Workshop III: Naturalistic Approaches to Artificial Intelligence

Part of the Long Program Mathematics of Intelligences
November 4 - 8, 2024

Overview

Many approaches to artificial intelligence are inspired by natural systems; for example, deep learning draws inspiration from biological neural networks. In recent years, researchers have looked to alternative biological inspiration. One paradigm develops distributed architectures for artificial intelligence, looking to biological evolution, insect swarms, and immune systems for models. Another paradigm argues that many forms of thinking are essentially equivalent to programming; this paradigm emphasizes the inference of (latent) programs from experience.

These paradigms have many appealing features. For example, evolutionary programming provides a powerful framework for search over machine learning architectures. Program synthesis approaches (in the “thinking as programming” paradigm) can learn latent, interpretable programs for complex tasks beyond conventional deep learning approaches.

These naturalistic approaches lack a detailed theory that explains their power. Both cases involve optimization (sometimes, combinatorial optimization) over high dimensional, complex objective functions. Both also involve basic objects (like string-based representations or programs) with rich structure and few obvious symmetries, which have only recently been studied as mathematical objects in their own right.

This workshop will draw together researchers creating new algorithms and architectures (e.g., active symbol architectures, evolutionary programming approaches, neural program synthesis) with mathematicians and theoretical computer scientists who specialize in non-convex optimization, the theory of programming languages, type theory, proof theory, and category theory. It aims to promote cross-fertilization between these paradigms and more traditional approaches, while stimulating the development of rigorous foundations for evolutionary computing, program synthesis, and other naturalistic approaches to AI.

Organizing Committee

Stephanie Forrest (Arizona State University)
Tom Griffiths (Princeton University)
Sumit Gulwani (Microsoft Research)
Josh Tenenbaum (Massachusetts Institute of Technology)