Why didn’t Japan invent generative AI?

Today's "AI arms race" is a sequel. The first played out between the US and Japan. And it isn't over yet.

The sudden appearance of China’s DeepSeek R1, a cheap, capable, open source large language model, sent the world into an absolute tizzy last week. Observers declared it the “Sputnik moment” in an ongoing AI arms race. Many questions remain about the actual costs involved in building DeepSeek R1, the datasets used for training, and the implications of using an AI that censors things that Chinese leadership doesn’t want the world talking about. But it got me thinking: why didn’t Japan get there first?

I don’t mean this in in the sense that China couldn’t or shouldn’t be allowed to build an AI. I’m not surprised in the slightest that Chinese innovators did so; “AI sovereignty” is a hot topic these days. But their coup only serves to highlight how little of an impression Japan has made in the field of AI since 2022. I pick that year because it is when the ChatGPT chatbot and DALL-E text-to-image generator captured global imaginations (or triggered nightmares, depending on your view). Western AI firms love using Japanese imagery to sell their products, but Japanese AI firms don’t play much of a role in the global dialogue.

This is really kind of weird, when you think about it. You’d be forgiven for not knowing this, but Japan was in fact a pioneer in the field of AI. And it has an even longer history of re-packaging foreign innovations into products that wow the world. So not being first to the game isn’t a historic disadvantage. Transistor radios, karaoke machines, televisions, automobiles, Walkmans, game consoles, anonymous imageboards — you name it, all of it based on tech foreign companies created first, all of it re-envisioned by Japan to great effect.

That did not happen with generative AI. It absolutely could have, as you’ll read. But it didn’t. And Japan is well aware: when a pair of North Americans won the 2024 Nobel Prize in Physics for research in neural networks, it led to some grumbling about Japan being written out of AI history.

To understand how that happened, let’s turn back the clock to the the 1980s. The capabilities of computers were advancing by leaps and bounds even then, and Japan seemed to have a lock on manufacturing them more cheaply than the West. “What do we want our kids to do? Sweep up around Japanese computers?” asked Walter Mondale in the run-up to his 1984 presidential campaign. Fear of a Japanese Planet, as it might be called, was a bipartisan sport. In 1987, Reagan imposed a 100% tariff on Japanese electronics to protect American chip manufacturers. But he forgot to include game consoles, so by 1991, Mondale’s nightmare came true, and there really was a Japanese computer in nearly every house. Thus when Apple president Michael Spindler was asked that year which company he feared most, he answered with a single name: “Nintendo.”

Despite all this hand-wringing, it wouldn’t be Japanese tech companies who would transform the planet. It would be Silicon Valley entrepreneurs who unleashed innovations ranging from the iPod and iPhone to the social media systems that now shape culture around the globe. We might add AI to this pile of missed opportunities.

Was it a lack of creative vision? Impossible. Artificially intelligent machines have long played a role in Japanese pop culture. The very first cartoon to be described as “anime,” 1963’s Astro Boy, starred an AI hero. Doraemon, another massively popular franchise, features an atomic-powered, time-traveling robot cat. These characters have in turn nourished the dreams and careers of many a Japanese engineer.

Nor could it be attributed to a lack of technological prowess. Those engineers made pioneering strides in neural networking, image recognition, and deep learning algorithms as far back as the Sixties. By 1982, the Japanese government had grown so convinced that its researchers were on the cusp of an AI revolution that it launched a massive artificial intelligence initiative called FGCS, for Fifth Generation Computer Systems.

Although Japan positioned this as an open, collaborative, global project, it so alarmed American authorities that the following year, the Reagan administration launched a billion dollar Strategic Computing Initiative designed to counter Japan’s lead. England followed suit with a similar but smaller-scale project called Alvey. For the next decade, scientists in Japan, the US, and Europe competed in what one pundit called “the race for the thinking machine.” Our current AI arms race, seemingly cutting-edge and of the moment, is actually a revival of this long-forgotten competition between Western and Eastern techno-powers.

The first AI race fizzled out in the early Nineties. For all, the money Japan spent, it failed to achieve the goal of a true thinking machine, or even commercialize any aspect of its project. As fears of Japan ebbed in the post-Bubble era, Britain mothballed Alvey in 1990, while America shuttered the SCI in 1993. So began what insiders call the “AI winter,” in which artificial intelligence was treated like a failed experiment.

But in fact, the Americans were implementing a great deal of AI tech behind the scenes. The Defense Department used findings from SCI to create “intelligent” logistics systems that played a key role in the first Gulf War, and autonomous land-navigation systems that are the direct predecessors of modern driverless cars.

In this greatly abbreviated story, you can see several reasons why Japan dropped the ball when it came to commercializing AI ahead of the West. One is that the FGCS was a consortium of bureaucrats and academics, never a great combination for coming up with hot products. It also seems to have suffered from infighting. In his 1983 book Inside the Robot Kingdom: Japan, Mechatronics, and the Coming Robotopia, Frederik Schodt quotes a former Tsukuba University robotics researcher who quickly gave up and moved to the United States out of frustration at the hierarchies and factionalism stalling the project. Some participants wanted to make standalone AI programs; others were interested in AI as a tool for understanding human cognition; still others wanted to integrate them into robots. This schism contributed to the project’s failure to come up with a killer app.

And “killer app” is particularly appropriate here. Where the Japanese saw dead ends, the Americans saw huge potential for defense applications. The long history of collaboration between the American military-industrial complex and Silicon Valley is no secret, with technologies initially utilized for defense eventually being repurposed for civilian uses or vice versa. It’s no coincidence that OpenAI recently named retired U.S. Army General Paul M. Nakasone to its board of directors.

So you might say Japan lost the first AI arms race because it wasn’t making arms at all. Japanese researchers were far more interested in “AI for social good” than they were in building weapons. Japan’s lack of a military-industrial complex (or at least one on any scale comparable to that of America or China) isn’t the only reason Japan failed to make a mark on post-ChatGPT AI (yet), but it’s a big factor.

None of this is to say Japan is out of the race for good. But hurdles remain. The country possesses a decided lack of “compute” compared to America and China. Supercomputers are needed to compile generative AI large language models. America and China have more than 150 each, many in the hands of private companies like Meta, Microsoft, and OpenAI. Japan has far fewer, and they’re mainly in the hands of governmental or academic institutions.

Still, it’s too early to count Japan out. As part of a “Generative AI Accelerator Challenge,” the government is helping fund a great number of startups and new machines, including an awesomely-named “Zeta-class” supercomputer, in an attempt to catch up with its rivals. And maybe it will. In fact, I wonder if the same society-centric focus that caused Japan to lose the AI arms race of the Eighties might prove an asset this time around. For “physical AI” — intelligent robots — are quickly becoming the Next Big Thing in the field, and it’s easy to imagine a future where Japanese robots take the world by storm. Who knows; maybe we’ll get our Doraemon after all.