The claim: every forecast about AI quietly assumes people will behave sensibly. They won't. Loss aversion, status anxiety, tribal loyalty and plain inertia sit underneath every other theory on this site, and they change the answers.
Why it deserves its own page
Most arguments about AI are built on models of rational actors: firms adopting what's efficient, workers retraining toward opportunity, voters backing policy that helps them. Behavioral economics spent fifty years establishing that this isn't how people work. Kahneman, Thaler and Ariely are the foundations.
Treat irrationality as a variable rather than noise and the other theories shift. Adoption runs slower than capability allows, because people resist tools that threaten their standing. Political responses to displacement follow anger rather than analysis. And some of the most-cited hopes about a "meaning economy" depend on irrationality being durable: people paying a premium for a human being on the other end, for no efficient reason at all.
Where it shows up across the debate
- Tyler Cowen — "mood affiliation." People pick an AI position that matches their tribe and reason backward. He names it as a problem with the quality of the discourse itself, not just a flaw in other people.
- Peter Diamandis — humans think linearly and the technology moves exponentially, which is why its impact is perpetually surprising even to people who follow it closely.
- Eliezer Yudkowsky — built an entire intellectual community around the premise that people are systematically miscalibrated, particularly about low-probability catastrophe.
- Mustafa Suleyman — "pessimism aversion": we're reluctant to plan for bad outcomes, which makes containment harder than it needs to be.
- Marc Andreessen — argues doom framing is status-driven rather than technical. Worth noting the blade cuts both ways: panic and complacency are both irrational responses.
- Aza Raskin — "the race to the bottom of the brainstem." Engagement design already exploits predictable biases, and AI scales personalized persuasion rather than correcting for it.
- Shoshana Zuboff — surveillance capitalism needs human behavior to be predictable and nudgeable. Irrationality isn't incidental to the business model, it's the raw material.
- Carl Benedikt Frey — the empirical link between automation exposure and political backlash. The original Luddites were responding rationally to their own wage data, but displaced populations often reach for scapegoats instead.
- Richard Dawkins — ideas spread through cognitive shortcuts rather than evaluation. Language models are, among other things, enormous engines for amplifying whatever spreads well.
- Jasmine Sun — what AI means to people drives politics more than what it does. The doomer-versus-accelerationist binary is a media artifact of tribal reasoning.
- David Shapiro — his "meaning economy" is explicitly built on people not being rational optimizers. The premium paid for human contact is the feature holding the whole sector up.
Cognitive surrender
The mechanism that matters most for everything else here: people hand decisions to a confident-sounding system with far less scrutiny than they'd apply to a colleague, and override their own judgment in the process.
Steven Shaw and Gideon Nave named it in a 2026 Wharton working paper, Thinking Fast, Slow, and Artificial, which adds a "System 3" of external, algorithmic thinking to Kahneman's two systems. Across three preregistered experiments, people's accuracy simply tracked the quality of the AI's advice — better with good answers, sometimes worse than no AI at all with bad ones — and their confidence went up either way, including when they were wrong.
It's the sharpest version of the problem because it cuts against the optimistic case directly. A tool that could extend someone's thinking can just as easily replace it, and the difference isn't a property of the tool. It's whether the person stays in the loop. Access to AI doesn't determine the outcome. What you do with it does.
What it changes, theory by theory
- Automation displacement — the political response will be driven by loss aversion and status anxiety, not by optimal policy design. That shapes what's actually achievable.
- The augmentation thesis — adoption depends on human factors in both directions: resistance from people who feel diminished, over-reliance from people who don't check the output, and organizational inertia slowing everything.
- Techno-optimism — exponential blindness means optimists are regularly proven right about direction. Confirmation bias also means they overweight the evidence that agrees with them.
- Post-labor economics — the meaning economy depends on that irrational premium holding. And the politics of redistribution are governed by loss aversion more than by arithmetic.
- Alignment and existential risk — miscalibration, mood affiliation and pessimism aversion are three different accounts of why this debate goes badly, offered by people inside it.
Open questions
- Should irrationality be a parameter inside every forecast, the way a discount rate is, or its own separate analysis?
- Does AI reduce human irrationality by supplying better decision support, or amplify it through manipulation, echo chambers and cognitive surrender?
- Is the meaning economy fundamentally an irrationality economy — and if so, is that a durable foundation or a fragile one?
All theories · clickbridge.com · Rich Price · last reviewed September 2026 · People linked here have their own pages at people.clickbridge.com