Mark Zuckerberg's Biohub is teaming up with Google and the federal government to build an AI that can predict how living cells behave. According to Axios, the project has a small problem: researchers don't yet have enough of the data it needs to work.
The Pitch: Test Experiments Without Touching a Petri Dish
Axios reports that the ultimate goal is an AI model that lets scientists test potential experiments virtually. The idea is to identify the most promising experiments before anyone spends the time and money to run them in a lab.
If that works, it would be a genuine shift in how biology gets done. Lab work is slow, expensive, and full of dead ends. A tool that screens out the losers before the pipettes come out is the kind of thing researchers have wanted for decades.
Proteins Are Hard. Whole Cells Are Much Worse.
Here's the part that should temper the hype. Axios notes that AI is already capable of understanding proteins and other pieces of biology. But modeling an entire living cell is, in the outlet's words, orders of magnitude more complex.
Think of it as the difference between understanding a single gear and simulating the whole watch while the watch is alive, self-repairing, and constantly changing. Getting the pieces right is not the same as getting the system right.
The Data Doesn't Exist Yet
The sentence that matters most in the Axios report is the one that gets cut off: researchers don't yet have enough of the raw material they need. That is the whole reason for the partnership. The plan, per Axios, is to use AI to generate vast quantities of biological data that can predict how cells behave.
So the model needs data that doesn't exist, and the plan is to produce that data with the help of AI. Maybe that works beautifully. It also means the project is building the foundation and the house at the same time, which is a sentence nobody in construction has ever said with a straight face.
Who Is Actually in the Room
Axios says Biohub is partnering with Google and the federal government. That is a lot of institutional weight behind a Zuckerberg-backed effort, and the excerpt available to us does not spell out what each partner is contributing, how much money is involved, or which federal agencies are participating.
Those details matter. Who funds this, who owns the data it generates, and who gets to use the finished model are not footnotes. We will have more once the full reporting is on the table, and you should be asking the same questions.
The Dingo Take
Everyone involved is describing a tool that will let scientists skip failed experiments and speed up discovery. What the reporting actually shows is a project whose central input does not exist yet. Researchers lack enough data to model a living cell, so the first job is manufacturing that data at enormous scale. That is a research program with a promise attached, not a product.
That doesn't make it a bad idea. Virtual experiments could save real time and real money, and if the cell model ever works, the payoff for medicine is hard to overstate. But the gap between "AI understands proteins" and "AI simulates a living cell" is enormous, and the people selling the second thing are leaning on the credibility of the first.
So watch the unglamorous questions. Who owns the data? Who gets access to the model? What does the federal government get for its involvement? A billionaire's foundation, one of the biggest tech companies on earth, and Washington are sitting at the same table, and the public should know exactly what it's being asked to bless.
Comments