The narrative is everywhere now. Artificial intelligence has "solved" protein folding. AlphaFold and its successors will revolutionize drug discovery, cure diseases faster, and render traditional structural biology nearly obsolete. The story is seductive because it contains real achievement wrapped in genuine technological progress. But the way this breakthrough is being sold to the public and policymakers deserves considerably more pushback than it's receiving.
Start with what's actually true. DeepMind's AlphaFold represents a legitimate advance in predicting three-dimensional protein structures from amino acid sequences. That's genuinely useful. Knowing a protein's shape matters for understanding how it functions and how drugs might interact with it. The achievement warranted recognition. But somewhere between genuine progress and transformative science, the narrative got away from itself.
The problem begins with prediction versus discovery. AlphaFold excels at predicting structures for proteins that share similarity with ones already in scientific databases. It's pattern-matching at superhuman speed. What it does not do is tell us what proteins do in living systems, why evolution selected them, or how they behave inside actual cells. A predicted structure is a starting point, not a destination. This distinction matters enormously for drug development, yet it's often glossed over in popular accounts.
Consider the funding implications. As AI protein prediction gets treated as inevitable progress, funding pressures mount on traditional structural biology. Universities and institutes face pressure to chase the AI narrative. We see this pattern repeatedly in science: emerging technologies get hyped, funding shifts, and entire research ecosystems reorganize around the new story before we actually know what we're betting on.
The hype also creates perverse incentives. If AI protein tools are presented as complete solutions, researchers have less motivation to ask hard experimental questions. Why spend years doing wet-lab validation when the algorithm has already predicted the answer? Except that prediction and reality often diverge in biology. Proteins fold differently in different cellular environments. They interact with other molecules in ways that pure structure prediction cannot capture. These complications aren't flaws in the research; they're reflections of how messy biology actually is.
There's also a deeper epistemological concern. Science progresses through cycles of surprise and explanation. When we develop powerful predictive tools, we can inadvertently lock in our existing frameworks and assumptions. AlphaFold was trained on existing structural data, which means it can extrapolate brilliantly from what we already know but may systematically miss novel biology. We don't yet have a clear picture of what kinds of biological truths might be invisible to this approach.
None of this means AlphaFold is a failure or that AI tools in biology are worthless. They're clearly useful. But "useful" and "revolutionary" are different categories. The revolution narrative serves powerful interests: venture capitalists seeking returns, technology companies expanding their reach into biology, and the simple human desire for stories of inevitable progress.
The skepticism I'm advocating isn't anti-science or anti-technology. It's pro-reality. It means asking harder questions before reorganizing entire research fields around emerging tools. It means funding traditional structural biology alongside AI approaches, not treating them as competitors. It means being honest that we don't yet know the full implications of these advances.
We've learned this lesson before. Not every transformative-seeming technology transforms immediately or in the ways predicted. Some do. Some don't. The way to find out isn't to accept the inevitability narrative. It's to maintain genuine, skeptical curiosity while the technology matures.
That's how science actually works best.