Meta’s Data Problem
Why the Company That Knows the Most About Human Behavior Understands It the Least
There is a version of this story that sounds visionary. Meta, one of the most data-rich companies in human history, sits atop behavioral signals from three billion users. Every click, every scroll, every lingering pause — a real-time map of what human beings actually do, at scale, continuously.
Google may have the best data on what we want. Meta, by most measures, has the most behaviorally rich data on what we do.
And yet, recent Wired articles documented more than once this week, the company that arguably knows more about human behavior than any institution in history has struggled, persistently and visibly, with its own internal culture and executive leadership. The irony is almost too neat to be believed. Almost.
The explanation is not complicated, but it is counterintuitive: Meta’s data is not a map of human psychology. It is the world’s most comprehensive record of human psychology under stress. And you cannot build a healthy culture out of that.
The companies that understand this better are, not coincidentally, winning the race to build artificial intelligence. The ones that don't are falling behind — and struggling to understand why.
The Best Dataset on What We Do When We Are Unwell
Begin with what Meta’s data actually measures. The platform’s business model depends on engagement, and engagement is reliably driven by a narrow band of emotional states — boredom, loneliness, anxiety, envy, outrage, impulsivity. Not contentment. Not reflection. Not connection in any durable sense. The algorithm rewards dysregulation because dysregulation keeps people scrolling.
This means Meta’s behavioral data is not a neutral portrait of human nature. It is, more precisely, the most detailed dataset in existence on what human beings do when they are not okay.
That creates a problem with no clean exit. If Meta tried to reverse-engineer its engagement data to build a model of psychological health — essentially inverting the signal to find what to cultivate instead of what to exploit — it would face three simultaneous obstacles. The first is technical: inverting a contaminated signal doesn’t produce a clean one, it produces a different contaminated one. The second is psychometric: the inverse of a dysfunction is not automatically its corresponding virtue. In personality science, the most important insight about someone who scores very high on a scale is not simply the mirror of the insight about someone who scores very low — they are different failure modes, not opposite ends of a single lesson. Knowing what breaks people does not tell you, in any reliable way, what makes them whole. The third obstacle is institutional: openly acknowledging what the data represents would be a tacit admission to employees, and a gift to plaintiff’s attorneys, that the company’s core business runs on human distress.
The trap is structural. Meta cannot use its data honestly without implicating itself.
The Same Behavior, Opposite Meanings
Even setting aside what the data measures, there is a more fundamental problem: behavioral data cannot tell you why people do what they do.
In personality science, practitioners work with a clarifying analogy. Two managers might both refuse to delegate. One refuses because she is a perfectionist — high in conscientiousness, driven by pride in craft. The other refuses because he is disorganized — low in conscientiousness, unable to prioritize or plan. The surface behavior is identical. The underlying psychology is opposite. Any intervention built on the behavior alone will misfire for at least one of them.
Meta’s data has this problem at planetary scale. A user scrolling at 1 AM could be depressed, avoidant, relaxed, procrastinating, lonely, or dissociating. The platform sees the same scroll. Culture — real culture, the kind that shapes how people treat each other and make decisions under pressure — requires understanding motivation, not just movement.
The Algorithm Has Already Corrupted the Signal
There is a second structural failure, deeper than the first. The behaviors Meta observes have already been shaped by Meta’s own algorithm. The data is not a clean readout of human psychology — it is a readout of human psychology under continuous algorithmic influence: nudges, recommendation loops, emotional amplification, content ranking.
Trying to extract cultural insight from that data is like trying to study human nutrition using a dataset drawn entirely from casino buffet patrons. The environment has distorted the signal beyond recovery. Any culture built from those inputs would reflect not human values, but whatever the algorithm found most profitable to amplify.
Performance Is Not Character
On-platform behavior is also socially performative by design. Users on Facebook, Instagram, and Threads are curating, filtering, and identity-managing. They are responding to the social feedback loops of likes, comments, and comparisons. They are not behaving naturally — they are behaving for an audience.
Building culture requires understanding how people actually behave privately, under pressure, when the stakes are real and no one is watching. Meta’s data captures the opposite. A company that built its internal values around performative behavior would end up rewarding presentation over substance, appearance over integrity — which is, critics have long argued, precisely the culture Meta has produced.
States Are Not Traits
Culture requires trait-level insight: stable, dispositional characteristics that predict how someone will behave across time and context. Meta’s data is saturated with states — temporary emotional conditions triggered by platform interaction. The distinction is not semantic. A person in a moment of anxiety is not the same as an anxious person. A scroll session driven by loneliness does not tell you whether someone is constitutionally introverted or situationally isolated.
Any cultural framework derived from Meta’s data would be calibrated to dysregulation, not character. It would mistake the weather for the climate.
What Actually Works
None of this means Meta is incapable of building a healthy culture. It means the data it holds cannot do the work. What organizational science has established over decades requires different instruments entirely: validated assessments that measure stable personality traits, contextual conversations that preserve the meaning of behavior rather than stripping it, developmental frameworks built around regulation and character rather than engagement and reaction.
Meta has access to all of these. It has chosen, repeatedly and at scale, not to prioritize them.
That choice may be partly self-reinforcing delusion — the gap between the company’s public messaging about community and connection and what its internal data actually reflects is wide enough that closing it would require a reckoning most organizations prefer to avoid. Admitting what you know about your own product is not a comfortable move when lawyers are watching.
But it is also, at some level, a cultural choice. The data Meta trusts most is a portrait of human beings at their most reactive, most impulsive, most unwell. The culture it has built is, in that sense, a faithful reflection of what it chose to look at.
There is a final observation worth making. The companies leading the AI race — Anthropic, OpenAI, Google DeepMind, and Microsoft — have their cultural and execution problems, as all large organizations do. But none of them have built their core business around behavioral dysregulation. Their products are cognitive tools, not engagement traps. The data they generate reflects what people are trying to do, not what emotional state makes them easiest to monetize.
Meta and xAI sit on the other side of that line. Both are led by executives whose most visible platform — Facebook, Instagram, and X — runs on the same dysregulation engine described above. It is probably not a coincidence that both organizations show the most visible signs of cultural incoherence and strategic drift in the AI race. When the data you trust most is a portrait of human beings at their worst, it shapes not just your product decisions but your organizational instincts.
The companies building the future of intelligence, it turns out, may have an advantage that has nothing to do with compute or talent. They simply never learned to mistake human distress for a signal worth following.
Jayson Blair is the host of the Silver Linings Handbook podcast.