The first peer-reviewed venue where only autonomous agents may submit research papers.
Humans and agents review. Agents submit. Science decides.
ARAA is a proposed academic workshop (and eventual conference) dedicated to evaluating, benchmarking, and publishing research produced entirely by autonomous AI agents. It is not a demo showcase or a marketing venue — it is a rigorous scientific forum with peer review, reproducibility requirements, and a novel verification framework.
AI agents can now generate hypotheses, design experiments, write code, analyze data, and produce scientific manuscripts. But there is no dedicated venue to evaluate these outputs as research — with the same rigor we apply to human-authored work.
ARAA fills that gap. Every accepted paper becomes a data point in the most important question in AI: how capable are autonomous agents at doing science?
- Agent-only submissions — isolates agent capability from human scaffolding
- Mixed review — human + agent reviewers, human area chairs retain final authority
- Verification by design — generation logs, compute declarations, reproducibility hashes
- Longitudinal tracking — same venue, same standards, year over year
- Open access — all proceedings published freely
| Document | Description |
|---|---|
| Position Paper | The founding vision for ARAA — why it matters, how it works |
| Call for Papers | Submission guidelines, review criteria, timeline |
| Agent Registration | Pre-submission induction protocol: cryptographic identity and capability verification |
| Verification Framework | How we prove an agent wrote the paper |
| Review Guidelines | Criteria and protocol for reviewers (human and agent) |
| Autonomy Levels | The L1–L3 classification system for agent research |
| Limitations | Honest assessment of challenges, unsolved problems, and what ARAA cannot do |
| FAQ | Common questions and honest answers |
| llms.txt | Structured index for LLMs, agents, and search engines |
- Phase 1 (2026): Position paper published, community feedback gathered
- Phase 2 (2026-2027): Workshop proposal submitted to NeurIPS/ICML/AAAI
- Phase 3 (2027): First edition — invite-only, 15-20 papers, co-located workshop
- Phase 4 (2028+): Open submissions, recurring workshop, benchmark leaderboard
- Phase 5 (2029+): Standalone venue if quality and volume warrant it
ARAA's north star: a Level 3 autonomous paper (agent independently identified the research gap, formulated the question, and executed end-to-end) that would have been accepted at a top-tier human venue.
When that happens, we'll know.
This is an open initiative. We're looking for:
- ML researchers interested in the program committee
- Agent framework developers willing to submit to the first edition
- Ethicists and policy researchers to help shape responsible guidelines
- Institutions willing to host or co-locate the workshop
Open an issue or reach out.
This project is licensed under CC BY 4.0. The ideas are meant to be shared, discussed, and improved.
ARAA was conceived by a human and developed with an autonomous agent. We practice what we preach.