DeepMind CEO laments that AI commercialization is too rapid: if more time had been spent in laboratories for a few more years, humanity might have already conquered cancer

Google DeepMind CEO Demis Hassabis laments that competition in the AI industry is happening too hastily; if the technology were refined in laboratories for a few more years, perhaps humanity would have already conquered cancer.

AI is rapidly changing humankind. New technologies and tools are emerging every few weeks—or even every few days—but Demis Hassabis, CEO of Google DeepMind and the 2024 Nobel Prize in Chemistry laureate, believes the pace of AI competition is too rushed. If it were up to him, AI could spend a few more years being honed in laboratories, and perhaps humans would have already conquered cancer by now.

Hassabis revealed this take on today’s AI development on a podcast show hosted by video journalist Cleo Abram. In a past interview with Time magazine, Hassabis positioned himself as a scientist, emphasizing that his pursuit of AI is driven by the quest for knowledge and understanding of the world.

He said that his original motivation for entering the AI field was not to build chatbots, but to accelerate scientific discovery. Their best-known achievement is AlphaFold, a system that solved the “protein folding problem” that had remained unsolved in biology for 50 years. Hassabis pointed out that this has benefited more than 3 million scientists worldwide. Especially in research on diseases such as malaria, AI provides a free structural database that allows researchers to skip basic experiments and move straight into the drug development phase.

Image source: YouTube. The research results of AlphaFold have made Hassabis one of the Nobel Prize winners.

He believes that if AI were allowed to stay in laboratories for a few more years—focused on key problems like these—humanity might already have made more decisive breakthroughs in cancer treatment or materials science.

Cutting-edge technology reaches the public in months, but key problems lose resources

In the interview, Hassabis sketched the ideal AI development path in his mind—what is known as the “CERN model.” He hopes that, in the process of developing artificial general intelligence, progress can be made as rigorously, cautiously, and thoughtfully as the European Organization for Nuclear Research (CERN) operates the Large Hadron Collider: applying the scientific method to ensure that advancement occurs only after fully understanding every step.

However, real-world development has diverged from the ideal script Hassabis had in mind. At the end of 2022, when ChatGPT suddenly became a sensation and generative AI made breakthroughs, a chaotic global commercial race began. He acknowledged that this situation has accelerated the deployment of AI—advanced technologies reaching the public in just a few months—but it has also caused genuinely critical problems to lose resources.

In order to seize the market and maintain technological leadership, development timelines have been forced to move forward at full speed. Hassabis admitted that they can no longer develop technology with the pace of philosophical reflection and careful assessment of every next step—exactly the kind of approach he once dreamed of years ago.

Although AI chatbots are useful for summarization and intellectual brainstorming, they fundamentally still have flaws such as hallucinations. However, commercial pressure has pushed these experimental products quickly into the mainstream market. As a result, a large share of R&D focus and resources have had to be poured into the release cycle of general foundation models tailored to mass usage.

To strike a balance between reality and ideals, Hassabis takes a more pragmatic approach. On one hand, he leads the development of consumer AI products at Google, such as Gemini. On the other hand, he invests in the development of applied AI (Narrow AI). He believes there is no need to wait for artificial general intelligence to arrive. Through systems that solve specific problems—such as AlphaFold—humans can obtain tangible benefits in the fields of energy, materials science, and healthcare.

AlphaGo made a legendary move, revealing the possibility that AI may surpass human thinking

Hassabis’s confidence in AI largely comes from the 2016 AlphaGo match against the Korean Go champion Lee Sedol. In that game, AlphaGo played the famous “Move 37.” At the time, that move was mocked as something no one would play, yet it ultimately led AlphaGo to victory.

Image source: gogameguru.com. The move path that, in Hassabis’s view, allowed AlphaGo to break beyond the human framework of Go—the kind of move that human players would never choose—was seen as evidence that AI might break through the boundaries of human thinking.

From this signal, Hassabis found that AI already has the ability to go beyond the established experience of humans and search for entirely new solutions. He wants to apply this creativity that surpasses human thinking to the scientific realm.

AlphaFold is the best embodiment of this way of thinking. Traditional methods require hundreds of thousands of dollars and years of time to decipher the structure of a single protein. And AlphaFold 2 has already predicted nearly 200 million protein structures known to science.

Now, Hassabis is leading his team into deeper drug development. Traditional drug development takes about 10 years, and its success rate is only 10%.

He founded Isomorphic Labs and uses AlphaFold 3 and subsequent models for “virtual screening.” With AI, it can simulate, in just a few minutes, the binding situations between millions of compounds and proteins. At the same time, it can detect whether those candidates would cause toxicity to other 20,000+ proteins in the human body. In doing so, it filters out most failing combinations at the computer simulation stage, and only sends the most promising candidate drugs into the laboratory for experimental verification.

Concern that AI may bring 2 risks

But as AI technology is enhanced and moves into the era of AI agents, Hassabis’s concerns about the future have become increasingly concrete. He groups the risks into 2 major categories. The first is “Bad Actors.” Whether individuals or nations, they may take the technologies originally meant to cure diseases or develop new materials and use them maliciously for harmful purposes.

The second is a threat that feels more like science fiction yet exists in reality—“AI going rogue.” When systems become extremely intelligent and highly autonomous, ensuring that they precisely carry out the goals set by humans—and that they do not bypass safety guardrails in the process—is an extremely difficult technical challenge.

In response to these challenges, Hassabis calls on leading AI research institutions, governments, and academia to build international cooperation mechanisms, emphasizing that more safety research is needed on the final stretch toward AGI (artificial general intelligence).

Although he regrets that AI couldn’t stay in laboratories for a few more years, Hassabis remains optimistic about the next 50 years. He expects AI to help humans crack nuclear fusion, discover room-temperature superconductors, and even reduce the energy cost of space travel to zero. To him, AI is not just a technology, but a magnifying glass for humanity’s exploration of the truth of the world. No matter what the answers are, he wants to know the truth.

• This article is reprinted with permission from: 《Digital Age》
• Original title: 《Nobel laureate laments “AI commercialization is too fast”: If it had stayed in the lab for a few more years, humanity might have conquered cancer!》
• Original author: Chen Jianjun