Regulating Social Media as a Public Good: Limiting Epistemic Segregation

ABSTRACT

The rise of social media has correlated with an increase in political polarization, which many perceive as a threat to public discourse and democratic governance. This paper presents a framework, drawing on social epistemology and the economic theory of public goods, to explain how social media can contribute to polarization, making us collectively poorer, even while it provides a preferable media experience for individual consumers. Collective knowledge and consensus is best served by having richly connected networks that are epistemically integrated: individuals with diverse levels of expertise should be relatively well connected to each other. In epistemically segregated networks, by contrast, we have reason to predict collective epistemic failures. Expert knowledge will be isolated from the majority, leading average opinion to be less informed than is socially optimal, and entrenching disagreements. Because social media enables users to very easily adopt homophilous network connections – connections to those with similar opinions, education levels, and social backgrounds – it is likely to have increased epistemic segregation compared to older media platforms. The paper explains the theoretical foundations of these predictions, and sketches regulatory measures – such as taxes – that might be employed to preserve the public good of a well integrated social media network.

KEYWORDS:

• Social media
• public goods
• polarization
• homophily

Acknowledgments

Thanks to Kevin Zollman, Robert Simpson, Birendra Rai, Daniel Akselrad and Colin Klein for helpful comments on earlier drafts. The paper also benefited from comments at the Trust in Digital Spaces Workshop, held at ANU in 2022.

Disclosure Statement

No potential conflict of interest was reported by the author.

Notes

1. Some recent examples of an economic approach in social epistemology include Alfano and Sullivan (2021); Zollman (2020).

2. Here I focus on user-user interactions on social media, which are particularly important on platforms such as Twitter and Facebook. On platforms like YouTube, user-content-user interactions are more typical, in which users interact in comments around a piece of content produced by a third party (Wattenhofer, Wattenhofer, and Zhu 2012).

3. In networks like Twitter or YouTube, the decision to follow can (usually) be made unilaterally, by the follower/subscriber alone; in networks like Facebook it requires bilateral agreement to form a link. But in no cases I am aware of is a third party involved. That said, platforms such as Twitter and Instagram make efforts to encourage users to adopt new links, so third party influence does exist.

4. Note the Condorcet jury theorem itself does not rely on informed rational voters – it has requirements that are in some sense weaker, but in other ways much less realistic.

5. A thought experiment to attempt to isolate the epistemic value of social media is to ask how much we would reduce our use of social media if we experienced amnesia immediately after every session. This is premised on the idea that non-epistemic benefits of social media use, such as affiliation and entertainment, can be enjoyed without lasting memories of the session, and conversely that epistemic benefits cannot be enjoyed without any memories. On these assumptions, amnesia removes epistemic benefits, while other benefits remain intact. I speculate that while for many users, consumption will be relatively unchanged, it is plausible that there would be an overall decline in the order of 1–10%. If correct, the opportunity cost of this time is quite substantial, and sets an impressive lower bound on the epistemic value of social media.

6. Coase (1960). An important caveat is that the theoretically better alternative arrangements might have transaction costs – understood as the costs of moving to a new institutional arrangement and thereafter maintaining it – which make it infeasible to achieve the efficiency gain.

7. See Gillespie (2018) for a fascinating sociological approach to the labor of content blocking.

8. Smith, McPherson, and Smith-Lovin (2014); though Eika, Mogstad, and Zafar (2019) suggest that in the US, homophily with respect to educational attainment has been relatively constant since 1980, having risen between 1960 and 1980.

9. See, e.g. Mijs and Roe (2021), 4. Notably, political polarization in the US is now highly correlated with education, and education may be more significant as an explanatory factor of political disagreement than income or wealth (Cowen 2017; Boxell 2020).

10. It is also worth observing that popular media consumption in the US was already, prior to the invention of online social networks, highly segregated by race (Anderson and Waldfogel 2015). In 1998 for instance, the most popular TV show among whites – Seinfeld – was the 51st ranked in popularity among blacks. And Between Brothers, the most popular show among black viewers, was 112th in popularity among whites. Similar racial segregation is also observed in radio and newspaper consumption.

11. Though this is a natural interpretation, it is certainly not essential. The nodes could for instance be individual posts, tweets, or other units of content.

12. StatCounter (2022). Similarly, Facebook and Google combined account for over 75% of political advertising revenue in the US.

13. Major news organisations (and sites like Wikipedia) would potentially benefit from such a subsidy, and this proposal might simultaneously address concerns about economic pressures on major news organisations. The Australian government recently passed legislation requiring social media companies to negotiate to pay media companies for the sharing of their content via social media (Treasury Laws Amendment [News Media and Digital Platforms Mandatory Bargaining Code] Bill 2021). See Grueskin (2022) for comment on the possibility of such legislation being adopted in other jurisdictions. To the extent that ‘old’ media companies produce content which is widely shared across epistemically diverse groups, this legislation has a similar effect as the present proposal: it subsidizes epistemic desegregation.

14. There is some evidence that social media is being overconsumed anyway, relative to what our informed selves would recommend (Allcott et al. 2020) – so measures that make social media less attractive might already be justified on overall welfare grounds.

Additional information

Funding

This work was supported by the Australian Research Council under Grant DP19010041.