AI Won a Nobel Prize. It Still Can't Run Your Lab.

AlphaFold just won the Nobel Prize in Chemistry, and it deserved it. But if you walked into a biology lab tomorrow and told the postdocs they were about to be managed by an algorithm, they’d hand you a mop and ask you to clean the cell culture hood first. We keep confusing a breathtaking tool with a turnkey replacement for the people wielding it.

Headlines will tell you that artificial intelligence has “solved” biology. Vendors are already rewriting their sales decks to suggest that scientific discovery is now a software problem. And if you only skim the news, I can’t blame you for believing it. The misconception is simple but costly: predicting what a protein looks like—the thing AlphaFold does brilliantly—is the same as doing science. It isn’t.

Science is mostly uncertainty, failure, and judgment calls made with incomplete data. AlphaFold gives you a static map. Running a lab is navigating a jungle with that map while your funding, equipment, and graduate students’ morale are all running low at the same time.

Let’s look at what actually happened. In 2024, the Nobel Committee awarded half the Chemistry prize to Demis Hassabis and John Jumper for AlphaFold, and half to David Baker for computational protein design [1]. AlphaFold can predict the three-dimensional shape of a protein from its amino-acid sequence—the order of its molecular building blocks—with staggering accuracy, cracking a problem researchers had chipped away at for half a century [2]. That is not a parlor trick. It changes how pharmaceutical companies screen drug targets and how biologists form hypotheses.

But here is what it does not do. It does not design the experiment to test whether that predicted structure matters inside a living mouse. It does not notice that your reagent shipment sat on a loading dock in July and is now useless. It does not sit with a grieving graduate student who just realized six months of data are contaminated. It does not decide, at two in the morning, whether an anomalous result is noise or the next breakthrough.

In my book, More Than Parrots, Less Than Gods, I describe AI as occupying a precise middle ground: far more than a mimic, but far less than an autonomous agent with judgment, context, and care. AlphaFold is the perfect example. It is a cognitive telescope. It lets us see protein landscapes we could never see before. But you still need the astronomer to know where to point it, and what the image means when it comes back blurry.

This pattern holds everywhere AI touches. The tool handles a narrow, structured task with superhuman speed. The surrounding work—coordination, ethics, interpretation, maintenance, office politics—remains stubbornly human. I see the same gap in business teams that expect a large language model to “do strategy.” It can draft the slide, but it cannot tell you if the slide is steering your company off a cliff. It has no nose for context.

So what do you do with a breakthrough like this? Three thoughts.

1. Find the 5 percent, protect the 95 percent.
AlphaFold likely handles a narrow slice of your workflow—structure prediction, literature synthesis, maybe code generation. Map where your team’s actual hours go. Automate that narrow slice ruthlessly, but pour your best human energy into the ambiguous majority that surrounds it.

2. Never confuse prediction with proof.
AlphaFold predicts shapes; it does not validate biological function. In any field, when an AI hands you an answer, build a feedback loop that tests it against physical reality—whether that means a wet-lab experiment, a customer interview, or an audited set of books.

3. Buy tools, not miracles.
Vendors will soon sell “AI-powered research platforms” that promise to run your lab. Ask them the boring questions: Who retrains the model when my protocols change? How does it handle a broken centrifuge or a contaminated cell line? What happens when funding priorities shift? If they cannot answer, you are buying the fantasy, not the product.

AlphaFold changed what we can see. What it has not changed is what it means to look, to doubt, and to decide under uncertainty. So here is my question for you: What is the one messy, unglamorous part of your work—the part no headline ever celebrates—that you suspect AI will be the last to figure out? I genuinely want to know.

[1] “The Nobel Prize in Chemistry 2024,” NobelPrize.org, Nobel Prize Outreach AB 2024.
[2] J. Jumper et al., “Highly accurate protein structure prediction with AlphaFold,” Nature 596 (2021): 583–589.