More experiments on social media users could provide us insights into human behavior
A Bright Side to Facebook's Experiments on Its Users — Facebook's disclosure last week that it had tinkered with about 700,000 users' news feeds as part of a psychology experiment conducted …
Context & Ripple Effects
Facebook's disclosure last week that its data science team manipulated the news feeds of about 700,000 users to test emotional contagion landed after two years of the company building its reputation as a behavioral research shop — the "What Facebook Knows" reporting in Technology Review back in 2012 framed that ambition openly, and the search tool work covered by the Times in early 2013 showed the same infrastructure turned toward product.
The immediate trigger for this op-ed was the public backlash: the study, published in an academic journal earlier in June with data scientist Adam Kramer among its authors, drew criticism for running on users who never explicitly consented, and Forbes reported days ago that Facebook had quietly added a "Research" clause to its data-use agreement four months after the experiment ran. Against that backdrop, the Times makes the counter-case: that large-scale feed experiments, run responsibly, are one of the few ways to get causal insight into human behavior at scale.
First-order effects
- Facebook and Kramer are absorbing the direct fallout — confirmed backlash over non-consensual emotional manipulation of nearly 700,000 users, with the company publicly dismissing the fuss rather than apologizing.
- Users whose feeds were altered have no recourse under the agreement as it stood at experiment time; the retroactive 'Research' wording only covers future studies.
Second-order effects
- Academic co-authors and journals now face pressure to apply human-subjects review standards to platform data, since the reputational cost lands partly outside Facebook.
- Every major consumer platform running A/B tests inherits the scrutiny — competitors must either defend their own experimentation practices publicly or move first on disclosure.
Third-order effects
- If the pattern holds, platform experimentation drifts from an informal data-science practice toward institutionalized research-ethics governance — consent language baked into terms of service, review boards for studies touching user mood, and a de facto norm that unconsented behavioral experiments are a liability rather than a perk of scale.
The trend: Consumer-internet companies that treat their user base as an experimental subject pool are being pushed from ad-hoc data science toward formal research-ethics rules, with each backlash episode raising the bar for what platforms can run silently.