AI Superpowers Summary

In March 2016, a computer program beat Lee Sedol, one of the finest Go players alive, and a quiet technical shift became public theater. The program was AlphaGo, built by DeepMind, and its victory told many people that artificial intelligence had crossed from promise into force.

Kai-Fu Lee watched that moment from a rare seat. He had led Google China, worked at Apple and Microsoft, invested in Chinese startups, and lived inside both Silicon Valley polish and Beijing street-fight business.

His core claim is simple and useful. The age of AI discovery is giving way to the age of AI use, and the countries that win will not only have clever scientists. They will have data, hungry founders, fast markets, and governments willing to build roads for the new traffic.

The spark mattered. The blast furnace mattered more.

Geoffrey Hinton’s team shocked the ImageNet contest in 2012 by using deep learning, a method that lets software learn patterns from many examples. Four years later, AlphaGo’s win over Lee Sedol made the same shift feel less like a research note and more like a weather event.

Lee’s point is that AI moved from the age of expert craft to the age of mass production. Earlier AI often needed hand-built rules, like a nervous cookbook written by engineers. Deep learning works differently. Feed it enough examples, give it enough computing power, and it can find useful patterns that humans did not spell out.

When the recipe starts improving itself, the chef is no longer the only star in the kitchen.

This matters because it changes who can compete. In the old model, a tiny group of top researchers held the rarest tools. In the new model, great research still matters, but strong execution matters more than before. A company that can gather data, test features, and ship fast may beat a company with prettier theory.

AI stopped being a lab prize and became industrial plumbing.

That is why Lee keeps pulling attention away from magic and toward deployment. The furnace is not fed by wonder. It is fed by repeated use. Every search, click, ride request, scan, and purchase can become training material if a company has the right system around it.

The broader consequence is sharp. AI power will gather where feedback loops are dense. The best model attracts more users, more users create more data, and more data improves the model. That loop can turn a lead into a wall.