AI as a Tutor: Desired Difficulty and the Risk of Outsourcing

The article analyzes the tension between the efficiency of AI in providing answers and the process of deep learning. The author invokes the concept of 'desired difficulty,' indicating that a certain amount of cognitive resistance and effort spent on retrieving information (retrieval practice) are essential for the permanent retention of knowledge and its subsequent transfer. The main risk of generative AI is the excessive reduction of cognitive friction, which leads to so-called cognitive outsourcing—a situation where the tool takes over thought processes instead of supporting them. The text emphasizes the difference between immediate performance and actual learning, suggesting that AI should act as a tutor managing the level of challenges, rather than an automaton for providing instant answers.