969: The Laws of Thought: The Math of Minds and Machines, with Prof. Tom Griffiths

2026-02-24 · Source: Super Data Science: ML & AI Podcast with Jon Krohn · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Mathematics & Computational Sciences · Depth: Intermediate, extended

Summary

Princeton Professor Tom Griffiths discusses his new book, "The Laws of Thought," which explores the mathematical models underlying both biological and artificial intelligence. He highlights how mathematical principles, traditionally used for the physical world, can also describe cognitive science, or "the internal world." Griffiths' research at Princeton's Computational Cognitive Science Lab and AI Lab focuses on understanding human cognition through computational ideas and enhancing AI with insights from human minds. He notes that while psychologists often view humans as irrational, computer scientists find human intelligence inspiring. The discussion covers the historical shift from symbolic AI to probabilistic and neural approaches, the impact of training data sensitivity on large language models (LLMs), and the critical role of inductive biases in human learning compared to AI's data-intensive methods. The conversation also touches on modeling curiosity in AI and the challenges of evaluating AI systems scientifically.

Key takeaway

For AI scientists and research scientists developing next-generation models, recognize that current LLMs' probabilistic nature makes them prone to outputting more common answers, even for deterministic tasks. Your systems will benefit from incorporating stronger inductive biases, potentially through techniques like meta-learning, to achieve human-like efficiency in learning from limited data and to foster genuinely novel, less probable outputs. This approach can help bridge the significant gap between human and AI learning efficiency.

Key insights

Mathematical principles can describe both external physical laws and internal cognitive processes in biological and artificial intelligence.

Principles

• Human cognition excels at inductive problems, making inferences from incomplete data.
• AI systems' behavior is highly sensitive to their training data and inherent probabilistic nature.
• Inductive biases are crucial for efficient learning, bridging the gap between human and AI data requirements.

Method

Cognitive science methods, including targeted experiments and large-scale online data collection, are used to study both human and AI systems by analyzing behavioral responses to specific problems.

In practice

• Use compositional prompting to explore lower probability outputs from LLMs.
• Consider meta-learning to instill human-like inductive biases in neural networks.

Topics

• Mathematical Cognition
• Large Language Models
• Inductive Bias
• AI Evaluation
• Curiosity in AI

Best for: AI Scientist, Research Scientist, AI Researcher, Data Scientist, Machine Learning Engineer

Related on AIssential

• Connecting Language and (Artificial) Intelligence: Princeton’s Tom Griffiths — Intelligence, human and artificial, is best understood through the complementary interplay of logic, neural networks, and probability.
• Why Every Brain Metaphor in History Has Been Wrong [SPECIAL EDITION] — Scientific models are useful simplifications, not necessarily literal truths about underlying reality.
• Uneven Evolution of Cognition Across Generations of Generative AI Models — Generative AI models exhibit uneven cognitive development, excelling in language but struggling with visual-perceptual reasoning.
• System 3 AI: No Humans Needed — System 3 AI integrates formal logic and verifiable code into LLMs for autonomous, complex reasoning beyond human feedback.
• Statistical Properties of Training & Generalization — Deep learning's success challenges classical statistics, driven by physics-informed choices and neural scaling laws.
• AI and the human mind: only one is a black box — Behavioral evidence alone may not sufficiently prove human-level intelligence in LLMs due to fundamental asymmetries.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Super Data Science: ML & AI Podcast with Jon Krohn.