Some AI tools don’t understand biology yet

Artificial Intelligence has made remarkable strides in various fields, including biology, where it has been harnessed to design enzymes capable of breaking down plastics and proteins that can neutralize snake venom. However, amidst the ongoing buzz surrounding AI, there is a tendency to overestimate its capabilities. Many might assume that simply unleashing AI on existing biological data would yield significant insights, potentially reducing the need for traditional experimentation, including animal testing. Yet, biology encompasses far more than merely analyzing protein structures. Recent studies remind us that it is premature to believe that AI can uniformly tackle all facets of biological inquiry. A noteworthy study conducted by researchers Constantin Ahlmann-Eltze, Wolfgang Huber, and Simon Anders from Heidelberg examined multiple AI software tools aimed at predicting gene activity under various cellular conditions. Their findings revealed that these AI systems did not outperform a basic predictive model. This research serves as a crucial reminder of the intricacy of biological systems. Just because AI shows effectiveness in one domain does not guarantee its success across the entire biological spectrum. The researchers noted that several other studies echoed their findings while their work was available on a pre-print server, further underscoring the challenges AI faces in biological research. Their straightforward methodology offers valuable insights into the ongoing dialogue about the role of AI in the life sciences.