At this year’s Google I/O keynote, Demis Hassabis, CEO of Google DeepMind, stood on stage and declared with a completely deadpan expression that the company hopes to “reimagine the drug discovery process with the goal of one day solving all disease.” It's the kind of statement that stops a room cold—and it was almost immediately met with a mix of awe, skepticism, and head-scratching.
Hassabis wasn't claiming that a single AI model would magically wipe out cancer, Alzheimer’s, or the common cold. What he was actually describing was “Gemini for Science,” a collection of experimental AI tools designed to help researchers accelerate the pace of discovery. But nuance rarely travels well in a keynote that also announces forty bajillion new AI agents and features. And for the average person, the line likely sounded like “Gemini is going to cure everything because that’s the power of AI.” That’s not how medical breakthroughs work.
What Hassabis really meant
To understand the weight of the statement, we need to look at the tools Hassabis mentioned: AlphaFold and AlphaGenome. AlphaFold, first unveiled in 2020, is a deep learning model that predicts protein structures with remarkable accuracy. Proteins are the molecular machines of life, and understanding their 3D shapes is key to unlocking how diseases happen—and how to stop them. Before AlphaFold, determining a single protein’s structure could take years of lab work. Now, it can be done in minutes. Researchers have already used AlphaFold to help develop malaria vaccines, identify a key protein behind “bad” LDL cholesterol, and understand a protein linked to early-onset Parkinson’s disease.
AlphaGenome, meanwhile, is a model that predicts mutations in human DNA sequences. It could help scientists understand why certain genetic diseases occur. However, Google has noted important limitations: the model hasn’t been validated for personal genome prediction and struggles with cell- and tissue-specific patterns. These are critical nuances for researchers but often get lost in translation.
The long road from lab to bedside
In many respects, Hassabis wasn’t speaking to you or me. He was speaking to fellow researchers and investors who understand that these AI models are tools, not miracle cures. Even with the most advanced AI, the path from a protein structure to an approved drug is long, expensive, and fraught with failure. A typical drug takes 10–15 years and billions of dollars to go from discovery to pharmacy. AI can compress some steps, but it cannot replace clinical trials, safety checks, or regulatory oversight. Most experts estimate that if AI truly accelerates medicine, we are looking at 20 years or more before seeing a transformed landscape.
For someone with a loved one battling a serious illness today, that timeline feels glacial. But in the world of rigorous scientific research, 20 years is aggressive. And it assumes that political, economic, and social factors don’t derail progress—which is far from guaranteed.
The danger of soundbite science
This brings us to the larger issue: science communication in the age of short-form video, declining media literacy, and wellness grifters. A bold statement like “solve all disease” travels wide and fast. It can easily be misinterpreted, and worse, co-opted by bad actors. For example, Health Secretary RFK Jr. recently told Congress that AI might make the FDA “irrelevant” because it could develop and approve drugs faster. That’s a radical oversimplification. Yes, AI can help with analysis, but it doesn’t eliminate the need for animal testing, clinical trials, or the careful balancing of risks and benefits that the FDA oversees.
When Hassabis’s comment is heard alongside such political rhetoric, average listeners might falsely assume that Google agrees with Kennedy’s view. That couldn’t be further from the truth. The Verge has previously reported why Kennedy’s AI-in-health takes are flawed, but the damage from misleading associations is already done.
From science to sciencewashing
There’s a reason why “sciencewashing” is everywhere. A few buzzwords – AI, longevity, optimization – lend a high-tech sheen to otherwise dubious claims. The same set of tools that might help a researcher discover a new protein can be used to sell you a fancy supplement or a biometric tracker that promises to “defeat death.” The gap between what AI can do for a lab and what it can do for your health is huge, but it’s rarely emphasized.
For the majority of consumers, AI health has been a craptacular experience: regurgitated metric summaries, hallucinated advice, and tedious hand-holding. We shouldn’t conflate these tools with the powerful models used in research, but that’s exactly what happens when a CEO gives a bold one-liner and the rest of us fill in the gaps with our hopes and fears.
Maybe AI will one day help solve all diseases. But if it does, the path will not be simple, fast, or without political and societal hurdles. Forgive me if, for now, I keep my optimism tempered.
Source: The Verge News