Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
The co -founder and executive director of Google Deepmind, Demis Hassabis, speaks during the Mobile World Congress, the largest annual meeting of the telecommunications industry, in Barcelona, Spain, February 26, 2024.
Barrena Pau | AFP | Getty images
London – Artificial intelligence that can match humans in any task is still somehow, but it is only a matter of time before it becomes a reality, according to the CEO of Google Deep.
Speaking in an informative session at the Deepmind offices in London on Monday, Demis Hassabis said he believes that artificial general intelligence (AGI), which is so intelligent or smarter than humans, will begin to emerge in the next five or 10 years.
“I think today’s systems are very passive, but there are still many things that they can’t do. But I think that in the next five to 10 years, many of those capabilities will begin to advance and we will begin to move towards what we call artificial general intelligence,” said Hassabis.
Hassabis defined AGI as “a system that can exhibit all the complicated abilities that humans can.”
“We are not there yet. These systems are very impressive in certain things. But there are other things that they still can’t do, and we still have a lot of research work to go before that,” said Hassabis.
Hassabis is not just by suggesting that Agi will take time to appear. Last year, the CEO of the Chinese technological giant Baidu Robin Li said that agi is “More than 10 years away“Retroading the excitable predictions of some of their classmates about this advance that takes place within a much shorter period.
Some time to go still
The Hassabis prognosis pushes the timeline to get to AGI in some way compared to what their industry companions have been drawing.
Dario Amodei, CEO of AI Startup Anthrope, told CNBC in the World Economic Forum in Davos, Switzerland in January that sees a form of AI that is “better than almost all humans in almost all” emerging tasks in the “next two or three years.”
Other technological leaders see that Agi arrives even before. The product director of Cisco, Jeetu Patel, believes that there is the possibility that we can see an example of AGI emerge as soon as this year. “There are three important phases” for AI, Patel told CNBC in an interview at the Mobile World Congress event in Barcelona earlier this month.
“There is the basic AI that we all experience at this time. Then there is artificial general intelligence, where cognitive abilities meet those of humans. Then there is what they call superintelligence,” Patel said.
“I think you’ll see significant evidence that Agi is at stake in 2025. We are not talking about years away,” he added. “I think that superintelligence is, at best, a few years.”
Super artificial intelligence is expected to arrive after Agi and overcome human intelligence. However, “nobody really knows” when such progress will happen, Hassabis said Monday.
Last year, the Tesla CEO, Elon Musk, predicted that Agi would probably be available for 2026, while the Operai CEO, Sam Altman, said Such a system could be developed in the “reasonably close future.”
What is needed to get to AGI?
Hassabis said that the main challenge with the achievement of artificial general intelligence is to bring today’s AI systems to a point of understanding the context of the real world.
While it is possible to develop systems that can break down problems and complete tasks autonomously in the field of games, such as the Complex strategy board game – Bringing such technology to the real world is more difficult.
“The question is, how fast can we generalize planning ideas and the type of agent behavior, planning and reasoning, and then generalize it to work in the real world, in addition to things like world models, models that can understand the world around us,” said Hassabis, “said?
“And I think we have progressed well with world models in recent years,” he added. “So, now the question is, what is the best way to combine that with these planning algorithms?”
Hassabis and Thomas Kurian, CEO of the Google cloud computing division, said that the so -called “multiple” systems are a technological advance that is gaining a lot of traction behind the scene.
Hassabis said a lot of work is being done to reach this stage. An example that he referred to is Deepmind’s work, that AI agents discover how to play the popular “Starcraft” strategy game.
“We have worked hard on that with things like Starcraft Game in the past, where you have a society of agents or a league of agents, and could be competing, they could be cooperating,” said Deepmind’s head.
“When you think about agent to agent communication, that’s what we are also doing to allow an agent to express … What are your skills? What kind of tools do you use?” Kurian said.
“Those are all the elements you need to ask an agent a question, and then, once you have that interface, other agents can communicate with it,” he added.
