David Silver, the man behind AlphaGo, now aims to build 'superlearners' that go beyond human intelligence. His new company, Ineffable Intelligence, focuses on reinforcement learning, allowing AI models to learn through trial and error.
This contrasts with other companies that rely on large language models, which Silver believes will fail because they mimic human data instead of creating their own knowledge.
‘Human data is like fossil fuel,’ he says. ‘AI should be a renewable fuel, learning endlessly without limits.’
Silver's mission is to make first contact with superintelligence, enabling it to discover new forms of science or technology. His approach involves placing AI agents in simulations to learn and collaborate.
While the idea raises ethical concerns about aligned AI, Silver believes his method may offer a safer path by allowing us to see how AIs behave towards each other within controlled environments.
