Changing the incentive structure of social media may reduce online proxy failure and proliferation of negativity

Abstract

Social media takes advantage of people's predisposition to attend to threatening stimuli by promoting content in algorithms that capture attention. However, this content is often not what people expressly state they would like to see. We propose that social media companies should weigh users’ expressed preferences more heavily in algorithms. We propose modest changes to user interfaces that could reduce the abundance of threatening content in the online environment.

Through millennia of evolution, the brain has developed an attentional system that preferentially orients people toward physical, emotional, and social threats (Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001; Öhman & Mineka, 2001; Rozin & Royzman, 2001). The proxy for threat is, then, negative or threatening stimuli for both the individual and their group. Indeed, people attend to physical threats such as snakes or heights (Dijksterhuis & Aarts, 2003; Öhman & Mineka, 2001), social threats such as anger or out-group animosity (Fox et al., 2000; Rathje, Van Bavel, & Van Der Linden, 2021), and moral threats such as outrage or disgust (Brady, Crockett, & Van Bavel, 2020; Brady, Gantman, & Van Bavel, 2020; Hutcherson & Gross, 2011), in part because they alert people to potential harm. Our attentional system acts as a regulator, and its goal is to attend to stimuli that are relevant for people's wellbeing. Proxy failure occurs when people attend to false threats, even though no harm is imminent (John et al.).

Thus, attention is biased by threat proxies. It is therefore problematic that social media in particular uses attention as a proxy for what people want to see, and incorporate that into their algorithms. By using attention as a proxy for interest, social media companies motivate users to try to “hack” innate threat detection systems that drive people's attention (Brady, McLoughlin, Doan, & Crockett, 2021; Crockett, 2017). In this way, negative, divisive, and threatening content may be preferentially spread by algorithms because of lower-level proxy failure.

Critically, people do not necessarily want negative and divisive content in their social media feeds. Although people acknowledge that negative, false, and hateful content goes viral online, they do not want such content to go viral (Rathje, Robertson, Brady, & Van Bavel, in press). Rather, they want accurate, positive, and educational content to go viral. Proxy failure may be partially responsible for this outcome – focusing too much on social media engagement (likes, shares, etc.) as a proxy for people's preferences leads to proxy failure, because engagement metrics often promote content that people say they do not like, such as false, negative, or hostile content (Brady, Wills, Jost, Tucker, & Van Bavel, 2017; Rathje et al., 2021; Robertson et al., 2023).

Furthermore, changing incentive structures on social media toward more positive and constructive content may have downstream positive consequences. When the incentive structure of social media changes, people adjust the type of content they share. For instance, when social media platforms reward the veracity or trustworthiness of a post, people are far more likely to share true information (Globig, Holtz, & Sharot, 2023; Pretus et al., 2023). Furthermore, this can be achieved with a relatively small tweak to the design features of a social media site. Adding “trust,” “distrust,” or “misleading” buttons to the standard “like” and “dislike” reactions led people to become more discerning of true and false information, and more likely to share accurate information.

When attention or watch time is used as the basis for newsfeed algorithms, the algorithms may become more likely to show harmful or negative content because of peoples’ attentional bias toward threats and engagement with divisive content (Milli, Carroll, Pandey, Wang, & Dragan, 2023). However, if people were given a simple way to curate their news feeds, they would be able to make their news feeds align more with their preferences. Thus, social media companies should use people's expressed preferences for what they want to see on social media (e.g., positive, nuanced, educational, or entertaining content) as a proxy for interest, rather than ambiguous metrics such as attention or engagement. This could be achieved with modest changes to existing social media platforms (Globig et al., 2023; Pretus et al., 2023).

Adding an accessible mechanism that lets users say “I don't want to see content like this” could substantially reduce people's exposure to unwanted threatening stimuli that an attentional proxy might promote. Most social media companies even already have this functionality, but it is often hidden in menus, cumbersome to activate, or the default is set to attention-based proxies. Positive results might be achieved by simply moving such a feature to the default or a more visible location in the platform design.

Overall, this type of intervention will only work if social media companies’ superordinate goals include and prioritize positive user experience and societal outcomes over monetary gain. Negativity and toxicity increase grab people's attention, moral outrage, and spill into offline behaviors such as hate speech and endorsement of antidemocratic action (Brady et al., 2021; Kim, Guess, Nyhan, & Reifler, 2021; Suhay, Bello-Pardo, & Maurer, 2018). Reducing such content may lead to greater user wellbeing and a reduction in negativity, but it also reduces people's use of social media (Beknazar-Yuzbashev, Jiménez Durán, McCrosky, & Stalinski, 2022). This may explain why whistleblowers have revealed that Meta considered implementing algorithms that help downregulate content that is “bad for the world,” but decided against it because it reduced time spent on the platforms by the users (Roose, Isaac, & Frenkel, 2020). If the superordinate goal is simply making money, then social media companies may continue to “hack” people's attentional systems and promote content that draws users’ attention through sensationalism and threat.