Announcing an automatic theorem proving project

Summary (AI generated)

Archived original version »

Tim Gowers announces a new project aimed at developing an automatic theorem prover using GOFAI (Good Old-Fashioned Artificial Intelligence), prioritizing algorithms that mimic human mathematical reasoning over machine learning or brute-force computation. The goal is to create systems capable of solving problems as humans do, emphasizing logical steps and strategic thinking rather than computational power alone.

Key Aspects: 1. Methodology:

  • Focus on human-like problem-solving, avoiding ML/black-box approaches.

  • Design algorithms that replicate intuitive mathematical reasoning, including recognizing “silly” or redundant steps to avoid inefficiency.

  1. Collaboration Structure:

    • A core team will meet regularly (e.g., twice weekly) for focused discussions and strategic planning.

    • Remote participation is encouraged through a dedicated website or moderated platforms, balancing openness with inclusivity for marginalized groups.

    • Emphasis on shared progress tracking, including open questions about algorithm design (“How should a computer do X?”).

  2. Target Candidates:

    • PhD/postdoc applicants with strong problem-solving experience (e.g., tackling research-level math problems) are ideal, even if coding skills are limited.

    • Prioritize deep understanding of algorithmic logic over programming fluency; coders can later implement designs.

  3. Inclusivity Considerations:

    • Acknowledges barriers for underrepresented groups in open collaboration (e.g., fear of criticism).

    • Plans include moderated spaces to ensure safe participation, possibly separating public/private contributions.

  4. Unique Focus:

    • Demystifying human intuition: Why certain steps feel “obvious” or “silly” to mathematicians.

    • Strategic prioritization of tasks to maximize progress while maintaining focus on core goals.

The project seeks a collaborative environment where diverse ideas are shared iteratively, aiming to bridge the gap between human mathematical insight and computational logic without relying on modern AI trends.