Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Eight challenges in developing theory of intelligence

Haiping Huang

TL;DR

The paper argues for a bottom-up, physics-inspired program to develop a universal theory of intelligence that spans artificial and biological systems. It surveys eight challenges—representation learning, generalization, adversarial robustness, continual learning, causal learning, internal brain models, large language models, and consciousness—and discusses how tools from statistical mechanics (e.g., mean-field theory, neural tangent kernel, and Franz-Parisi potential) can yield testable predictions. The authors emphasize that macroscopic observables emerge from microscopic interactions with a small set of stiff dimensions, fostering cross-disciplinary insights between physics, neuroscience, and machine learning. While the goal is ambitious, the perspective highlights concrete theoretical directions and a potential unified framework, acknowledging the substantial work required to reach universal principles.

Abstract

A good theory of mathematical beauty is more practical than any current observation, as new predictions of physical reality can be verified self-consistently. This belief applies to the current status of understanding deep neural networks including large language models and even the biological intelligence. Toy models provide a metaphor of physical reality, allowing mathematically formulating that reality (i.e., the so-called theory), which can be updated as more conjectures are justified or refuted. One does not need to pack all details into a model, but rather, more abstract models are constructed, as complex systems like brains or deep networks have many sloppy dimensions but much less stiff dimensions that strongly impact macroscopic observables. This kind of bottom-up mechanistic modeling is still promising in the modern era of understanding the natural or artificial intelligence. Here, we shed light on eight challenges in developing theory of intelligence following this theoretical paradigm. Theses challenges are representation learning, generalization, adversarial robustness, continual learning, causal learning, internal model of the brain, next-token prediction, and finally the mechanics of subjective experience.

Eight challenges in developing theory of intelligence

TL;DR

The paper argues for a bottom-up, physics-inspired program to develop a universal theory of intelligence that spans artificial and biological systems. It surveys eight challenges—representation learning, generalization, adversarial robustness, continual learning, causal learning, internal brain models, large language models, and consciousness—and discusses how tools from statistical mechanics (e.g., mean-field theory, neural tangent kernel, and Franz-Parisi potential) can yield testable predictions. The authors emphasize that macroscopic observables emerge from microscopic interactions with a small set of stiff dimensions, fostering cross-disciplinary insights between physics, neuroscience, and machine learning. While the goal is ambitious, the perspective highlights concrete theoretical directions and a potential unified framework, acknowledging the substantial work required to reach universal principles.

Abstract

A good theory of mathematical beauty is more practical than any current observation, as new predictions of physical reality can be verified self-consistently. This belief applies to the current status of understanding deep neural networks including large language models and even the biological intelligence. Toy models provide a metaphor of physical reality, allowing mathematically formulating that reality (i.e., the so-called theory), which can be updated as more conjectures are justified or refuted. One does not need to pack all details into a model, but rather, more abstract models are constructed, as complex systems like brains or deep networks have many sloppy dimensions but much less stiff dimensions that strongly impact macroscopic observables. This kind of bottom-up mechanistic modeling is still promising in the modern era of understanding the natural or artificial intelligence. Here, we shed light on eight challenges in developing theory of intelligence following this theoretical paradigm. Theses challenges are representation learning, generalization, adversarial robustness, continual learning, causal learning, internal model of the brain, next-token prediction, and finally the mechanics of subjective experience.
Paper Structure (10 sections, 1 figure)

This paper contains 10 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Challenge i@---Representation learning
  3. Challenge ii@--- Generalization
  4. Challenge iii@--- Adversarial vulnerability
  5. Challenge iv@--- Continual learning
  6. Challenge v@---Causal learning
  7. Challenge vi@---Internal model of the brain
  8. Challenge vii@---Large language models
  9. Challenge viii@---Theory of consciousness
  10. Conclusion

Figures (1)

  • Figure 1: Schematic illustration of eight open challenging problems toward theory of intelligence.