Feasibility of fully AI-driven research that adheres to scientific values

Determine whether fully AI-driven scientific research—executed autonomously end-to-end by artificial intelligence systems—can be conducted while adhering to the key scientific values of transparency, traceability, and verifiability.

Background

The paper investigates whether LLM agents can autonomously execute the complete scientific research workflow—from hypothesis generation and data analysis to writing and assembling a manuscript—while maintaining rigorous standards of transparency, traceability, and verifiability.

To explore this question, the authors developed data-to-paper, a platform that orchestrates multi-agent LLM interactions with rule-based checks, code execution, literature retrieval, and information tracing. Their case studies show promise for simple research goals but highlight failures and the need for human co-piloting for more complex tasks, leaving the broader feasibility across diverse contexts as an unresolved question.

References

As AI promises to accelerate scientific discovery, it remains unclear whether fully AI-driven research is possible and whether it can adhere to key scientific values, such as transparency, traceability and verifiability.

— Autonomous LLM-driven research from data to human-verifiable research papers  (2404.17605 - Ifargan et al., 2024) in Abstract (page 1)

Systems that take an idea through experiments to a manuscript are already running (Lu et al., 2024; Schmidgall et al., 2025; Tang et al., 2025), and what remains open is whether their results hold under independent verification, the capability RECLAIM measures and the one that recursive self-improvement would depend on.

— RECLAIM: Can Agents Reproduce the Claims of Machine Learning Papers?  (2609.28850 - Salunkhe et al., 23 Sep 2026) in Section 7, Conclusion, p. 9

Existing efforts remain highly fragmented, with many systems focusing on isolated capabilities, while similar functionalities are often independently developed20-22 and the pathway from workflow automation toward fully autonomous materials discovery remains unclear.

— From Automated Simulation to Autonomous Discovery: A Hierarchical Framework for Agentic Computational Materials Science  (2609.36469 - Zhu et al., 29 Sep 2026) in Section 1, Introduction