· concept
Wikipedia (opens in new tab)

Demis Hassabis on Consciousness

Computable consciousness believer (strong)

TL;DR

Demis Hassabis generally posits that consciousness is likely computable, suggesting that AGI could eventually mirror it through complex information processing.

Key Points

  • He explicitly stated that current AI systems do not feel self-aware or conscious in any way as of April 2025.

  • Hassabis believes consciousness is likely computable unless physics proves otherwise, which frames his work on AGI as an empirical test.

  • He once debated the late Daniel Dennett, noting that assuming an AI running on silicon has the same substrate as a human makes it hard to verify its subjective experience.

Summary

Demis Hassabis, the CEO of Google DeepMind, approaches the topic of consciousness from an empirical and scientific standpoint, largely aligning with the view that it is a phenomenon capable of being computed. He has stated that if the entire universe adheres to physical laws that are computable, then consciousness, as an emergent property of the brain, should also be modellable by classical systems, though he acknowledges the current lack of a settled definition. His research focus on Artificial General Intelligence (AGI) serves as a potential method to test this hypothesis, viewing a sufficiently advanced AGI as a mirror against which human cognitive functions, and thus consciousness, can be compared. He suggests that if an AI can replicate the behaviors associated with consciousness, it provides strong evidence that the phenomenon is fundamentally rooted in information processing.

He has also explored the possibility that the subjective experience of consciousness—or qualia—might be unique to biological substrates like the carbon-based human brain, which is a point of debate with figures like Roger Penrose, who leans toward quantum explanations. Despite the philosophical complexity, Hassabis remains committed to the scientific method, suggesting that building and comparing advanced AI to the human mind is the best path to understanding what, if anything, is truly special about human consciousness. He has mentioned the definition of consciousness as "the way information feels when we process it," framing the ultimate answer as an exciting mystery to be uncovered through ongoing AI research.

Key Quotes

I don't think any of today's systems feel self-aware or conscious in any way

...if we mean self-awareness, and these kinds of things… I think there's a possibility that AI one day could be.

Frequently Asked Questions

No, Demis Hassabis has stated that he does not believe any of today's AI systems feel self-aware or conscious in any way. He argues that current systems are capable but still lack genuine subjective awareness or self-experience. However, he does not entirely rule out the possibility that future systems might acquire some level of self-awareness.

Demis Hassabis's working hypothesis, often framed in the context of the universe being an informational system, is that consciousness is computable. He suggests that if an AGI can replicate all human cognitive functions, it serves as evidence that consciousness is essentially a very complex form of information processing. He finds the mystery of consciousness fascinating and believes building AI will help uncover its nature.

While his fundamental view leans toward computability, his public statements show an evolution in acknowledging potential future states. He once suggested there was a possibility AI could one day be self-aware, which contrasts with his more recent firm statement that current systems are not conscious. His focus remains on ensuring AI is safe, useful, and aligned with human values, regardless of its eventual subjective state.