AI companies and researchers are increasingly turning to large genomic models—such as Evo 2, Genos, and Google DeepMind's AlphaGenome—to make sense of the human genome. These models are trained on vast amounts of DNA sequence data and are used to predict how genetic differences affect biology and disease risk. But as a recent analysis in Quanta Magazine, via WIRED, points out, the genome is not a tidy blueprint. Only about 2 percent of it codes for genes, and the regulatory systems governing how those genes are used are deeply intricate.
The source explains that gene regulation in humans involves overlapping controls, transcription factors that work in groups and change behavior by context, and logic that is far messier than the simple on-off switches seen in bacteria. AI models may capture correlations in this complexity, but they operate as black boxes. For businesses building on genomic AI, that means a tool that can flag patterns may still not explain why those patterns matter—or reliably predict outcomes in new contexts.
The article quotes biologist Karen Adelman of Harvard Medical School saying the genome is not what we would design at a drawing board. That sentiment carries a warning for the growing genomics industry: if the genome's logic is not algorithmic in the way AI assumes, then even powerful models may hit a ceiling. The source does not dismiss AI's usefulness, but it suggests that genuine understanding—and perhaps reliable clinical translation—will require more than scaling up computation.