Terry Tao on how university students should control AI diet
I watched an insightful presentation by mathematician Terence Tao on essentially the topic in the title of this blog post. I found them very insightful and decided to use AI to summarize the main takeaways below.
This lecture, delivered by Professor Terence Tao as part of the 2026 EMS Lecture Series on Mathematics Education, explores how university students should manage their "AI diet" in an era of cognitive abundance. Drawing a parallel to nutrition and physical exercise, Tao discusses the shift from an age of cognitive scarcity—where tasks were difficult and provided natural mental exercise—to one where AI tools can easily bypass the struggle required for skill development.
Key Takeaways:
-
The Risks of Unrestricted AI Use (23:36-40:39):
- Deskilling: Over-reliance on AI can lead to the atrophy of foundational problem-solving and critical thinking skills.
- Learned Helplessness: Students may feel unable to start or solve problems without AI assistance.
- Loss of Cognitive Diversity: AI models often converge on mainstream, popular, or repetitive answers rather than unique, diverse, or organic human insights.
- Sycophancy: AI's tendency to prioritize affirmation over accuracy can hinder constructive learning.
-
Recommendations for Education (41:49-51:38):
- De-emphasize the "Correct Answer": Focus on the process of discovery and verification rather than just the final result, which AI can easily provide.
- Encourage "Productive Failure": Normalize the experience of struggling or failing first, as this is where true learning occurs. Teachers should create tasks where AI is used to assist or be coached, rather than to replace effort.
- Spicing Up Curriculum: Use AI sparingly to present concepts in unusual or creative formats to keep students engaged, avoiding a monotone, textbook-only approach.
-
Blue Team vs. Red Team (52:54-54:55): Tao advocates for a strategy where students use AI for creative production (Blue Teaming) only if they have the skills to rigorously critique and verify the output (Red Teaming). Students must be capable of explaining their work and justifying their use of AI tools in class (55:36-56:16).
Conclusion:
Rather than banning technology, Tao suggests a hybrid approach—treating AI like a vehicle that should be used for specific tasks to free up time, while still "walking" (using traditional mental efforts) to maintain core intellectual fitness. Open, transparent dialogue between educators and students about the healthy use of these tools is essential to navigating this transition.