David Silver gave the world’s first glimpse of superintelligence.
In 2016, an artificial intelligence program he developed at Google DeepMind, AlphaGo, learned to play the famously tough game Go with a level of mastery that went far beyond mimicry.
Silver has since founded his own company, Ineffable Intelligence, which aims to build more general forms of artificial intelligence superintelligence. Silver says the company will do this by focusing on reinforcement learning, which is when artificial intelligence models learn recent capabilities through trial and error. The vision is to create “super learners” who go beyond human intelligence in many areas.
This approach contrasts with how most AI companies plan to build superintelligence by leveraging the coding and research capabilities of large-language models.
Silver, speaking to WIRED from his office in London, says he thinks this approach will fail. While LLM programs are amazing, they learn from human intelligence rather than building your own.
“Human data is like a fossil fuel that provides an incredible shortcut,” says Silver. “Systems that learn on their own can be thought of as a renewable fuel — something that can just learn, learn, learn forever, without limits,” he says.
I’ve met Silver several times, and despite this declaration, he always struck me as one of the more humble people in the field of artificial intelligence. Sometimes, when talking about ideas he considers stupid, he flashes a mischievous smile. But now he’s deadly solemn.
“I think of our mission as making first contact with superintelligence,” he says. “By superintelligence, I really mean something amazing. It should discover new forms of science, technology, government or economics on its own.”
Five years ago, such a mission might have seemed absurd. Yet technology CEOs now routinely talk about machines surpassing human intelligence and replacing entire categories of workers. The idea that some recent technological twist could unlock the superhuman capabilities of artificial intelligence has recently spawned a string of billion-dollar startups.
To date, Ineffable Intelligence has raised $1.1 billion in seed funding at a valuation of $5.1 billion, a huge sum by European AI standards. Silver also recruited top AI researchers from Google DeepMind and other pioneering labs to join his venture.
Silver says he will donate all the money he makes from his stake in Effable Intelligence – a sum that could amount to billions if he is successful – to charity.
“Creating a company focused on superintelligence is a huge responsibility,” he tells me. “I think this should be done for the good of humanity, and any money I make from Inefable will be donated to influential charities that save as many lives as possible.”
Total concentration
Silver met Demis Hassabis, CEO of Google DeepMind, at a chess tournament when they were children, and they later became lifelong friends and colleagues.
They remained close after Silver left Google DeepMind, which he only did because he wanted to forge a completely recent path. “I think it’s really important that there is an elite AI lab that actually focuses 100 percent on this approach,” he says. “That this is not just a corner of another place dedicated to LLM.”
Silver argues that the limitations of the LLM approach can be seen with a straightforward thought experiment. Imagine going back in time and releasing a vast language model into a world that believed the world was flat. Without the ability to interact with the real world, the system, he argues, would remain a staunch flat-earther even as it continued to refine its own code.
However, an artificial intelligence system that can explore the world itself could make its own scientific discoveries.