Demis Hassabis, CEO of Google's DeepMind group, believes that the prospect of artificial intelligence being able to match humans in every task is still a long way off, but it is only a matter of time before it becomes a reality.
Specifically, speaking at a press conference at DeepMind's office in London, the person who holds an important leadership role in Google's AI division predicted that artificial general intelligence (AGI) - a form of AI capable of performing almost any job that humans can do - will soon appear in the next 5-10 years.
Artificial super intelligence equal to humans
“I think that the systems today, they are very passive and there are still a lot of things they can't do. But I think that in the next 5-10 years, it's likely that that will start to happen and we'll start to move towards what's called general artificial intelligence,” Hassabis said.
DeepMind's boss defines AGI as “a system capable of performing all the complex activities that a human can do.”
Hassabis isn’t the only one who thinks AGI will take a while to arrive. In 2024, Baidu CEO Robin Li shared his vision of AGI being “more than 10 years away” from becoming a reality.
Creating artificial general intelligence, or AGI, is the stated goal of all major AI research labs. It is a branch of superintelligence that is capable of cognition and can learn and improve its abilities over time.
 |
Demis Hassabis, CEO of Google's DeepMind group, believes that the prospect of artificial intelligence being on par with humans will appear in the next 5-10 years. Photo: Bloomberg. |
According to ChatGPT, AGI has the potential to outperform humans at “nearly every task of economic value.” In a blog post in early 2024, Altman wrote at length about AGI, talking about the incredible new capabilities it could bring.
“We can imagine a world where everyone has access to ask for help with tasks that require cognitive intelligence. This provides a huge pool of human ingenuity and creativity,” said the OpenAI CEO.
If AI systems can reason on their own, it will improve consistency in queries as well as enhance their ability to solve many practical problems.
However, for many experts, AGI remains a vague concept. Many argue that achieving AGI is a huge challenge. According to TechCrunch , the end point of AI may be smarter models, but not necessarily models with human-level understanding.
Mixed predictions
Hassabis's forecast is actually a step back in terms of the timing of achieving AGI compared to what his industry peers have outlined.
Dario Amodei, CEO of AI startup Anthropic, told CNBC at the World Economic Forum in Davos, Switzerland that a form of AI that is “better than most humans at most tasks” will emerge in the “next two or three years.”
 |
As AI's automation capabilities increase, the risk of it replacing humans increases. Photo: New York Times. |
Other tech leaders are envisioning AGI coming even sooner. Cisco’s Chief Product Officer Jeetu Patel said it’s possible the world could see an example of AGI as early as 2025.
“There are three main stages, including the basic AI that everyone is experiencing today. Then there is artificial general intelligence, where cognitive capabilities meet those of humans. Finally, there is what they call superintelligence. I think you will see meaningful evidence of AGI in operation by 2025. We are not talking about many years from now,” Patel said at the Mobile World Congress event in Barcelona.
Elon Musk also predicted that fully capable AGI would appear by 2029. According to the billionaire, the capabilities of new artificial intelligence models will surpass human intelligence by the end of 2025, as long as the power supply and hardware power can meet the needs of this technology.
But development has been slowed by bottlenecks in the supply of microchips, especially those made by NVIDIA, which are essential for training and running AI models. Those constraints have slowed things down somewhat, but new models are also being tested in other data center and power grid equipment, Musk said.
 |
Artificial general intelligence (AGI) is a term used to describe machines that demonstrate cognitive abilities similar to humans. Photo: The Conversation. |
Meanwhile, Hassabis argues that the main challenge to achieving AGI is getting current AI systems to the point where they can understand context from the real world.
While it has been possible to develop systems that analyze problems and complete tasks automatically in the gaming domain, bringing such technology to the real world is becoming increasingly difficult.
DeepMind's CEO added that there's still a lot of work to be done to get to this stage. One example he mentioned was work to help AI agents figure out how to play the popular strategy game Starcraft .
“We’ve done a lot of experiments like playing Starcraft in the past. That’s where you have a society of agents or a federation. They can compete and they can also cooperate,” the DeepMind head said.
Source: https://znews.vn/giam-doc-ai-google-du-bao-thoi-diem-sieu-ai-xuat-hien-post1539171.html
Comment (0)