The National Institutes of Health (NIH) has developed an artificial intelligence tool to analyze gene data sets that the agency says is highly accurate and does not carry the risk of hallucinations.
The system dubbed “GeneAgent” was developed by NIH researchers and checks its generated predictions against third-party expert-curated databases while returning a verification report detailing its successes and failures when interpreting molecular data.
“LLMs [large language models] are not designed to verify truth, meaning AI-generated content can be false, misleading, or fabricated, a phenomenon called AI hallucinations,” said NIH. The agency also noted that LLMs are also prone to circular reasoning and fact-checking generated results against their own data.
“GeneAgent mitigates this issue by taking its own claims and independently comparing them to established knowledge compiled in external, expert-curated databases,” said NIH.
The agency explained that the AI agent analyzed over 1,100 gene sets and compared its predictions against trusted expert databases. It then produced verification reports marking whether each claim was supported, partially supported, or refuted – resulting in 92 percent of its self-checks to be accurate.
“While LLMs such as GeneAgent are still limited by the information they can use and their inability to reason as humans, GeneAgent’s ability for self-driven fact-checking shows remarkable promise in mitigating AI hallucinations,” said NIH.
Researchers added that additional testing on real-world mouse gene sets provided new insights into gene functions, meaning that the agent could eventually be used “for things such as potential new drug targets for diseases like cancer.”
The agent will primarily be used to help researchers “interpret high-throughput molecular data” while identifying “relevant biological pathways or functional modules,” with aims of having a “better understanding of how different diseases and conditions affect groups of genes individually and together.”