What if AI goes well?
On Friday, I walked you through how AI could kill us, and I promised you something more positive this week.
So here's the other side: what life could feel like if this goes right, and how we'd get there.
- What would a good future actually feel like?
Start with today. Someone in your life finds a lump, and that can mean scans, waiting for answers, and lying awake at 3 a.m. doing math about the odds.
Now imagine that same day if AI goes well. The scan gets read that afternoon. The doctor says, "We have a treatment made for exactly this," and talks about it the way you'd talk about a broken arm.
And that night, you'd actually sleep, knowing it was going to be okay.
- How would AI get us there?
Finding a treatment means testing many ideas, and only a few make it through and actually become something.
Here's my guess at the good version. Picture millions of AIs, each as good as the best expert in the field, working on one disease around the clock. They come up with tens of thousands of ideas no one has tried before, split them up, check each other's work, and drop the dead ends quickly. That doesn't exist yet... but the first signs of it are already here.
- Isn't that a stretch?
Look at how fast this has moved. In 2020, the leading AI language models (the kind behind ChatGPT) often got basic arithmetic wrong .
This year, an AI solved a math problem that had stumped the world's leading mathematicians for eighty years . One of the top mathematicians alive said that if a human had sent that work to a leading math journal, he'd have recommended accepting it . And it wasn't a one-off. Two weeks ago, OpenAI said its AI had produced a proof for another famous problem, one about how fluids move that had gone unanswered for roughly ninety years.
Now think about that same intelligence, scaled up to an extreme degree, and instead of pointing it at math, we point it at cancer.
This month, Dario Amodei, who runs the AI company Anthropic, predicted that AI could help cure most major diseases in the next five to ten years .
- Has any of it reached medicine yet?
Yes, in small ways. A drug for a deadly lung disease was designed with AI, and it showed promising results in a small early trial in patients .
And in a Swedish trial of more than 100,000 women , AI-assisted screening caught a higher share of cancers, with the same false-alarm rate as the usual screening by two radiologists.
Those are just two examples of AI starting to help in medicine, and I think many more are coming.
- Then why isn't cancer cured already?
Because a promising idea still has to work in real patients, and testing a new drug for safety and benefit takes quite a while.
What I expect AI to change is everything before that. Part of it is skipping the years spent on ideas that were never going to work. The bigger part, I think, is coming up with promising ideas we haven't thought of.
Ten years ago, an AI playing the board game Go against Lee Sedol, one of the world's best players, made a move that expert commentators thought was a mistake . It turned out to be part of a winning strategy. The move helped the AI win the game, and players still study it today. Now imagine that kind of inventiveness aimed at cancer research.
- Is this just about medicine?
Not even close. I expect the same kind of AI to fundamentally reshape how we educate kids.
It's more than a patient tutor for every kid. Everyone learns differently. Now imagine the best teacher in the world paired with one kid, figuring out exactly how that kid learns, and teaching them that way. I've helped build the AI systems at Alpha School, and I've seen kids make remarkable progress.
Early versions of this already exist elsewhere too. In Nigeria, six weeks of after-school AI tutoring , with teachers involved, produced gains researchers estimated were worth well over a year of normal schooling there.
- Won't only rich people get it?
This is exciting to me, because not every technology makes its way to everyone. In this case, I think it will.
Imagine paying $100 for help, then getting just as capable a helper for 35 cents. That gives you a feel for what's happened with AI. Between late 2022 and late 2024, the price of using AI at roughly the original ChatGPT's level fell by more than 280 times .
So say a few years from now there's an AI that can make real progress on one type of cancer, and at first only a few people can afford to point it at the disease they care about. A couple of years later, most people could afford it too, and a couple of years after that it could be close to free. Imagine the cost of pointing that kind of AI at any disease being nearly free. That's my hope, and I think it's a realistic one.
- What about someone who's sick right now?
Someone who's sick today may not have years to wait. But we're already seeing what this can mean for someone's life.
A man with ALS, whose speech had become hard for even his wife to understand, is talking with his wife again . A brain implant uses AI to turn his brain signals into words, and they come out of a speaker in his own voice. By the end of the video, he's saying "I love you." It's one person, it's still early, and it's worth watching.
- So what's the catch?
Amodei, the same Anthropic CEO who predicted the cures, wrote in that essay that the industry needs to slow down how fast it makes its most powerful AI smarter. Built the wrong way, it's dangerous (that's what Friday's email was about).
If we all die, none of this matters. If it takes a few extra years and we're all still here to live in it, that's obviously the better deal.
I think a world where someone you love finds a lump, and you still sleep that night, is really on the table. I want us all around to see it.
What questions do you still have? Hit reply and tell me. I read every one, and reply to as many as I can.
One more thing: you may have seen that Anthropic released Opus 5.5 , its new flagship model, yesterday. I've been testing it, and I'll have a review for you soon.
— Matt