What if it’s not “just a game”?
In AI, games are more than just fun—they can reconceive what’s possible.
Demis Hassabis, cofounder of DeepMind and 2024 Nobel Laureate, knows this firsthand. In 2016, after AlphaGo—the AI program developed to play the ancient Chinese game of Go—defeated world champion Lee Sedol, Hassabis saw a new horizon for AI. The victory wasn’t just about beating an 18-time international champion, it was a demonstration of AI’s ability to “think” creatively.
“Famously, move 37 of game two in that big challenge match was a move never seen before,” Hassabis told Fortune. “That showed some form of originality and creativity, and the kind of new system-generated ideas that would be really useful in science and medicine. I’d been waiting not only for us to crack Go at the world champion level, but for the system to be able to invent new strategies. So, when AlphaGo did both those things, that’s when I knew we were ready to take on something like protein-folding.”
At Google DeepMind, Hassabis and John M. Jumper led the development of AlphaFold, an AI system that predicts protein structures—a feat scientists had been trying to achieve for decades. This groundbreaking achievement won them the 2024 Nobel Prize in Chemistry. (Hassabis cofounded DeepMind with Mustafa Suleyman and Shane Legg, and it was acquired by Google in 2014.)
AlphaFold’s success would pave the way for Isomorphic Labs, a spinoff focused on AI-driven drug discovery. Isomorphic was founded in 2021, but it wasn’t until yesterday that the company announced its first external funding—a $600 million round, led by Thrive Capital with participation from GV. The company has existing partnerships with pharma giants Novartis and Eli Lilly, which help (among other things) deliver a vital piece in Isomorphic’s pursuit—data.
“You need data for any AI model,” said Hassabis. “I think AlphaFold 2 showed there’s a lot more one can squeeze out of existing data, if you really, really try, and you’re really innovative on the architecture and algorithm side.”
So, Hassabis’s approach to data is multi-dimensional, spanning multiple sources, including public and purchasable datasets, partner-specific data, and internal data garnered from real-world collection. In short, there’s a lot of data out there, and now it’s time to figure out how to best use it.
“We still need better tools to understand biology,” said Krishna Yeshwant, GV managing partner. “We need ways to make our whole system work better… I do think there’s a real role for machine learning in figuring out biology, figuring out targets, figuring out how we’re going to be able to drug them in unique ways.”
To date, Isomorphic has been cautious about disclosing the diseases it’s targeting, though Hassabis offered some color—broadly, oncology and immunology are in the company’s sights. The idea is to start by tackling diseases with challenging chemistry problems, but where we fundamentally understand how the disease works. (Comparatively, Hassabis’s wife studies Alzheimer’s, which he says has too many unknowns to be a target off the bat.)
“Isomorphic is going to take all these processes that happen physically and represent them computationally,” said Vince Hankes, partner at Thrive Capital. “It’ll radically change the entire industry, where you’ll look back and it seems obvious. Imagine building an airplane without CAD, doing all the drawings and simulation by hand. We’d never conceive of doing that today, nor would you want to fly in an airplane built by hand versus built by a computer. It’ll be the same thing with drug discovery.”
The history of medicine is gnarly, filled with practices we can no longer fathom, let alone imagine being standard. Think: lobotomies, bloodletting, and rotational therapy—the Victorian belief that if you twirl a patient vigorously in a suspended chair, it could cure mental illness. For Hassabis, success for Isomorphic is nothing short of curing all diseases, creating a new normal.
“To be clear, modern medicine has done incredible things, but we need a lot more of it,” said Hassabis. “Before penicillin and antibiotics, people could die from a scratch, right? They’d get infected and that was it. Child mortality was massive, because children could die from earaches. Now, we take it for granted. We give them some antibiotics, and they’re fine, without a second thought… In another 10 to 20 years, I want to look back at heart attacks or brain aneurysms and say: ‘Wow, it’s crazy people used to die from that, we can detect it now. Or, with cancer, we can detect it early and deal with it.’ It’ll be like taking an antibiotic, where it’ll be so matter-of-fact. It won’t seem extraordinary anymore.”
Isomorphic’s name is drawn from Ancient Greek, a scientific term that references disparate things with the same structure. Hassabis’s ambition is grand, but it’s also a reminder that something seemingly small can spark seismic change—just like a series of Go games have a throughline to a possible revolution in drug discovery.
The largest funding round ever…Yesterday, OpenAI announced it raised $40 billion at a $300 billion valuation. According to CNBC, $30 billion came from round leader SoftBank, while $10 billion came from a group of investors that included Microsoft, Coatue, Altimeter, and Thrive. (Notably, Hankes is Thrive’s OpenAI investor, along with firm founder Josh Kushner.) OpenAI’s valuation is now on par with ByteDance, and is only surpassed in valuation by SpaceX.
See you tomorrow,
Allie GarfinkleX: @agarfinksEmail: [email protected] a deal for the Term Sheet newsletter here.
Nina Ajemian curated the deals section of today’s newsletter. Subscribe here.
This story was originally featured on Fortune.com
