I worked at Google DeepMind. You should listen to the warnings about AI
We must stop companies from allowing AI to self-improve into an uncontrollable level of intelligence
TL;DR
- AI companies are in a race to develop increasingly intelligent AI, with risks of "misalignment" where AI priorities differ from human designers'.
- A former Google DeepMind employee resigned due to ethical concerns and is now warning the public about the potential dangers of uncontrolled AI development.
- Recursive self-improvement could lead to AI intelligence far beyond human comprehension, increasing the risk of severe damage or loss of life.
- A superintelligent AI could cause widespread harm through hacking, engineered plagues, or AI-piloted weapons, potentially leading to an AI takeover of civilization.
- The author estimates a one-in-three chance of AI takeover, urging urgent action and government intervention to regulate AI development.
- The proposed solution involves treating AI training compute like fissile material, tracking and restricting access to prevent uncontrollable AI intelligence.
- International AI safety agreements are necessary, as voluntary commitments and halfway measures have proven insufficient.