AI's Future: From Digital Realm to Physical World, Says Hassabis

The current development direction of AI technology is to use multimodal approaches to enable machines to deeply understand the world, thereby achieving true service to users. Artificial intelligence actually acts as a bridge connecting machines with human needs.

The true direction and path of this round of AI technology is to enable machines to understand the world through multimodality, thereby truly serving users. AI is a bridge connecting machines and humans.

Demis Hassabis's view on AI is more measured and noteworthy than those of Sam Altman and Elon Musk. He doesn't believe AGI will be achieved in the short term, but rather that AI capabilities are moving from passive systems to active systems.

Hassabis mentions that AI capabilities need to move from the currently limited bit world to the atomic world, i.e., the physical real world. He believes this is an area where AI will make tremendous progress in the coming years. The significance of multimodal capabilities lies in enabling AI to understand the surrounding world, and then to deal with real-world problems, truly helping users. This is the fundamental capability and challenge of AGI.

In the interview, Hassabis discussed the following key points:

  1. Current AI systems still lack planning, memory, and tool use capabilities, which are necessary for active systems.

  2. Games are important for AI development because they have clear goals, facilitating testing and validation of algorithms.

  3. "Embodied intelligence" is important, such as self-driving cars and robots that can interact with the real world.

  4. The AlphaFold project solved the major biological challenge of protein folding, which is significant for understanding biology and designing drugs.

  5. Gemini's multimodal capabilities allow it to process various types of information including text, images, and audio, which is crucial for AI to understand the world.

  6. Open-source is important for scientific progress, but for powerful systems approaching AGI, potential negative impacts need to be considered.

Overall, Hassabis believes that AI development should focus on enabling machines to truly understand and serve the real world, rather than merely staying at the level of content generation.