Tom Davidson

About me

I research risks from advanced AI at Forethought.

Recently, I’ve focussed on two main themes:

1. AI-enabled human coups: Could a few people use superhuman AI to seize power over a country, or even the world?
2. AI takeoff speeds: How long will it take to go from broadly human-level AI to superintelligent AI?

I’ve previously worked at Open Philanthropy and the UK’s AI Security Institute.

I’m on X, @TomDavidsonX. I’m also on Google Scholar.

Research top picks

AI-enabled human coups

• AI-enabled coups: how a small group could use AI to seize power
• Podcast with 80,000 hours on AI-enabled coups
• Human takeover might be worse than AI takeover

Takeoff speeds

• Will AI R&D Automation Cause a Software Intelligence Explosion?
• Three types of intelligence explosion
• Continuous doesn’t mean slow
• Podcast with 80,000 hours on how quickly AI could transform the world

Economics

• Could advanced AI drive explosive economic growth?
• The Industrial Explosion
• What are the social returns globally to spending on R&D?

Other research

• AI capabilities can be significantly improved without expensive retraining
• Will compute bottlenecks block a software intelligence explosion
• Will training new AI systems from scratch block a software intelligence explosion
• Could Advanced AI Accelerate the Pace of AI Progress? Interviews with AI Researchers
• A compute-centric framework for AI takeoff speeds
• Should There Be Just One Western AGI Project?
• Semi-Informative Priors over AI Timelines