Gemini’s latest advancements hint at a future of artificial general intelligence, with Google DeepMind CEO Demis Hassabis outlining his vision for AGI and its potential to revolutionize personal assistants, humanoid robots, and human-like AI.
The Future of Artificial General Intelligence: A Vision from Google’s AI Boss
Google DeepMind CEO Demis Hassabis has outlined his vision for artificial general intelligence (AGI), a long-sought goal in the field of artificial intelligence. According to Hassabis, reaching AGI will require significant advancements in areas such as reasoning, agency, and world-modeling capabilities.
Demis Hassabis is a British computer scientist and artificial intelligence (AI) expert.
He co-founded the AI company DeepMind in 2010, which was later acquired by Alphabet Inc., Google's parent company.
Hassabis holds a degree in Psychology from King's College London and has a background in neuroscience.
He is credited with developing AlphaGo, a program that defeated a human world champion in Go, a complex strategy board game.
Building Blocks of AGI
Hassabis points to Google’s flagship Gemini models as a key area of focus for developing AGI. ‘Gemini Flash’ and ‘Gemini Pro’ are the fastest and most capable AI models from Google, respectively. Hassabis notes that Gemini Pro outperforms other models on LMArena, a widely used benchmark for measuring AI abilities.
Reasoning and Agency
Gemini’s nascent reasoning, agentic, and world-modeling capabilities have the potential to enable more capable and proactive personal assistants, humanoid robots, and eventually AI that rivals human intelligence. Hassabis believes that these capabilities are crucial for developing truly intelligent machines.
Simulated Reasoning and World Modeling
Deep Think, a new simulated reasoning capability, is being developed for the Pro model. This technology uses more compute time and undisclosed innovations to improve upon traditional large language models. ‘Deep Think enables AI models to break down problems and deliberate in a way that resembles human reasoning.’
Humanoid Robots and Practical Applications
Hassabis emphasizes the need for humanoid robots with general intelligence to operate reliably in complex environments. Google is collaborating with companies like Apptroniks and Tesla on humanoid robotics projects. Hassabis believes that the key to developing practical applications lies not in the robot itself but in its understanding of its physical context.
The Path to AGI
Hassabis estimates that it may take five to 10 years for machines to master everything a human can do. However, he acknowledges that this timeline is still relatively short-term and that the pace of progress will depend on various factors. Hassabis believes that AI must become more inventive if it is to replicate human intelligence faithfully.
Expanding Creativity in AI
Google is exploring ways to enhance creativity in AI models. The recent unveiling of AlphaEvolve, a coding agent capable of generating new algorithms for longstanding problems, is an example of this effort. Hassabis envisions AI playing games inside realistic 3D worlds as a means to expand creative capabilities beyond math and coding.
The Road Ahead
Hassabis’s vision for AGI highlights the need for significant advancements in reasoning, agency, and world-modeling capabilities. As Google continues to develop its AI models, it is essential to prioritize these areas to create truly intelligent machines that can augment human capabilities.
