Glossary of the Cognitive Risks of Generative AI

The set of skills and knowledge required to use AI effectively, understanding its capabilities, limitations, and risks. Since February 2, 2025, Article 4 of the AI Act requires providers and deployers of AI systems to take measures to ensure a sufficient level of AI literacy among their staff.

The tendency to trust recommendations from an automated system beyond what its reliability warrants, to the point of no longer verifying them or ignoring contradictory information. A concept synthesized by Parasuraman and Manzey (2010).

A decrease in vigilance when dealing with an automated system deemed reliable. It often precedes automation bias (Parasuraman and Manzey, 2010).

A paradox described by Lisanne Bainbridge in 1983: the more we automate, the more difficult the tasks left for humans become, even as their skills erode due to lack of practice.

The deficit in understanding and memorization that accumulates when the effort of thinking is delegated to AI. Measured by Nataliya Kosmyna's team at the MIT Media Lab (2025) and theorized in software engineering by Margaret-Anne Storey (University of Victoria).

According to Margaret-Anne Storey, the loss of traceability for the objectives and reasons behind a decision when the work is produced by AI. The PROBE© method is specifically designed to make intent explicit.

The gap between what a person produces with AI and what they actually understand. It makes one incapable of correcting work whose structure is not mastered.

The loss of acquired skills from repeatedly delegating them to an automated system. Observed in endoscopists after exposure to AI (The Lancet Gastroenterology & Hepatology, 2025) and analyzed in computer-aided design by Shukla et al. (CHI 2025).

The act of entrusting part of the mental effort—memorizing, calculating, reasoning—to an external tool. Useful in small doses, it becomes a risk when it replaces learning; Chirayath et al. (Frontiers in Psychology, 2025) study its effect on the ability to cope with difficulties.

The tendency to delegate the regulation of one's own learning—planning, monitoring, evaluating—to AI. Described by Yizhou Fan and his colleagues in the British Journal of Educational Technology.

The ability to think about one's own thinking: observing how one learns and reasons while working, including with AI. Mustapha Benkalfate considers it a key career skill.

Effort that is deliberately maintained in a task to preserve a skill. The Revenu IA methods are designed to maintain minimal but real friction.

AI-generated content that looks like finished work but lacks substance, shifting the burden of correction to the person who receives it. A term popularized by the Harvard Business Review in 2025.

The mental cost of shifting from producing content to evaluating AI-generated content. Simkute, Tankelevitch, Sellen et al. (Microsoft Research, 2024) show that it can negate expected productivity gains.

The use of AI tools by employees that are not authorized by the organization. According to the 2024 Work Trend Index from Microsoft and LinkedIn, 78% of AI users at work bring their own tools.