Google’s Demis Hassabis tasked with turning AI research into profits

For an academic credited with some of the most important advances in AI over the past decade, Hassabis is extremely clear about the task ahead: bringing the latest AI technologies to every corner of the world. the world of Google to serve its billions of users.

“We’re like the engine room of the company,” Hassabis told CNBC, speaking about his new integrated AI unit within Google.

Last month, Alphabet CEO Sundar Pichai merged Hassabis’ DeepMind with Google Brain, a separate AI team, and chose Hassabis to lead the group. It’s now up to Hassabis to reestablish Google as a leader in generative AI after the company was caught off guard by the rapid emergence of Microsoft-backed OpenAI.

There may be no more important task at Google, especially as new generative AI services offer consumers alternative, more creative ways to search for information online. The business question is: Can a long-time researcher like Hassabis be the right person to ship products that consumers love?

Geoffrey Hinton, known as the “Godfather of AI,” says Hassabis’s will to win is beyond doubt.

“I don’t think I’ve ever met anyone more competitive,” said Hinton, who advised Google to buy DeepMind about a decade ago. “Demis is competitive at the level of people who win gold medals in the Olympics.”

He developed this trait from a young age. Hassabis was a child chess prodigy who, at one point, was the No. 2 player in the world. He also participated in the World Series of Poker.

Hassabis says he also has consumer experience on his side. At just 17 years old, he released a hit video game in the 1990s called “Theme Park.” At that time, games had to be fun and easy to navigate to be successful, Hassabis said. After Theme Park, he created his own video game company, Elixir Studios.

Hassabis went on to co-found DeepMind, which became widely recognized as the world’s leading AI research lab, attracting some of the most prominent experts in deep learning. When Google acquired the lab for $500 million in 2014, DeepMind was given a long leash to operate independently.

Under Hassabis, DeepMind was known for developing its technology through games like Breakout and AlphaGo, an AI program that beat the world’s best Go player.

There was a practical reason to focus on games. Hassabis told CNBC that “the simulations are completely safe, have no consequences, but they can still learn from them.”

During his career at DeepMind and then Google, Hassabis dominated the field of AI.

Even Tesla CEO Elon Musk reportedly told OpenAI co-founders in 2018 that they would need billions of dollars to have a chance of competing with Hassabis and Google.

Google, however, would lose this advantage as competitors like OpenAI and Microsoft brought headline-grabbing products like ChatGPT and Copilot to market.

A person who worked with Hassabis but wished to remain anonymous said that, for a time, Hassabis may have been more interested in winning academic honors than in releasing products that people could use. Nature magazine, one of the most influential scientific journals in the world, has featured Hassabis’ work several times over the past decade.

“Accolades were never the end goal,” DeepMind said in a statement to CNBC. “They simply reflect the importance and impact of the research they have recognized.”

One of DeepMind’s most important products, AlphaFold, was a revolutionary technology that used AI to help scientists predict the structure of proteins, a major challenge in biology for decades.

DeepMind open source AlphaFold, distributing it essentially for free.

In 2017, a team of Google researchers, separate from DeepMind, published a groundbreaking study on Transformers, a way for AI models to better process text used for training.

This sparked the wave of AI innovation that followed, most notably at OpenAI. Transformers is the “T” in ChatGPT.

Critics argued that Google was giving away key products and research and helping its larger competitors.

“They had the advantage and they were very careful,” Hinton said. “They’ve been very careful about both generative AI images and large language models. And when OpenAI partnered with Microsoft and ChatGPT was used by Microsoft, Google couldn’t no longer allow yourself to be careful.”

Hassabis and Google have gone on the offensive.

The Washington Post reported in May 2023 that Google was internally announcing a significant change from its previous policy. Employees had to stop sharing their research with the world, settling for publishing articles after the research was turned into products.

AlphaFold has become a huge business opportunity, securing commercial partners like Eli Lilly and Novartis for drug discovery.

In a TED talk, Hassabis said the ChatGPT moment demonstrated that the public found value in LLMs and was ready to adopt them.

“When we work on these systems, we mainly focus on the flaws, the things they don’t do, and the hallucinations,” he said. “We wanted to improve these things first before releasing them. But interestingly, it turned out that even with these flaws, tens of millions of people still find them very useful.”

Hassabis said it was time to take these products “beyond the rarefied world of science.”

Investors are now waiting to see if Google can accomplish what they consider the most important feat and turn this cutting-edge science into profits.