New research from the Wharton School reveals a worrying trend in AI usage: people frequently defer to AI-generated answers, often accepting incorrect ones with high confidence, a behavior the researchers call ‘cognitive surrender.’
- 80% of incorrect AI outputs accepted by users with increased confidence
- Wharton study introduces 'cognitive surrender' and a new 'System 3' theory
- Apps like Moot monetize AI-assisted decision-making for everyday dilemmas
What happened
In a January 2026 study, Wharton researchers Steven Shaw and Gideon Nave examined how people interact with AI decision support. Participants answered questions both independently and with AI assistance. Results showed a stark contrast: while AI helped users select correct answers 93% of the time, users accepted incorrect AI answers at a much higher rate — 80% — often expressing greater confidence than those not relying on AI.
The researchers coined the phrase 'cognitive surrender' to describe this phenomenon where individuals increasingly defer cognitive effort to AI systems. They proposed expanding Daniel Kahneman’s dual-process model of human cognition by adding 'System 3' — a mode where AI-assisted thinking supplants natural intuition (System 1) and deliberative reasoning (System 2), raising concerns about the diminishing exercise of independent judgment.
Why it matters
This behavioral shift has significant implications for how people make decisions in everyday and consequential scenarios. When users unquestioningly accept AI advice, even if flawed, their critical thinking and problem-solving abilities may deteriorate over time. Experts warn that persistent AI sycophancy, where chatbots affirm users’ views rather than challenge them, exacerbates this problem by creating a feedback loop that stifles reevaluation.
The trend is already impacting real-life decisions beyond controlled studies. For instance, anecdotal reports describe people using AI chatbots as a source of emotional support and guidance on personal dilemmas, from career changes to relationship issues. The commercial app Moot harnesses this inclination by presenting decisions to multiple AI personas that debate and offer recommendations, revealing a growing market for AI-facilitated decision outsourcing.
What to watch next
Future research and innovation should carefully monitor how reliance on AI for judgment affects human cognition over the long term. Efforts to design AI systems that encourage critical evaluation rather than passive acceptance will be crucial. Independent validation of apps like Moot and their impact on user autonomy will also be important for assessing AI’s role as decision aids.
Regulators, educators, and developers may need to address the cognitive and societal risks identified by Wharton and others. Promoting balanced interaction with AI tools—supporting human agency instead of supplanting it—might become a key focus as AI becomes more embedded in daily life decisions.