AI is doing something weird to Science

2025-09-10 · Source: The Computist Journal · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Research Methodology & Innovation · Depth: Advanced, extended

Summary

AI is fundamentally altering scientific discovery by integrating large language models (LLMs) into a human-AI "loop" that accelerates candidate generation. This loop, comprising a human poser, an AI proposer, an independent verifier, and a human curator, is exemplified by cases like Donald Knuth's "Claude's Cycles" (combinatorics with Claude Opus 4.6), Terence Tao's mathematical proofs with LLMs and Lean, AlphaFold's protein structure predictions, and GNoME/A-Lab's materials discovery. The critical shift since 2022 is that general-purpose LLMs now efficiently occupy the proposer role, offering open-ended program synthesis and cross-domain transfer, making candidate generation cheaper and faster. This contrasts with earlier computational assistance, where proposers were domain-specific. The article stresses that the verifier's reliability is paramount, as demonstrated by Galactica's failure without one, and argues that the loop, not just the AI or human, drives discovery.

Key takeaway

For research scientists building AI-assisted discovery systems, recognize that the verifier is the critical component, not the AI proposer. You should invest heavily in designing robust, independent verification mechanisms, whether formal proof checkers or physical experiments, to ensure scientific validity. Prioritize hiring individuals skilled in posing incisive questions and constructing reliable verifiers, as these human-centric roles remain indispensable for generating trustworthy scientific advances.

Key insights

Scientific discovery now thrives in a human-AI loop, with AI proposing and robust verifiers confirming.

Principles

• The loop (poser, proposer, verifier, curator) drives scientific discovery.
• Verifier reliability is critical; a weak verifier yields confident nonsense.
• LLMs make the proposer role cheaper and more general across domains.

Method

A human poses questions, an AI model proposes candidates, an independent verifier filters, and a human curates the surviving results.

In practice

• Build robust, independent verifiers for AI-generated proposals.
• Prioritize question-posing and verifier-design skills in research.
• Scrutinize AI-in-science results lacking a clear, external verifier.

Topics

• AI in Scientific Discovery
• Large Language Models
• Formal Verification
• Materials Synthesis
• Protein Folding
• Research Methodology

Best for: AI Scientist, Research Scientist, Director of AI/ML

Related on AIssential

• TAI #192: AI Enters the Scientific Discovery Loop — AI models are transitioning from tools to active participants in scientific discovery, particularly when coupled with robust verification.
• AI ‘scientists’ promise to accelerate research — how do they work? — AI agents accelerate research, yet human involvement is essential for sound scientific practice.
• 🔬 Automating Science: World Models, Scientific Taste, Agent Loops — Andrew White — AI agents like Cosmos are automating scientific discovery by integrating literature, data analysis, and experimental feedback loops.
• The Year in Physics — AI generates novel physics experiments, while new astronomical and geological findings challenge established scientific paradigms.
• Mar 23, 2026ScienceVibe physics: The AI grad student — AI can significantly accelerate expert-guided theoretical physics research, but requires rigorous human oversight.
• Inference: When Will We Solve Intelligence — Solving intelligence requires a hybrid approach combining formal verification with large language models for robust reasoning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Computist Journal.