AI 'Generate' Buttons Mirror Slot Machine Psychology, UX Expert Warns

The uncomfortable parallel

Christian Kuhn, a UX practitioner with over twenty years of experience, has identified a troubling pattern in generative AI products: the core interaction model—type a prompt, click generate, wait for results, then re-roll if unsatisfied—mirrors the psychological mechanics of slot machines.

Writing for AI Watch, Kuhn describes the moment between clicking "Generate" and seeing results as functionally identical to what casino researcher Natasha Dow Schüll calls "the zone" in her fifteen-year study of Las Vegas slot players. The difference is that AI users are paying per spin through token-based pricing models.

The behavioral science foundation

Kuhn grounds his argument in three established research findings. B.F. Skinner's 1957 work on reinforcement schedules showed that variable ratio rewards—where payoffs arrive after unpredictable numbers of attempts—produce the most persistent behavior in animals. Wolfram Schultz's 1998 research demonstrated that dopamine neurons fire not when rewards arrive, but when they exceed predictions. Kent Berridge's ongoing work has separated "wanting" from "liking" in the brain, showing that dopamine drives the urge to repeat an action without necessarily increasing satisfaction.

These three mechanisms, Kuhn argues, combine to create the same behavioral signature found in slot machines and social media feeds—and now in AI generation tools.

Tokens as casino chips

Most generative AI services have moved away from flat subscription pricing toward token-based models where users pay per generation. Midjourney sells GPU minutes, Runway sells generations, and ChatGPT imposes message caps. Kuhn characterizes these tokens as "casino chips"—purchased in advance, spent on uncertain outcomes, and depleted through repeated re-rolls.

The friction-free re-roll button, present in nearly every consumer image generator, functions as what casinos call a "max bet" lever. Users who have invested time refining prompts experience what behavioral economists call the IKEA effect—placing higher value on outputs they helped create, even when quality is poor. This keeps them spending tokens to inch toward an acceptable result.

The evidence gap

Kuhn acknowledges that his strongest claim—that AI tools produce clinically measurable compulsive use patterns—outpaces current research. While 2025 studies have developed scales for "Generative AI Dependency" and "Problematic ChatGPT Use," the longitudinal population studies comparable to social media or loot box research don't yet exist. The structural evidence is clear; the clinical harm data is preliminary.

He advocates for immediate commissioned research rather than waiting for regulators to mandate it.

Design interventions

Kuhn proposes concrete changes based on established human-centered AI principles. Products could detect when users click re-roll more than four times in five minutes and offer a gentle interruption: "You've generated nine variants. Want to refine the brief?" This returns control without imposing hard stops.

The shift from subscription to metered pricing, Kuhn notes, was a choice—not a technical necessity. Flat-rate pricing remains viable for AI products, just as "unlimited" mobile plans eventually became standard after regulatory pressure.

Why it matters

As generative AI moves from novelty to daily workflow tool for knowledge workers, the cost structure and interaction patterns shape not just user experience but professional economics. If the dominant pricing model converts every creative decision into a metered transaction, and the interface exploits known compulsion mechanics, the field risks repeating the mistakes of social media—building engagement systems that extract value while eroding user agency. The window to choose different design patterns is narrow and closing.

The analysis was first reported by Christian Kuhn in AI Watch, drawing on his work running a UX practice and teaching human-centered AI design principles.