"AI will solve all diseases."
These words came from Google DeepMind CEO Demis Hassabis, on the main stage at Google I/O.
The Verge Victoria Song wrote a pushback piece with a blunt title: "Solve all diseases, you say? Not so fast."
I agree with Victoria judgment, but I want to approach this from a different angle: why this kind of talk does more harm to the industry than to the public.
This is not the first time, and it will not be the last
The AI "cure all diseases" narrative has been cycling for at least five years:
- 2021: AlphaFold solves protein folding, media says "drug discovery is about to be revolutionized"
- 2023: AI discovers new antibiotics, media says "superbugs are saved"
- 2025: AI-assisted clinical trial design, media says "new drug development cycle will shrink by 80%"
- 2026: Demis Hassabis says "AI will solve all diseases"
Every time it is partly true plus a lot of exaggeration. Protein folding was indeed solved, but there are more than ten gaps between that and "curing all diseases."
The problem is not AI capability, it is the scale of the narrative
DeepMind has genuinely done remarkable work in AI + healthcare. AlphaFold is real, AlphaMissense is real, their progress in drug discovery is real.
But the problem with "solve all diseases" is not that AI cannot do it — maybe in 50 years we will look back and AI did massively accelerate medical progress.
The problem is: when you say "solve all diseases," you are redefining what "solve" and "all" mean.
- The causes of Alzheimer disease are still not fully understood
- The heterogeneity of most cancers means "curing cancer" is actually "curing dozens of different diseases"
- Rare diseases have too few patients, commercial incentives are not enough to drive R&D
- Healthcare system accessibility — even if there is a drug, how many people can access it?
These are not technical problems. They are biology, economics, sociology problems. AI cannot solve these problems, at least not by "training a bigger model."
Misleading entrepreneurs and investors
The most harmful aspect of this talk is that it creates unrealistic expectations for entrepreneurs and investors.
If you are an AI healthcare startup founder, hearing DeepMind CEO say "AI will solve all diseases" — you will feel your赛道 has infinite potential. Then you discover:
- FDA approval process does not speed up just because you used AI
- Phase III failure rate does not decrease just because your model has more parameters
- Hospital procurement cycles are so long you question your life choices
Then your burn rate runs out, your company shuts down. You did not fail because AI does not work. You failed because you were misled by the narrative.
I respect DeepMind work
Let me say it again: DeepMind team is doing genuinely important research. AlphaFold changed the rules of structural biology. No question about it.
But their executives public statements are increasingly sounding like the tech industry common overpromising. The side effect of this kind of talk is:
It detaches public expectations of AI healthcare from scientific reality, crushing researchers who are doing the hard work under unrealistic expectations.
My advice
Next time you hear "AI will cure X," ask three questions:
- Which specific X are they talking about? "All diseases" is not specific. "Alzheimer disease" is specific.
- How long from lab to clinic? Usually 10-15 years.
- Are they doing scientific research or giving a fundraising pitch? The meaning of "cure" in these two contexts is completely different.
AI will transform medicine. I have no doubt about this.
But "solve all diseases" is not a scientific prediction. It is a PR slogan.
The distinction matters.
Main sources: