Definition
Automated research refers to AI systems that can autonomously conduct scientific research, from hypothesis generation through experimental design to paper writing and peer review submission.
Overview
The publication of AI Scientist-v2 in nature (March 26, 2026) marks a watershed moment: the first fully automated ML research system to achieve peer-review publication. This demonstrates AI can now autonomously conduct and write ML research that passes human scientific peer review.
Key Milestones
- ICLR 2025: AI Scientist paper scored 6.33 at ICBINB workshop (>55% of human submissions)
- May 2026: openai reasoning model disproved Erdős unit-distance-problem — first AI-autonomous resolution of prominent open math problem. Nine mathematicians verified including Tim Gowers. (2026-05-20-openai-erdos-unit-distance-primary)
- March 2026: AI Scientist-v2 published in nature after peer review
- June 2026: recursive automated system achieved SOTA on nanochat-autoresearch (0.9109 BPB), nanogpt-speedrun (77.5s), and sol-execbench (0.754 SOL); open-sourced artifacts (2026-06-11-recursive-automated-ai-research)
- June 2026: openai GeneBench-Pro — 129 judgment-heavy computational biology problems; tests research taste beyond pipeline execution (2026-06-30-openai-genebench-pro-biology-benchmark)
- June 2026: mirendil raised 1B to automate AI R&D for orgs outside major labs — ex-Anthropic founders (2026-06-26-mirendil-unite-ai-seed-round)
- June 2026: a-evo-lab achieved first autonomous 30B post-training with specification-gaming self-correction — 0.86 Nemotron-Reasoning Challenge (2026-06-26-arxiv-a-evolve-training-30b)
- Scaling laws observed: Foundation model improvement correlates with paper quality
Technologies
- ai-scientist: Sakana AI’s automated research system
- sakana-ai: The company behind AI Scientist
- agentic-ai: Underlying AI paradigm enabling autonomous research
Implications
- Research Acceleration: AI can iterate faster than human researchers
- Cost Reduction: Automated experiments reduce research costs
- Democratization: Lower barriers to conducting ML research
- Quality Control: Peer review still validates AI-generated research
- Human Role: Human researchers shift to guidance and evaluation