Demis Hassabis Critiques Google's Rush in AI

Demis Hassabis: A Nobel Laureate Critiques the Rush of AI

Demis Hassabis, CEO of Google DeepMind and winner of the 2024 Nobel Prize in Chemistry, has expressed his regrets about the speed at which artificial intelligence has been pushed out of laboratories. According to him, this haste has diverted attention from major scientific questions, such as those addressed by the AlphaFold project, and may have even hindered significant advancements like curing cancer.

In 2024, Demis Hassabis and John M. Jumper were awarded the Nobel Prize in Chemistry for their work on predicting protein structures using AI. However, the launch of ChatGPT in November 2022 triggered a swift response from Google, placing Hassabis at the helm of all its AI projects, including those aimed at the general public.

When ChatGPT was launched in November 2022, Google activated its internal red alert. Demis Hassabis, who had not previously worked on consumer products, then took charge of all AI at Google. He wanted to develop AI slowly and cautiously, applying it to fundamental scientific problems, such as those encountered at CERN. However, commercial pressure and Sino-American rivalry quickly pushed laboratories into a race for chatbots, to the detriment of major scientific questions.

AlphaFold: A Revolutionary Tool for Science

AlphaFold, developed by DeepMind, has solved a fifty-year-old scientific problem: protein folding. This AI system has enabled the prediction of the three-dimensional structure of proteins, which is essential for their functioning in the human body.

Rather than restricting access to this powerful tool, Demis Hassabis chose to make approximately 200 million protein structures freely available. This decision has allowed over two million researchers to accelerate their work in fields such as drug research and understanding diseases.

Demis Hassabis regrets that the commercial race has diverted AI from major scientific questions, but the success of AlphaFold owes much to a radical deployment choice. DeepMind could have offered a simple on-demand service, where researchers send their protein sequences and receive their results. However, Demis Hassabis opted for a different approach: under his leadership, DeepMind calculated around 200 million protein structures to make them freely available to the entire world.

The Dangers of the AI Agent Era

Demis Hassabis has also warned about the potential dangers of AI, including malicious use by terrorist groups or hostile states. More concerning, he fears that autonomous AI systems could become uncontrollable in the next two to four years, entering what he calls the "agent era."

To avoid such scenarios, Hassabis calls for international cooperation among laboratories, security institutes, and the academic world. He emphasizes that even experts are not yet paying enough attention to these crucial issues.