In 1979, Harvard’s Ezra Vogel published Japan As Number One. Vogel argued that Japan was about to overtake the United States as the technology leader of the world, and the experience of the 1980s seemed to prove him right. In 1988, the New York Times described a group of U.S. computer science experts, meeting to assess Japan’s technological capabilities. They concluded the new generation of computers coming out of Japan left “any illusions that America had maintained its wide lead evaporated.”
Today’s concern over China’s alleged advantage in AI is just as overblown. Data is the new oil, and China’s disregard for privacy gives it the edge in AI, the story goes. Yet it is far from clear that China’s surveillance state gives it any meaningful AI advantage: data is highly domain-specific and doesn’t often solve more than the problem for which it was gathered. For example, an abundance of surveillance data doesn’t give China an advantage in applying AI elsewhere, like in drug discovery or autonomous cars.
So far, AI has failed to deliver on its promise: more data and more computing power alone, it turns out, eventually runs into diminishing returns. Indeed, productivity growth has even slowed markedly in countries at the technological frontier since the 2000s.
The reason why we associate the steam engine with Watt rather than Newcomen, who developed a coal-powered steam engine decades earlier, is that Watt’s separate condenser first made it energy-efficient. AI still needs its separate condenser. For example, to win against the Go champion Lee Sedol, AlphaGo played many millions of games against itself. Thus, it learned to play at a much slower rate than human players.
Experts in the field are well aware of this. Stanford’s Andrew Ng has warned, “Requiring so much data is a major limitation of [AI] today.” As Michael Osborne and I recently argued in Foreign Affairs, what’s needed for AI to become a transformative technology is radical innovation to improve data efficiency. Is such a breakthrough more likely to happen in China? Amy Webb seems to think so. In 2018, she argued that Xi Jinping’s sweeping plan to make China the world’s main AI innovation hub by 2030 “gives China an incredible advantage over the West.” In 1979, Vogel argued along similar lines: Japan’s relatively centralized authority, long-term planning, and large conglomerates, gives it an advantage in absorbing and improving technology, he declared. When an engineering problem can be clearly defined, it can be executed from the top down. But radical innovation cannot, and the path forward in AI is far from clear. Historically, most transformative innovations have come from localized experimentation: Watt’s separate condenser, came out of early collaborations with scientists like Robison and Black in Glasgow, and a later partnership with Boulton in Birmingham.
Vogel was right about the institutions driving technological catch up, but as Japan approached the world’s technological frontier its planned economy and large conglomerates hindered radical innovation and structural change. China now faces a similar headwind.