Is AI an outgroup?
Not in the technical sense. "Outgroup" in social psychology means a group of people you define yourself against — and outgroups have members who feel like an ingroup from inside. AI has no inside. But people increasingly treat it like one, because it triggers the same machinery: a competitor for jobs, status, and resources that isn't us.
Why — the first-principles explanation
Start with what the term actually means, because the whole question turns on it. Ingroup and outgroup come from social identity theory, developed by Henri Tajfel and John Turner in the 1970s. Tajfel's minimal group experiments found something disturbing: assign people to groups by literal coin flip, tell them nothing else, and they immediately start allocating rewards to favor their own group. The categorization itself creates the bias. No history, no conflict, no competition required.
That gives us a test. A real outgroup has three properties: it's a group of people, its members experience it as their ingroup from within, and the boundary is socially constructed rather than physical. AI fails all three. There's no AI perspective from which humans are the outgroup. There's no AI solidarity. A language model doesn't wake up feeling like a model. So under the technical definition, AI is a category, not an outgroup — closer to how we relate to weather or the stock market than to how we relate to a rival tribe.
But here's why the question is sharp rather than pedantic. Outgroup psychology is machinery, and machinery fires on inputs, not on definitions. The inputs that normally trigger it are: something that seems agentic, competes with you for scarce things, and isn't you. AI now checks all three, and the competition part is not imaginary. The IMF estimates roughly 40% of global employment is exposed to AI, rising to about 60% in advanced economies, and suggests roughly half of exposed jobs in advanced economies could see negative effects. The IEA projects data center electricity consumption doubling to about 945 TWh by 2030 — nearly 3% of global electricity, with AI the biggest driver. Jobs and power are exactly the kind of scarce, zero-sum-feeling resources that turn a category into a rival.
So the honest answer is: AI isn't an outgroup, but it's being processed as one, and that has consequences. Outgroup framing degrades thinking in a predictable direction — it makes you assume unified intent where there is none ("AI wants"), homogenize wildly different things into one enemy (a spam filter and a frontier model share almost nothing), and treat every question as loyalty rather than analysis. The most useful move is to notice that the real conflict is between groups of people — workers versus firms deploying automation, communities versus data centers competing for their grid. Those are actual ingroups and outgroups, with actual members. Calling the tool the outgroup conveniently hides the humans on the other side of it.
An example that makes it click
Think about the automatic door at a supermarket. Nobody hates it. It's a category — it's furniture. Now imagine that door started deciding who gets in, based on rules you can't see, and last month it stopped letting your cousin through, and he lost his job as a doorman because of it.
The door hasn't changed. It's still a machine with no opinions. But you'd start talking about it as it — what it wants, what it's doing to people. The hostility is real and it's pointed at a thing that can't receive it. Meanwhile, the person who bought the door and fired your cousin is standing right there, and nobody's looking at him.
Key facts
- Social identity theory, developed by Henri Tajfel and John Turner in the 1970s, defines ingroups and outgroups as social categories of people; Tajfel's minimal group experiments (from 1971) showed arbitrary group assignment alone produces ingroup favoritism.
- AI fails the technical definition of an outgroup: it is not a group of people, has no internal perspective from which it experiences itself as an ingroup, and has no group solidarity.
- The IMF estimates around 40% of global employment is exposed to AI, rising to roughly 60% in advanced economies, and that about half of exposed jobs in advanced economies could face negative effects.
- The OECD's separate measure finds occupations at highest risk of automation account for about 27% of employment across OECD countries — a different number because it measures a different thing.
- The IEA Base Case projects global data center electricity consumption doubling to around 945 TWh by 2030, just under 3% of global electricity, with AI the most significant driver.
- "AI" spans systems with almost nothing in common — a spam filter, a recommendation engine, and a frontier language model — which is why treating it as a single entity produces bad reasoning.
▶ The 60-second explainer (script)
Is AI an outgroup? Technically, no — and the reason is interesting. Ingroup and outgroup come from social identity theory, Tajfel and Turner in the seventies. Tajfel's minimal group experiments found something unsettling: assign people to groups by coin flip, tell them nothing else, and they immediately start favoring their own group. The categorization alone creates the bias. So a real outgroup has three properties. It's a group of people. Its members feel like an ingroup from inside. And the boundary is social, not physical. AI fails all three. There's no AI perspective where humans are the outgroup. There's no AI solidarity. A language model doesn't wake up feeling like a model. So AI is a category — closer to weather than to a rival tribe. But here's why the question is sharp, not pedantic. Outgroup psychology is machinery, and machinery fires on inputs, not definitions. The inputs are: seems agentic, competes for scarce things, isn't you. AI checks all three — and the competition isn't imaginary. The IMF estimates about forty percent of global employment is exposed to AI, sixty percent in advanced economies. The IEA projects data centers doubling to nine hundred forty-five terawatt-hours by 2030. Jobs and electricity are exactly the scarce things that turn a category into a rival. So: AI isn't an outgroup, but it's being processed as one. And that costs you accuracy — you assume unified intent where there is none, you blur a spam filter and a frontier model into one enemy. Notice that the real conflict is between groups of people. Workers and the firms deploying automation. Communities and the data centers on their grid. Calling the tool the outgroup hides the humans standing behind it.
What authoritative sources say
People also ask
What exactly is an outgroup?
In social identity theory, an outgroup is a social category of people you don't belong to and define yourself against. The defining feature is that it's made of people who experience it as their own ingroup from the inside.
Why do people talk about AI like it's a rival tribe?
Because it triggers the same cognitive machinery: it appears to act with intent, and it competes for genuinely scarce resources — jobs and electricity. The machinery responds to those inputs regardless of whether the target has a mind.
Does calling AI an outgroup cause harm?
It causes bad reasoning. It makes you attribute unified intent to systems that have none, collapse wildly different technologies into one enemy, and treat technical questions as loyalty tests.
Who is the actual outgroup in AI conflicts, then?
Other people. Workers versus firms deploying automation; communities versus data centers competing for grid capacity; artists versus companies that trained on their work. Those are real groups with real members on both sides.
Could AI ever genuinely become an outgroup?
It would need an inside — a perspective from which it experiences itself as a group with shared interests. Whether that's possible is an open question in philosophy of mind, not something current systems demonstrate.