AI can name lab equipment but can't spot hazards: Study
A new study recently published in Nature Computational Science shows that while advanced AI can handle basic stuff—like spotting lab equipment—it struggles when things get tricky, especially with scientific reasoning and putting information together.
Researchers from IIT Delhi and FSU Jena warn this could be risky if AI is left unsupervised in research labs.
MaCBench test shows AI's strengths and weaknesses
The team, led by NM Anoop Krishnan and Kevin Maik Jablonka, created a new test called "MaCBench" to see how well AI performs in chemistry and materials science.
Turns out, AI got a solid 77% right on naming equipment but dropped to just 46% on spotting safety hazards.
The takeaway? These systems are good at pattern-matching but don't really "get" the science yet.
The researchers say better training—and always keeping humans in the loop—are key for safe use of AI in research.