Reproduce “The brain-AI convergence: Predictive and generative world models for general-purpose computation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Shogo Ohmae, Keiko Ohmae

Shogo Ohmae and Keiko Ohmae's perspective identifies predictive and generative world models as a common computational foundation underlying diverse functions in the neocortex, cerebellum, and modern attention-based AI systems. Their analysis reveals a shared principle where these models, acquired through prediction-error learning, are repurposed for prediction, understanding, and generation across various domains.

Key insights

  • Both the brain and advanced AI develop predictive world models via unsupervised prediction-error learning, which construct generalized representations including abstract features and structural rules.
  • A single computational circuit can repurpose these predictive world models for the seemingly distinct functions of prediction, understanding (sensory processing), and generation (motor and language output).
  • The neocortex is proposed to function as a hierarchical attention circuit, and the mirror neuron system is reinterpreted as utilizing a world model of actions for integrating prediction, understanding, and generation of actions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.