Mustafa Suleyman 深度访谈

Mustafa recounts his journey from co-founding DeepMind in 2010 with Demis Hassabis and Shane Legg, when AI was considered sci-fi and dismissed by academics and big tech alike. His motivation: creating technologies that truly understand humans and improve the human condition.

Mustafa explains AI in simple terms: the physical world converted into an information world. Algorithms learn the structure of information. The key breakthrough was scale — more data and compute enabling learning algorithms to capture the structure of the universe.

Mustafa reframes hallucinations as the creative and imaginative capability of AI — what we actually want. A system that only reproduces training data is just a spreadsheet. We want interpolation, invention, and creativity.

Mustafa argues for regulation not as a brake but as sculpting — chipping away at the pain points of technology in the collective interest. Governments have the responsibility for the common good, which no individual corporate actor will take on alone.

The fundamental question of our era: can we manage these technologies so we coexist safely? Mustafa discusses the two camps — AI will destroy humanity vs. AI will solve everything — and argues for a middle path of responsible development with strong guardrails.