Shane Legg, co-founder of Google DeepMind, has cautioned that rapidly advancing AI capabilities should not outpace safety measures, as calls grow to slow the development of cutting-edge models.
“We're living in a period now where capabilities are advancing very, very quickly,” Legg said in an interview with the Financial Times. “But we can't let capabilities get ahead of safety.”
Legg described Anthropic CEO Dario Amodei's proposal to slow, but not halt, the rollout of frontier models due to safety concerns as “interesting directionally” and “worth considering”.
Also Read | Layoffs Are Rising In Tech, But These Skills Are Helping People Unlock Better Pay
“But we need to really work through the details of that and how that would work in practice,” he added.
ALSO READ | OpenAI Weighs Funding Round At Over $1.2 Trillion Valuation
Legg made the remarks as he unveiled the DeepMind Institute, which aims to examine the implications of artificial general intelligence. Google DeepMind describes AGI as a system possessing the cognitive capabilities of the human brain.
Despite the technology's progress, Legg said it was too early to declare that AGI, a term he helped coin, had been achieved, despite recent assertions by executives at Nvidia and OpenAI.
Legg said he remained “comfortable” with his longstanding prediction that there was a 50% chance of achieving “minimal” AGI by 2028.
Legg is now the chief AGI scientist at Google DeepMind. Last month, Sir Demis Hassabis moved from his role as CEO of the Google-owned AI lab to become its chair. The third co-founder, Mustafa Suleyman, now holds a senior executive position at Microsoft.
Legg said the new DeepMind Institute would seek to “elevate” the discussion on AGI by making the company's technical research available to a wider audience. The other two directors are Hassabis and James Manyika, Google's senior vice-president.
“The public discourse is really lighting up now. And people are taking the possibility of powerful general intelligence very, very seriously,” Legg said. “So it seems like this is really the moment to start to engage in a broader public discussion around many of these fascinating topics.”
Acknowledging that different people define AGI in various ways, Legg noted that it was unclear whether Astra fulfilled OpenAI's specific definition, namely, the ability to perform most economically valuable cognitive tasks.
“Obviously they're accomplishing some remarkable things, but it's not yet at a stage where I believe it can even satisfy their own criteria,” he said.
Legg said the institute would address a broad spectrum of topics connected to AGI, including science, education, society, policy, philosophy and human flourishing. Many of these essays would be authored by Google staff, while others would come from external experts to foster a wider range of perspectives.
The institute is initially releasing three essays on AI safety, economic policy and principles for a “new utopianism”.
Also Read | OpenAI Says It's Working With Anthropic, Google On AI Safety
An essay by DeepMind researchers Rohin Shah and Anca Dragan, discussing so-called “chain of thought” reasoning models, criticised the rapid rollout of OpenAI's GPT-6 Astra model, arguing that it exhibited “a concerning downward trend” in transparency. They noted that competitive pressures might discourage transparency and proposed that regulation could mitigate this.
However, they contended that stronger models could be created provided that transparency architecture was incorporated into their design.
“For all the ways we still don't understand AI systems, we shouldn't overlook how fortunate it is that the best reasoning models today think out loud in a way humans can understand,” they wrote.
Manyika said the only AI competition Google wanted to join was “the race to get it right.”
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.
