At the conference “Future of Britain 2024”, organized by the Global Changes, Demis Hassabis, General The director and co-founder of Google Deepmind, shared his vision of the current state and future artificial intelligence (AI). In an interview with the former Prime Minister of Great Britain, Tony Blair, Hassabis touched on the theme of the development of artificial general intelligence (AGI) and made an unexpected comparison.
“In terms of general intelligence, we have not yet reached even the level of a cat,” Hassabis noted, answering the question of the progress of Deepmind in the AGI field. This statement aroused a keen interest among the participants in the event.
According to him, although modern AI is able to create texts, images or music, convincingly imitating a person, even pets are still better trained and adapt to various situations.
“Today we are still far from the human level in all respects,” he said. “However, in separate highly specialized areas, for example, in games, AI already exceeds the best players in the world.”
Despite the current restrictions, Hassabis expressed confidence in the huge AI potential. In his opinion, the influence of smart technologies on society will be comparable to the effect of the industrial revolution or the development of fire and electricity. He suggests that AI will become a key factor in the acceleration of scientific discoveries in the energy, material science, health care, climatology, and mathematics.
AI already makes a significant contribution to scientific research in these fields. The researcher also drew a parallel with the beginning of the 2000s when everyone spoke about “big data” (AI systems were a response to the calls related to the processing of huge arrays of information).
Hassabis paid special attention in his speech to the DeepMind project called Project Astra. The purpose of this initiative is to create a “universal AI-agent” that can go beyond the capabilities of ordinary chatbots. As planned, the model will take into account the context of the situation, the environment of the user, preferences, and the history of interactions, which will make the technology much more useful in everyday life.
Among the key obstacles to the AGI era, Hassabis identified the ability to plan, efficient use of memory, the ability to use tools, and ask meaningful questions.