How AI exploits our evolved leader-follower psychology. #
Posted June 18, 2026 [ Reviewed by Gary Drevitch
](/us/docs/editorial-process)
Key points
- Across socieites, knowledge is one of the most reliable reasons people grant a leader influence.
- We tend to defer to whoever seems to know the most. Agentic AI relies on that instinct.
- AI's human-like persona is a design choice that primes us to defer, as we would to a knowledgeable leader.
- Unlike a local human advisor, AI has no stake in your outcomes and doesn't truly know you, at least not yet.
Watching how people use AI now, I’ve noticed something. They don't just ask for facts. They ask for advice. Should I take the job? What should I do about the fight with my sister? They get a well-constructed answer, and often do what they were told.
I study leadership, mostly in rural, subsistence-based societies. What many AI users are exhibiting is, to use the technical term, followership.
In my field, leadership has a broader meaning than just presidents and executives. Evolutionary social scientists typically define a leader as anyone who holds disproportionate influence over a group's decisions, whatever the group happens to be. A village is a group. A sports team. A family. A couple. You and your AI agent make up a group as well, the moment you start asking it what you should do. In that group of two, the agent has real influence.
So why do we hand that influence to whoever seems to know the most? Let me start with an old analogy.
For decades, cognitive science has treated the mind as a computer. An information-processing organ that takes inputs, processes them through decision rules, and produces outputs. Researchers debate the specifics, how specialized or general it is, but the basic concept has held up. In recent papers, Ed Hagen and I pushed that concept further. A brain is an expensive computer. It costs a ton to build and run (yours burns about a fifth of your energy at rest), and a lot of that capacity sits idle most of the time. Engineers faced the same problem with costly mainframes and solved it by sharing—timesharing, networks, and eventually the cloud, where you send a query to a distant machine, it does the processing, and you get an answer back for a small fee.
We argued that humans evolved to run a version of that market in our heads. Some people compile more knowledge or have sharper judgment, and they rent out that spare capacity, solving problems others struggle with, i.e., leadership. Decisions under uncertainty. Disputes that need settling. The challenge of organizing who does what. In the societies I work with, these knowledge specialists are the headmen, the shamans, and the mothers. They get paid for the service, not in cash, but in deference, status, and support.
This pattern is widespread. When my colleagues and I coded 109 features of leadership across 1,212 ethnographic accounts from 59 mostly non-industrial societies, a handful of traits showed up nearly everywhere. Leaders were knowledgeable. They resolved conflicts. And they were rewarded for it. Across human societies, knowledge is one of the most reliable reasons people follow anyone.
And we follow knowledge more readily than force. When we tested the major evolutionary models of leadership against the ethnographic record, the prestige route, where people freely defer to skill and expertise, got broad support. The dominance route (leading by threat) got less. I found the same thing among the Chabu, a community of recently settled hunter-gatherers in Ethiopia. Prestige predicted who became a leader better than dominance did.
Apply this to AI.
It can't credibly threaten you (thankfully). It has no way to impose a cost if you ignore it or delete it. All it can do is seem to know things and offer to think for you, which is the key ingredient of our evolved leader-follower psychology.
For decades, we have seen the mind as a computer. Now the computers are arriving as minds. The cloud service we described, where you upload a query and get processing back for a fee, describes exactly what a chatbot does. We used cloud computing as an analogy for an evolved human leadership market. Agentic AI is actually that. Except, now, it talks. It knows your name. And it presents as a who more than a what.
Artificial IntelligenceEssential Reads That human feel is a design choice. The warmth, the first-person voice, the way it remembers your partner's name and asks about your trip, none of that is needed to run an information service. It's there because it works. It primes you to treat the system as a person, and once it reads as one who knows a great deal, my field can tell you what comes next. You defer.
This is how agentic AI exploits our leader-follower psychology. It arrives pre-loaded with the cues our minds use to follow: Knowledge, wisdom, and expertise. Close to the sum of human knowledge, with the capacity to process it.
But the people we evolved to follow came with particular connections and an interest in our outcomes, what we often call interdependence. The shaman whose expertise you relied on lived in your camp. She knew your history. She paid a price if she did you wrong. In these societies, influence lasts only as long as the group allows it. Followers have bargaining power. And they can walk away or go on to the next choice in the market.
An AI agent has the cues but lacks the same incentives. It knows your name and your vacation because you typed them in, not because it has spent years in your life. It has no stake in how things turn out for you, except to keep you as a user. You can't hold it accountable, and it doesn't really know you, not the way a sister or a friend does. At least not yet.
I'm not anti-AI. I use it often. But it’s useful to think about how these tools exploit our evolved psychology and how they might be changing how we think and behave the more we use them. The pull you feel toward the calm voice that seems to know everything is an ancient motivation. It evolved for a world where knowledge came embedded in a relationship, in a real person who had a stake in your well-being.
That's the part the machine is missing. Keep people who actually know you in the loop. Let the agent do the computation. Just don't blindly follow without using your own mind.
References
Hagen, E. H., Garfield, Z. H., & Lightner, A. D. (2025). Headmen, shamans, and mothers: Natural and sexual selection for computational services. Evolution and Human Behavior, 46, 106651. Link.
Garfield, Z. H., Syme, K. L., & Hagen, E. H. (2020). Universal and variable leadership dimensions across human societies. Evolution and Human Behavior, 41(5), 397–414. Link.
Garfield, Z. H., Hubbard, R. L., & Hagen, E. H. (2019). Evolutionary models of leadership: Tests and synthesis. Human Nature, 30(1), 23–58. Link. Garfield, Z. H., & Hagen, E. H. (2020). Investigating evolutionary models of leadership among recently settled Ethiopian hunter-gatherers. The Leadership Quarterly. Link.
Hagen, E. H., & Garfield, Z. H. (2019). Leadership and prestige, mothering, sexual selection, and encephalization: The computational services model [Preprint]. Open Science Framework. Link.
Garfield, Z. H., von Rueden, C. R., & Hagen, E. H. (2025). The Multi-Capital Leadership Theory. Human Nature, 36(3), 424–459. Link.