New Lex Fridman Insight: Vladimir Vapnik: Predicates, Invariants, and the Essence of Intelligence
Sent June 11, 2026
Key Insights
- Vladimir Vapnik argues that understanding intelligence is a philosophical problem, distinct from engineering intelligence, which imitates human activity.
- Vapnik suggests that using predicates can significantly reduce the amount of data needed for tasks like digit recognition, potentially needing 100 times fewer examples.
- He emphasizes the importance of discovering good predicates in machine learning to improve performance and reduce the set of admissible functions.
- Vapnik believes human-level intelligence may not have a closed-form solution, implying reliance on heuristics and philosophical understanding.
- The conversation highlights the influence of Vladimir Propp's structural analysis of narratives on understanding human behavior and intelligence.
How the conversation moved
The episode begins with Vladimir Vapnik framing the problem of intelligence as fundamentally philosophical, distinct from the engineering challenge of replicating human-like behavior. He argues that while engineering intelligence focuses on creating devices that imitate human actions, understanding intelligence involves grappling with abstract ideas and philosophical questions. This sets the stage for a deeper exploration of how predicates and invariants play a role in both understanding and engineering intelligence.
Vapnik's main argument centers around the use of predicates to improve machine learning efficiency. He suggests that discovering good predicates can drastically reduce the amount of data needed for tasks like digit recognition, potentially requiring 100 times fewer examples. This claim is supported by references to Vladimir Propp's work on narrative structures, which Vapnik believes can provide insights into human behavior and intelligence, thereby informing AI development.
Despite the compelling nature of Vapnik's argument, there is little direct pushback from the host. However, the conversation does touch on potential skepticism regarding the feasibility of reducing data requirements so drastically. Vapnik's assertion that logic-based systems alone cannot find good predicates without understanding reality invites implicit critique, as it challenges the prevailing reliance on purely logical or data-driven approaches in AI.
The conversation concludes with Vapnik reflecting on the philosophical nature of intelligence and the role of heuristics. He suggests that human-level intelligence might not have a closed-form solution, implying a need for philosophical understanding and heuristic methods. The discussion also highlights the influence of literary analysis, particularly Vladimir Propp's structural approach, on understanding intelligence, leaving open questions about how these insights can be practically applied in AI.
Surprising moments
In-depth
Predicates in Machine Learning
- Predicates can reduce the amount of training data needed by narrowing the set of admissible functions.
- Good predicates are crucial for improving machine learning performance.
- Vladimir Propp's narrative structures offer insight into human behavior and intelligence.
Philosophical Challenges of Intelligence
- Understanding intelligence is a philosophical problem distinct from engineering it.
- Human-level intelligence may not have a closed-form solution, relying on heuristics.
Notable Quotes
Engineering is imitation of human activity. You have to make a device which behave as human behave, have all the functions of human.
Still open
- Vapnik questions whether human-level intelligence can ever be fully understood or replicated without philosophical insight.