, in the interest of allowing authors to disseminate their work without unnecessary delay and added to a curated page for the collection of articles. An editorial reflecting on their insights will be published later in 2024.
Submission process
Authors should submit articles through the Data & Policy ScholarOne site. Use the 'Generative AI for Sound Decision-makings' dropdown response to the Special Collection question.
Before submission, authors should familiarise themselves with the Instructions for Authors (which include LaTeX, Overleaf and Word templates for author convenience). Note that Data & Policy publishes the following types of articles, which authors will be prompted to select from on submission:
- Research articles that use rigorous methods that demonstrate how data science can inform or impact policy by, for example, improving situation analysis, predictions, public service design, and/or the legitimacy and/or effectiveness of policy making. Published research articles are typically reviewed by three peer reviewers: two assessing the academic or methodological rigour of the paper; and one providing an interdisciplinary or policy-specific perspective.
- Commentaries are shorter articles that discuss and/or problematize an issue relevant to the Data & Policy scope. Commentaries are typically reviewed by two peer reviewers.
- Translational papers are contributions that show how data science principles, techniques and technologies are being used in practice in organisational settings to improve policy outcomes. They may present original findings but are less embedded in the scholarly literature as research articles. They are typically reviewed by two peer reviewers, who assess the rigour and policy significance of the paper.
- Data papers that provide a structured description of an openly available dataset with the aim of encouraging its re-use for further research
Note that the journal encourages authors to make replication data and code available in an open repository, where this is possible (see the research transparency policy). All authors must provide a Data Availability Statement in their article that explains where the replication material resides, if it is available, and if not, the reason why it cannot be made accessible.
Editors
- Jacqueline CK Lam (The University of Hong Kong)
- Victor OK Li (The University of Hong Kong)
- Lawrence Cheung (Chinese University of Hong Kong
- Jon Crowcroft (University of Cambridge & Alan Turing Institute) - Data & Policy Editor-in-Chief