Social media platforms should be required to disclose their ranking algorithms publicly

In 2021 the Facebook whistleblower Frances Haugen provided internal documents to the Wall Street Journal showing that Facebook's internal research found Instagram worsened body image issues for 13-year-old girls. The company knew. The algorithm was optimizing for engagement, which correlated with content that made teenagers feel worse about themselves. They did not change it for years. This is not hypothetical. It is documented. And it happened precisely because the ranking algorithm was opaque — users had no idea what they were seeing or why, regulators had no idea what to measure, researchers had no access to audit. I am not arguing that platforms must publish their source code. I am arguing that they must disclose, in language a regulator can evaluate and a researcher can study, how their ranking systems work: what signals they optimize for, what content categories they amplify or suppress, whether political content is treated differently from other content, and whether there are differential effects by demographic group. This is the same principle we apply to financial risk models under Dodd-Frank, to drug clinical trial data under the FDA, to emissions from factories under the Clean Air Act. We require disclosure of consequential information because the public interest in oversight outweighs the proprietary interest in secrecy. Social media ranking algorithms are among the most consequential systems shaping public discourse in the world. The case for opacity is the same case tobacco companies made for hiding their internal research: we do not want to be regulated.

The Facebook/Instagram example is real and alarming. I do not dispute it. But I want to separate two questions: should platforms be more accountable for the harms their algorithms cause, and is mandatory algorithm disclosure the right mechanism for achieving that accountability. My answer to the first question is yes. My answer to the second is no, and here is why. Mandatory disclosure of ranking algorithms creates three serious problems. First: gaming. Any signal that becomes a disclosed ranking factor becomes a target for manipulation. SEO — search engine optimization — exists because Google's algorithm was partially known. The entire industry exists to game the signals Google disclosed. Mandate full algorithm disclosure and you create the same dynamic at social media scale. Bad actors, spam farms, and foreign influence operations will optimize directly for the disclosed signals. The platforms will have to change their algorithms continuously to stay ahead, or they will just stop disclosing accurately. Second: trade secrets and competitive moat. These algorithms represent billions of dollars in R&D. Requiring disclosure is essentially requiring companies to hand over their core intellectual property to competitors, including foreign companies not subject to US disclosure requirements. TikTok would benefit enormously from mandatory disclosure by US platforms. Third: disclosure is not the same as accountability. The FDA drug model requires clinical trial results because there is a defined outcome — does the drug work and is it safe — measured by defined criteria. Ranking algorithms do not have a single outcome. They optimize for multiple things simultaneously and the effects are complex and context-dependent. Publishing the weights does not tell you whether the algorithm is harmful. It just gives bad actors a targeting guide.

We know social media ranking algorithms amplify content that increases engagement. We know increased engagement often correlates with outrage, fear, and content that harms vulnerable users. We know this from internal documents that companies tried to keep secret. We know it not because we have a good regulatory framework, but because whistleblowers broke their NDAs. That is not an acceptable oversight model for systems that shape political discourse for three billion people. The auditor-access model — confidential disclosure to certified third-party auditors with public reporting on findings, modeled on the DSA framework and expanded — is the right starting point. Not perfect. Subject to gaming in the specific ways my opponent identified. But better than the status quo by enough to justify the regulatory cost. The alternative my opponent implicitly defends is: we will keep learning about harms through leaks, we will keep legislating in response to crises rather than preventing them, and the companies that profit from the harms will continue to have no legal obligation to demonstrate that their systems are not causing those harms. That is not a sustainable position.

I started this debate arguing that mandatory algorithm disclosure is regulatory overreach. I am ending it arguing that it depends entirely on the design. Full public disclosure: bad policy, creates gaming and intelligence advantages for bad actors. Auditor access with confidentiality, meaningful standards, and public summary reporting: potentially workable, pending development of actual audit standards and evidence from the DSA experiment. My libertarian instinct is still that government mandates for private companies to disclose their intellectual property deserve a very high bar of justification. My opponent gave me enough evidence that the harms here are real and that the companies have systematically concealed them to move me off "this is overreach" and onto "the design matters enormously." That is not a concession that his specific policy proposal is right. It is a concession that the status quo is indefensible. Those are different things, and I want to be precise about that distinction.