A consistent pattern across fields
Looking back across protein structure prediction, drug discovery, and materials science, the same pattern repeats. Each field has a search space too large to explore by brute force, whether it's possible protein foldings, drug-like molecules, or material compositions. AI models trained on existing data learn to recognize which regions of that space are promising, and they propose a much smaller, ranked set of candidates for real-world attention. Experiments, whether that's a lab synthesis, a clinical trial, or a structural determination, still have to confirm whether a given candidate actually works the way the model predicted.
This is worth stating plainly because it's easy to overstate what AI is doing here. AI accelerates the search, it doesn't replace the lab. The value is in dramatically cutting down how much expensive, slow experimental work is needed to find promising candidates, not in eliminating the need for that work altogether.