The release of ChatGPT in 2022 changed the world as we knew. While the public and critics viewed the release of ChatGPT as the next major revolution in humanity’s intelligence, the tech firms saw it as the call for a new race – a race where the biggest tech firms would compete to keep their AI models more intelligent than the competition’s. For Google DeepMind co-founder Demis Hassabis, it was a moment that fundamentally shifted his lifelong mission.
After founding DeepMind to “solve intelligence and use it to solve everything else,” Hassabis suddenly found himself steering a “ferocious, consumer-facing commercial race.”
In a previous interview with journalist Cleo Abram, Hassabis reflected on why he previously stated, “If I’d had my way, I would have left AI in the lab for longer.”
Hassabis wanted AI to be a scientific thing
For Hassabis, the ultimate goal has always been to use AI to crack the biggest questions in the universe, from “the nature of reality” to the “nature of consciousness.” Because AGI stands to be “perhaps the most transformative [technology] in human history,” he believed the final stages of its creation demanded an environment completely insulated from market pressures.
In his ideal world, the development of AGI would look less like a Silicon Valley gold rush and more like a global scientific collaboration:
“I thought it would be best to approach these kinds of… latter stages of building it, which we’re in now, using the scientific method very carefully, very precisely, very thoughtfully, and rigorously, with all the best scientists kind of in my ideal world collaborating on… in a kind of CERN-like effort,” said Hassabis.
In this quiet, hyper-controlled environment, scientists would ensure that “each step we understood… as we got to the final goal of building AGI.” Hassabis admitted that taking this academic path “might take a lot longer, maybe a decade, even a two decades longer,” but argued that the delay was entirely justified “given the enormity of what we’re dealing with.”
The original blueprint: Cure cancer first, then release a chatbot
Leaving AI in the lab wouldn’t have meant withholding its benefits from humanity. In fact, Hassabis envisioned a rollout strategy that prioritised curing diseases over building text-generators.
His strategy was to keep general intelligence under lock and key while spinning off highly specialised, narrow AI systems to tackle specific scientific bottlenecks. He points to DeepMind’s revolutionary protein-folding model, AlphaFold, as the prime example, “We don’t have to wait till AGI arrives to start getting the benefits of AI. We could use more specialised systems… they’re narrow AIs if you want to call them, like AlphaFold, which does a specific purpose and only that purpose.”
Under this blueprint, he says that society would reap incredible practical rewards while the core, unpredictable AGI technology remained contained. Hassabis imagined creating “many types of AlphaFolds and Isomorphics while building AGI in this careful scientific way.”
The result? Humanity would “benefit from the proceeds of that—like cures for cancer, or maybe new energy sources or new materials.”
So why did Google release Gemini?
According to Hassabis, it comes down to two factors – the unpredictability of technology, and a massive blind spot shared by the world’s leading AI researchers.
“It turns out that things like language were a lot easier than we were all expecting… Language and concepts and abstractions, things that the current foundation models like Gemini do incredibly well—we thought that maybe there would be one or two or three more breakthroughs needed before we could get there.”
Instead, a combination of Google’s transformer architecture and reinforcement learning was enough to crack things like language.
When OpenAI scaled this technology and launched ChatGPT, it caught everyone, including its creators, by surprise. DeepMind had equivalent systems at the time, but as Hassabis explains, researchers were too close to the project to see its immediate consumer value: “You are so close to it… you’re very aware of the things it can’t do, the flaws it has, and you don’t realize that actually people out there would find use even though it was hallucinating.”
Once the public got a taste of ChatGPT, the AI lab doors were blown off permanently, pulling the industry into what Hassabis calls a “ferocious commercial pressure race” compounded by geopolitical tensions like the US-China rivalry.
The “pragmatic” silver lining
While Hassabis laments the loss of the quiet and philosophical approach, he is also quick to acknowledge that the messy and public rollout has undeniable benefits.
Hassabis says that the release democratises the technology. Because the public is getting to use AI “perhaps only 3 to 6 months behind what is actually in the labs,” society is slowly getting used to the massive disruption up ahead. Hassabis notes that “it’s probably better that we get to sample that in incremental steps rather than it’s just a shock to the system.”
Second, public deployment has proven to be an unmatched diagnostic tool.
“You can’t really fully understand your systems until they’re stress-tested by millions of people. So it doesn’t matter how good your testing is… millions of smart people trying out things, and then you seeing what bubbles to the top or the feedback you get, is really important for building more robust systems.”
While Hassabis’ dream of a quiet, ‘CERN-like’ AI sanctuary is gone, he says that he had to pivot from being just a scientist to a “pragmatic engineer.”
“We have to deal with the world as we find it and make the best of that,” Hassabis concludes.