“In my spare time … I work on physics theories about information being the most fundamental unit of the universe – not energy, not matter, but information.” – Demis Hassabis – CEO DeepMind
Demis Hassabis: The Visionary Behind DeepMind
Demis Hassabis stands at the forefront of artificial intelligence, serving as co-founder and CEO of Google DeepMind, a pioneering AI research laboratory. Born in 1976 in London to a Greek Cypriot father and Chinese Singaporean mother, Hassabis displayed prodigious talent from a young age. At just five years old, he mastered chess, eventually competing at a professional level and ranking among the world’s top players in his youth. By 13, he won the junior British Chess Championship. His early fascination with intelligence extended beyond games; he pursued neuroscience, earning a degree from Cambridge University and conducting research at University College London on the brain’s hippocampal system, which underpins memory and imagination.
In 2010, Hassabis co-founded DeepMind with Shane Legg and Mustafa Suleyman, driven by a mission to solve intelligence and build artificial general intelligence (AGI) – machines capable of human-like reasoning across diverse tasks. Acquired by Google in 2014 for approximately $500 million, DeepMind has achieved landmark breakthroughs, including AlphaGo’s 2016 defeat of world Go champion Lee Sedol, AlphaFold’s revolutionary protein structure predictions (earning Hassabis the 2024 Nobel Prize in Chemistry), and recent advances in video generation with Veo and fusion energy research. Hassabis emphasises ethical AI development, insisting on ‘red lines’ against uses in surveillance or weaponry, and views DeepMind as a ‘cathedral to knowledge’ dedicated to advancing humanity.1,2,5
The Quote’s Context: A Glimpse into Hassabis’s Spare-Time Pursuits
This intriguing quote emerged during a conversation with mathematician and broadcaster Hannah Fry, where Hassabis discussed DeepMind’s trajectory towards AGI. Amid talks on world models, simulations, and AI’s role in uncovering physical laws – from protein folding to nuclear fusion – he revealed his personal passion: developing physics theories positing information as the universe’s bedrock. Far from casual musing, this reflects Hassabis’s conviction that reality operates as a vast information-processing system, with AI tools like AlphaFold and Veo reverse-engineering its underlying ‘software’. He argues the universe exhibits ‘learnable manifolds’ – stable, predictable structures that survive evolutionary pressures, modelable by deep learning.1,3
Hassabis envisions AI not merely mimicking reality but discovering its rules, addressing challenges like hallucinations in models by generating ground-truth physics benchmarks from game engines and simple simulations (e.g., pendulums or three-body problems). This aligns with DeepMind’s pursuit of AGI attributes: robust reasoning, hierarchical planning, long-term memory, and hypothesis generation.3,4
Leading Theorists: Pioneers of ‘It from Bit’
Hassabis’s ideas echo foundational work in theoretical physics, particularly John Archibald Wheeler’s seminal 1989 proposal of ‘It from Bit’. Wheeler, a prominent American physicist (1911-2008) who collaborated with Niels Bohr and Richard Feynman, argued that every physical ‘it’ (particle, field, or spacetime) derives its ultimate significance from ‘bits’ – binary yes/no answers to questions, making information ontologically primary. This ‘participatory universe’ concept suggests observers co-create reality through measurement, bridging quantum mechanics and information theory.1
- Claude Shannon (1916-2001): The father of information theory, Shannon quantified information as uncertainty reduction via bits in his 1948 paper ‘A Mathematical Theory of Communication’. His work underpins digital computing and links entropy in physics to informational entropy, influencing views of the universe as a communication system.
- Rolf Landauer (1927-1999): Extended thermodynamics to computation, proving in 1961 that erasing information dissipates heat (Landauer’s principle), tying information irrevocably to physical laws and energy.
- John Archibald Wheeler: Synthesised these into ‘It from Bit’, inspiring digital physics and simulation hypotheses. Modern extensions appear in the holographic principle (e.g., Juan Maldacena’s AdS/CFT correspondence) and Seth Lloyd’s ‘ultimate laptop’ model of the universe as a quantum computer processing information.1,6
Contemporary thinkers like Hassabis integrate these with AI, proposing the universe as a ‘computational’ entity where physics emerges from informational laws, not vice versa. This framework posits stability (‘what survives’) as the sieve shaping reality, from cosmic structures to biological forms – a philosophy driving DeepMind’s quest to simulate and solve the world’s root problems.1
Implications for Physics and AI
If information is fundamental, AI becomes a tool for decoding the cosmos’s source code. DeepMind’s models exemplify this: AlphaFold uncovers protein ‘rules’, Veo simulates video physics, and fusion research targets energy abundance. Yet challenges persist – ensuring simulated physics avoids plausible-but-false hallucinations requires verifiable data and benchmarks. Hassabis cautions on AGI’s path, stressing consistent cognition over narrow prowess, while navigating ethical tensions in an era of rapid scaling.2,3,4
References
1. https://www.glbgpt.com/resource/the-universe-is-code-demis-hassabis-on-ai-games-and-the-future
2. https://time.com/6246119/demis-hassabis-deepmind-interview/
3. https://www.youtube.com/watch?v=PqVbypvxDto
5. https://www.youtube.com/watch?v=-HzgcbRXUK8
6. https://www.bigtechnology.com/p/demis-hassabis-and-sergey-brin-on
