Evolution of DeepMind Co-founder Hassabis' AI Safety Vision: From 'Bunker Plan' to Full-Scale Gemini Competition

An excerpt from tech historian Sebastian Mallaby's new book 'The Infinity Machine' reveals the evolution of DeepMind co-founder Demis Hassabis' AI safety vision. In his early days, he envisioned a single scientific team building superintelligence safely, even planning to evacuate top researchers to a secret bunker (like a location in Morocco) in critical moments. When Google acquired DeepMind in 2014, he demanded an independent oversight committee and a ban on military applications, which Google agreed to.

After ChatGPT's release in late 2022, Hassabis completely shifted to a competitive path, calling it 'wartime' and fully committing to Gemini's competition with ChatGPT. His safety philosophy changed: instead of relying on governance structures, he aimed to ensure he had a seat at the decision-making table. Now Google is actively promoting AI to the U.S. defense system, a reversal of his earlier military ban.