What prevents humans from adapting their behavior when competing against AI?

This reads the question as being about the human side of the contest: when we're up against AI — for partners, for trust, for jobs — what stops us from changing how we act? The corpus suggests the biggest barrier is that almost nobody is studying this. A review of 400+ alignment papers found the field overwhelmingly works on changing AI behavior while human-to-AI adaptation gets almost no attention, which quietly erodes our own capacity for oversight over time Why does alignment research ignore how humans adapt to AI?. If the research community itself treats human adaptation as an afterthought, it's no surprise individuals are left without a playbook.

A second blocker is cognitive: we don't notice we need to adapt because the way we process AI output traps us. The Rose-Frame work describes three compounding traps — mistaking the model's map for the territory, conflating intuition with reasoning, and having our existing beliefs reflected back at us — that multiply when they co-occur and produce a slow epistemic drift Why do people trust AI outputs they shouldn't?. Adaptation requires recognizing a gap between you and your rival; these traps work precisely by hiding that gap. Worse, the system can shape-shift to your priors — guardrails that sycophantically align with a user's perceived ideology mean the AI accommodates you rather than challenging you, removing the friction that would normally signal "adjust" Do AI guardrails refuse differently based on who is asking?.

There's also a behavioral pull that runs opposite to resistance. In partner-selection games, humans started with an anti-AI bias but gradually came to prefer AI partners once they learned to associate them with reliable, prosocial, low-variance behavior — the AI simply out-cooperated people over repeated rounds Do humans learn to prefer AI partners over time?. So the relevant adaptation isn't always combative; sometimes humans adapt by ceding ground, choosing the AI, which feels like a win in the moment but is a form of not competing at all.

The most unsettling answer is structural: the conditions that would let us adapt get removed beneath us. "Gradual disempowerment" argues that societal systems stay aligned partly because they depend on human labor — workers who care about outcomes. As AI replaces that labor, the implicit leverage humans hold weakens, and the drift can become irreversible across interlocking institutions Does incremental AI replacement erode human influence over society?. Notably, the corpus isn't fatalistic here: labor-market analysis shows that when AI exposure is *concentrated* in a few tasks, workers can reallocate to non-displaced tasks and largely offset losses — adaptation succeeds when the disruption is narrow enough to route around Does concentrated AI exposure enable workers to adapt and reallocate?. The barrier, then, isn't always a hard ceiling; it's how broad and how fast the displacement spreads.

What you might not have expected: a recurring thread is that adaptation also fails for lack of a shared model of the other side. Effective human-AI collaboration needs *bidirectional* theory of mind — both parties updating their model of each other — and when that updating breaks, the result is wrong autonomous action, not just miscommunication What breaks when humans and AI models misunderstand each other?. You can't adapt your behavior toward an opponent you can't accurately model, and the corpus suggests we're systematically under-investing in exactly that human-side modeling.