DeepMind wants to reinvent drug discovery and defeat disease

At Google I/O 2026, DeepMind CEO Demis Hassabis announced with a completely straight face that the company hopes to "reinvent the drug discovery process with the goal of one day solving all disease." It's an ambitious goal, and frankly, it sounds unrealistic. But before dismissing it, let's look at what's behind it and why journalists at the Verge are already holding Google to its word.

What Google showed at I/O

Google DeepMind unveiled a full package of tools to speed up biological research. The main three are Gemini for Science (a specialized version of its flagship LLM for scientific work), AlphaFold 3 (a protein structure prediction system), and the brand-new AlphaGenome.

AlphaFold 3 can predict the three-dimensional structure of proteins and their interactions with other molecules with high accuracy. That matters because a protein's shape often determines whether a drug will work.

AlphaGenome extends the approach: the system can now work with genomic data and predict the genetic factors behind diseases.

Gemini for Science integrates both systems into a single platform. It's an LLM that "understands" the context of biological research, can read scientific papers, and offers recommendations directly inside an interface scientists already know.

How it works in practice

Here's a rough scenario: a researcher asks Gemini for Science something like "design an antiviral compound for virus X." The system analyzes what's known about the virus, proposes candidate molecular structures, uses AlphaFold 3 to predict how those molecules interact with viral proteins, and forecasts likely side effects.

On paper, that sounds like a leap in productivity. A search like this used to take months. Now it may take hours or days.

But there's a catch: Google talks about "acceleration" but doesn't give concrete timelines. How many percent faster? At what cost? That stays off-screen.

Why the skeptics outnumber the believers

Here's why "solving all disease" is, to put it mildly, a very bold claim:

• Going from molecule to drug is not a single task. Even if AI predicts a molecule perfectly, it still has to be synthesized in a lab, tested on cell cultures, then on animals, then on humans (clinical trials: 3 phases, 5-10 years), and only then approved by a regulator.
• "All disease" isn't 100 conditions, it's 10,000+. Some are genetic in nature (within AI's reach), but others are infectious, autoimmune, psychiatric, or environmental. Molecular biology AI won't help much with those.
• AI history is full of promises that didn't pan out. Deep Blue promised a revolution in analytics. IBM Watson promised a revolution in medicine. Reality is usually more complicated than marketing slides.
• Regulators slow the process down on purpose. The FDA requires proof of safety. That isn't a whim; it protects patients. An AI suggestion is only the first step on a very long road.

"The distance between big promises and results in a real clinic is enormous," the

Verge's editors write, and they're not alone in that view.

What this actually means

Google DeepMind's tools are genuinely useful and can speed up individual phases of drug development, especially the candidate search stage. But "solving all disease" isn't a technology problem. It's a question of how complex biology is, how strict regulators' requirements are, and how much money gets invested in each area.

AI will be a good assistant to scientists. But it won't be a wizard. Hassabis's promises are probably just a Google I/O speech. A useful speech, but it's worth staying realistic.