Conjecture Machines: AI agents and the new validation bottleneck in science
Don Wallace, Conor Griffin, Sean O’Neill, Thang Luong and Owen Larter. Google DeepMind, July 2026.
This policy analysis draws on conversations with ten Google DeepMind researchers and engineers. It argues that AI-assisted ideas and candidate solutions are becoming more abundant while scientific validation remains costly and slow. It discusses experimental testing, formal verification, peer review and people’s understanding of results. Its account motivates investigation of checking capacity; it is not a controlled measurement of a universal bottleneck or an evaluation of 1Cademy.
Retrieved 2026-11-11.
0
1
Contributors are:
Tags
Scientific Oversight of AI-Assisted Research
1Cademy
Related
Working with academic communities to build scientific understanding
A new golden age of discovery
AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms
Claude discovers a novel enzyme system with CRISPR-like repeats
Intentmaking and Sensemaking: Human Interaction with AI-Guided Mathematical Discovery
Theorem Proving in Lean 4: Propositions and Proofs
What is Microsoft Discovery?
Independent Assessment of AI R&D Acceleration at Anthropic in October 2026
Improving mathematical reasoning with process supervision
Tracing the thoughts of a large language model
Investigating vault particles with Claude
Evaluating Claude’s bioinformatics research capabilities with BioMysteryBench
BioMysteryBench Changelog — v11 (2026-07-06)
AI-assisted teams outperform AI-led teams but not human-only teams in assessing research reproducibility in quantitative social science
AI for scientific discovery: How researchers use Claude
Measuring how AI agents handle everyday wet-lab science
Deep learning-guided discovery of an antibiotic targeting Acinetobacter baumannii
Related
Working with academic communities to build scientific understanding
A new golden age of discovery
AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms
Claude discovers a novel enzyme system with CRISPR-like repeats
Intentmaking and Sensemaking: Human Interaction with AI-Guided Mathematical Discovery
Theorem Proving in Lean 4: Propositions and Proofs
What is Microsoft Discovery?
Independent Assessment of AI R&D Acceleration at Anthropic in October 2026
Improving mathematical reasoning with process supervision
Tracing the thoughts of a large language model
Investigating vault particles with Claude
Evaluating Claude’s bioinformatics research capabilities with BioMysteryBench
BioMysteryBench Changelog — v11 (2026-07-06)
AI-assisted teams outperform AI-led teams but not human-only teams in assessing research reproducibility in quantitative social science
AI for scientific discovery: How researchers use Claude
Measuring how AI agents handle everyday wet-lab science
Deep learning-guided discovery of an antibiotic targeting Acinetobacter baumannii