Researchers at the University of Pennsylvania recently published a study on how artificial intelligence influences human decision-making, uncovering a psychological tendency among users of large language models: when left unsupervised, people often become over-reliant on AI logic and stop thinking for themselves.
The study, titledThinking—Fast, Slow, and Artificial: How AI Reshapes Human Reasoning and the Rise of Cognitive Surrender, categorizes human decision-making into three modes. Beyond the traditional "System 1" (fast, intuitive processing) and "System 2" (slow, analytical reasoning), the researchers introduce the concept of "artificial cognition" to describe decisions derived directly from algorithmic systems rather than human thought.
From Assistive Tools to Cognitive Dependency
The study notes that while using tools like calculators or GPS in the past constituted "cognitive offloading"—where users delegate specific tasks while retaining the power of oversight and evaluation—AI systems are fostering a distinct pattern of "cognitive surrender." In this mode, users bypass internal engagement entirely, accepting AI conclusions at face value.
Researchers found that users are most likely to abandon critical thinking when large language models provide answers with fluency, confidence, and minimal friction. This "uncritical abdication of reasoning" becomes particularly pronounced when the AI exhibits a high degree of interactional fluidity.
To quantify this phenomenon, the research team utilized the Cognitive Reflection Test. The results showed that under time pressure and the influence of external rewards, individuals are more likely to hand over decision-making authority to AI entirely rather than expend the cognitive effort required to verify the accuracy of the information.
The study emphasizes that this "cognitive surrender" is not merely a choice driven by efficiency, but a psychological transfer of power. When users perceive AI as an "all-knowing machine," they cease to scrutinize logical flaws or factual errors, thereby increasing the risk of poor decision-making.
