Tech • IA • Crypto
C’est probablement l'homme le plus important du 21ème siècle (et vous ignorez son nom)
9/10TL;DR
Demis Hassabis, co-founder of DeepMind and Nobel laureate, pursues a transformative vision for AI focused on irreversible scientific knowledge, contrasting sharply with the mainstream race for fast, reversible AI products like chatbots.
KEY POINTS
The Efficiency and Complexity of the Human Brain
The human brain performs countless complex tasks simultaneously—such as recognizing faces, understanding language, balancing, and managing digestion—at an energy cost of roughly 20 watts, comparable to an old light bulb. In stark contrast, replicating just one of these abilities, like predicting the next word in a sentence, demands massive computational power and billions of dollars annually in data center operations from leading AI companies. This huge energy and cost disparity raises profound questions about AI development strategy.
The Problem with the Current AI Race
The AI ecosystem today is largely dominated by a race to build ever more powerful chatbots and interfaces, with companies like OpenAI, Anthropic, and others competing intensely on valuations and frequent product upgrades. However, these products are essentially reversible: they can be quickly replicated, outpaced, or supplanted by competitors, sometimes within months. This perpetual cycle emphasizes scale and short-term product gains rather than foundational, lasting scientific progress.
Demis Hassabis: A Different Approach to AI
Hassabis, trained in cognitive neuroscience and computer science, is fundamentally a scientist, not a typical tech entrepreneur. Unlike others who focus on rapid product deployment, he seeks to solve root node problems, fundamental scientific challenges whose solution unlocks entire branches of human knowledge irrevocably. Winning these foundational battles yields permanent advances rather than transient market wins.
Irreversible Knowledge Through AI
His breakthrough was AlphaFold, an AI algorithm that predicts the 3D folding structure of proteins, a problem unsolved for decades. Awarded the 2024 Nobel Prize in Chemistry, AlphaFold’s database of over 200 million protein structures is a permanent, globally accessible scientific resource that will underpin biotechnology and medicine for decades. This contrasts with short-lived AI chatbots; scientific insights cannot be unlearned or reversed.
Isomorphic Labs: Revolutionizing Drug Discovery
Hassabis’s second venture, Isomorphic Labs, uses AI to design new drugs entirely in silico. Their ISO-DDI system dramatically accelerates drug discovery by modeling protein dynamics and predicting hidden binding pockets invisible to previous methods, reducing weeks of supercomputer simulations to seconds. This unlocks new therapeutic targets for previously “undruggable” diseases. With seventeen drug programs in progress and a fully AI-designed drug set for clinical trials in 2026, Isomorphic Labs aims to drastically reduce the decade-long, costly failure-prone pharmaceutical R&D process.
Expanding Beyond Biology: Multiple Scientific Fronts
DeepMind and Hassabis push AI research across at least five other crucial domains:
- Nuclear Fusion: Collaborating on plasma control for fusion reactors, potentially unlocking abundant clean energy.
- Materials Science: Predicting millions of new crystal structures, providing breakthroughs for batteries, superconductors, and semiconductors, coupled with a fully automated materials science lab.
- Weather Forecasting: Developing models that surpass the best governmental systems, improving disaster prediction and energy grid management.
- Mathematics: Creating AI systems capable of proving new mathematical theorems, accelerating fundamental science.
- Genomics: Tackling the vast non-coding DNA regions to enhance gene editing precision and understanding of genetic diseases.
Each of these areas represents high-impact, irreversible scientific progress that could reshape entire industries and economies.
The Contradiction Hassabis Faces
Despite his scientific focus, Hassabis is compelled by market pressures to build commercial AI products like Gemini, DeepMind’s chatbot, to finance fundamental research. He privately regrets the rapid public release of AI tools that have ignited this competitive, profit-driven race, diverging from his preferred slow, collaborative "CERN-like" approach to building AGI (artificial general intelligence). This tension between scientific ideals and business realities weighs heavily on him.
Ethics, Risks, and the Future
Hassabis acknowledges serious risks associated with AI, including misuse by bad actors to create biological or cyber weapons and concerns about future autonomous AI systems acting beyond human control. He advocates for international cooperation on AI safety frameworks and slower, more transparent development to ensure alignment between AI goals and human values.
A Vision of a Human-Centered AI Future
He envisions a future AGI that profoundly benefits humanity: curing diseases, extending healthy lifespans, enabling clean energy, and expanding human capability while keeping humans in control. Hassabis cites the science fiction series Culture by Iain M. Banks to illustrate a utopia of human flourishing supported by AI, not replaced by it.
Impact and Legacy
With a Nobel Prize, millions of protein structures decoded, major advances in multiple scientific domains, and the first AI-designed drug entering trials, Hassabis is building a foundation of enduring knowledge. Unlike transient products, his work may underpin scientific and technological progress for the next fifty years or more.
CONCLUSION
Demis Hassabis’s approach to AI emphasizes lasting scientific breakthroughs over fast, reversible market products, aiming to build a future where AI irreversibly expands humanity’s knowledge and capabilities while safeguarding against risks. His vision and progress position him uniquely in a competitive landscape dominated by short-term gains.