Agi – how many years are for a breakthrough?
Hassabis defines Agi as a system capable of performing all complex tasks that people can do. Although the current AI models impress in some areas, they still have high restrictions and require further research.
The head of Deepmind is not the only expert who indicates that we will wait for Aga. Robin Li, Chinese Baidu, believes that This technology will appear in over 10 years. Dario Amodei, in turn, Anthropic startup CEO, predicts that Ai capable of performing most human tasks will be available within 2-3 years. Even more optimistic is Jetu Cisco Patel, which suggests that the first signs of Aga may appear as early as 2025.
Superintelligence
After reaching Aga, the next step will be artificial superintelligence (asi), So systems that exceed human intelligence. Hassabis emphasizes, however, that no one can determine when such a breakthrough will happen. Elon Musk predicts Aga for 2026, and Altman from OpenAI speaks of “close future”.
According to Hassabis, the main obstacle to the development of Aga is to teach a and understanding context in the real world. AI models are great at solving problems in games, such as strategy of it, but moving these skills to reality turns out to be much more difficult.
AI multi-agent systems-the key to the future?
Google Deepmind is intensively working on the so -called AI multi-agent systems that could work more effectively and communicate with each other. An example is a project in which AI learned to play Starcraft strategy, competing and cooperating as part of the “League of Agents”.
Hassabis emphasizes that the development of AGA requires further integration of models with planning and communication algorithms between AI agents. Only in this way will it be possible to create systems that will be able to act in real conditions.
