Google DeepMind co-founder and chief executive officer Demis Hassabis stated that there is still some way to go before the emergence of such a configuration of artificial intelligence that is able to cope with any task on a par with humans, but it is a matter of time when the corresponding digital cognitive systems will become a reality.
Mr. Hassabis made the mentioned statement at a briefing in DeepMind’s London offices on Monday, March 17th. In his opinion, artificial general intelligence, which in terms of cognitive abilities is not inferior to humans or even smarter than them, will begin to emerge within the next five or 10 years.
Demis Hassabis stated that today’s machine intelligence systems are very passive, but there are still many things that these functional platforms cannot do. In his opinion, in the next five to 10 years, a lot of those capabilities will start coming to the fore, and the movement towards what is meant by the definition of artificial general intelligence will begin.
Mr. Hassabis interprets AGI as a system able to exhibit all the complex capabilities that humans can. According to him, at the moment there is not quite progress in the relevant technological direction. He noted that advanced digital intelligence systems are impressive in some things. At the same time, according to him, there are other things that the mentioned systems cannot yet do, and there is still quite a lot of research work to go before that.
It is worth noting separately that there is a possibility that at some point in its technological evolution, artificial intelligence will gain the ability to develop independently. It is also noteworthy that the horizons of the potential capabilities of machine intelligence configurations that surpass the human mind are still unknown. It is possible that humanity will not be able to understand this, since a more advanced cognitive system is implied in the relevant context than people’s brains. It can also be tentatively assumed that artificial general intelligence will become something like a new phase of the evolution of thinking. This means that the AGI will probably be able to advance in exploring the world in a global philosophical sense, not limited by focusing on researching only the existence of human civilization.
It is worth noting that Demis Hassabis is not the only proponent of the view that it will take some time for emerging artificial general intelligence. In this case, it is worth mentioning that in 2024, the chief executive officer of one of the largest Chinese technology companies, Baidu Robin Li, stated that, in his opinion, emerging AGI is still more than 10 years away. The mentioned point of view is not consistent with more optimistic expectations. In this context, it implies predictions according to which an artificial intelligence configuration that surpasses the human mind may become a technological reality in the coming years.
The trustworthiness of expectations about when artificial general intelligence will become a full-fledged functioning virtual thinking system is currently something that belongs to the category of hypotheses rather than strictly practical probability. An unambiguous answer to the question of when AGI will become a fact of reality is still virtually impossible.
Dario Amodei, chief executive officer of the startup Anthropic, which operates in the artificial intelligence industry, during a conversation with media representatives at the World Economic Forum in Davos, Switzerland, in January, said that he sees a form of AI that is better than almost all humans, copes with almost all tasks emerging in the next two to three years.
It is worth noting that other technology leaders are expecting artificial general intelligence to arrive even sooner. Cisco’s chief product officer Jeetu Patel suggests that there is a chance that the example of AGI emergence will be observed as early as the current year. During a conversation with media representatives at the Mobile World Congress event in Barcelona this month, he stated that artificial intelligence has three major phases. According to him, there is a basic AI that people are experiencing right now. The next stage, according to Jeetu Patel, is artificial general intelligence, in which cognitive abilities will meet those of humans. There is also, according to him, superintelligence.
It is worth noting separately that there is currently no unanimous common definition of artificial general intelligence. Some experts define AGI as a form of digital intelligence that has the same cognitive abilities as humans. At the same time, there are supporters of the view that artificial general intelligence will surpass the human mind.
Jeetu Patel stated that this year there will be meaningful evidence of AGI being in play in 2025. Superintelligence, in his opinion, will emerge in a few years at best.
Demis Hassabis on Monday announced the expectation that artificial superintelligence will arrive after AGI and become a more advanced thinking system regarding the level of cognitive development of the human mind. According to him, no one really knows when the corresponding breakthrough will occur.
Last year, Tesla chief executive officer Elon Musk predicted that artificial general intelligence would most likely be available by 2026. At the same time, OpenAI chief executive officer Sam Altman stated that a corresponding virtual thinking system can be developed in the reasonably close-ish future.
Demis Hassabis stated that the main challenge associated with the achievement of artificial general intelligence is to get the current digital cognitive systems to a point of understanding the context from the real world.
Although in the realm of games, it is already possible to develop systems capable of solving problems and completing tasks on their own, for example in the complex strategy board game Go, to bring such technology to the real world is proving harder.
Demis Hassabis stated that the question is how quickly it will be possible to generalize the ideas of planning and agentic kind of behaviors, planning, and reasoning, and then generalize that over to working in the real world, on top of things like world models – models that can understand the world around them. In his opinion, significant progress has been made in the world models over the past couple of years. He stated that the current question is, what is the best way to combine this with planning algorithms.
Demis Hassabis and Thomas Kurian, chief executive officer of Google’s cloud computing unit, said the so-called multi-agent artificial intelligence systems are a technological advancement that is gaining a lot of traction behind the scenes.
Mr. Hassabis noted that lots of work is being done to get to this stage. In the relevant context, he referred as one example to the work of DeepMind for getting machine intelligence agents to figure out how to play the popular strategy game Starcraft. According to him, a lot of work has been done in the past on things like Starcraft, where there was a society of agents or a league of agents and they could compete or collaborate.
Moreover, Demis Hassabis said agent-to-agent communication is what DeepMind does to allow the agents to express themselves.
Despite significant differences in predictions about when artificial general intelligence to arrive and different approaches to defining the corresponding concept, it is already obvious that virtual thinking systems, in terms of what can be described as an evolutionary stage, are approaching the verge of a new leap in development. It is worth noting that mental superiority in the context of a kind of competition between human cognitive abilities and the potential horizons of thought available to machine intelligence is not something that can be unambiguously characterized. The capability to operate with data stored in memory according to a specific context is an option of the mind, which, within the framework of some concepts of determining the level of intellectual development, may actually be the only subject of interest. At the same time, it is also clear that without the active function of abstract comprehension and reasoning, the thinking process will not be complete. Artificial intelligence has already demonstrated high-quality abilities related to working with huge amounts of information materials as part of a discussion on the theme that users choose. At the same time, the limits of the capabilities of digital thinking systems in the context of abstract perception of facts, phenomena, and items of the surrounding reality, and generating conclusions on certain issues in the context of a generalized, beyond the strictly utilitarian approach to the object of study, method are still unknown.
As we have reported earlier, OpenAI to Give Microsoft Access to Its AGI.
