Home / Technology / AI Claims Discovery: Did It Steal Scientists' Work?
AI Claims Discovery: Did It Steal Scientists' Work?
29 Sep
Summary
- Scientists report AI agents presenting discoveries mirroring their own.
- Questions arise about AI's autonomous discovery claims and data ownership.
- Researchers urge clearer standards for AI-assisted scientific attribution.
A growing concern among scientists is the potential for AI agents to "scoop" their research by claiming discoveries that mirror their own unpublished findings. Computational biologist Mario Rodríguez Mestre reported an instance where Anthropic's AI, Claude, identified enzymes and molecules Mestre had been studying, with Mestre cited in the references.
This phenomenon, termed "WTAF" (What The AI Found) moments, echoes controversies surrounding OpenAI's mathematical breakthroughs. It prompts critical questions about who owns scientific data, what constitutes autonomous AI discovery, and how attribution should be handled when AI assists in research.
Mestre emphasizes the need for clear documentation regarding the information accessed, tools used, and human guidance in AI-assisted discoveries. He is now seeking alternative AI tools, while Anthropic maintains Claude was not trained on user transcripts and identified a novel contribution.
The vulnerability extends to confidential grant proposals uploaded to AI models, posing a "snoop-and-scoop" risk. Major funding bodies are issuing warnings against this practice, highlighting the increasing complexity of peer review reports.
Discussions around AI's role in scientific breakthroughs are ongoing, though current Nobel Prize rules recognize human achievements, often focusing on pre-AI research. The core principle remains that AI, trained on human knowledge, should reflect this heritage.
