What AI innovation needs is a stage, not a blueprint

Lee Soo-hwa

The author is a research professor of the Big Data Innovation Convergence College at Seoul National University and the head of the AI Center at DLG Law Corporation.

Around the time the 2008 global financial crisis battered Wall Street, a quieter crisis was unfolding in Silicon Valley. Graphics chipmaker Nvidia saw its share price plunge by more than 80 percent after a laptop chip defect known as "Bumpgate." Yet investors were even more skeptical of something else: CEO Jensen Huang’s insistence on pushing CUDA, a general-purpose parallel computing platform introduced in 2006 whose value the market had yet to recognize.

At roughly the same time, a laboratory at the University of Toronto occupied the margins of academic research. The neural network group led by Prof. Geoffrey Hinton had fallen out of favor. Once celebrated in brain science, neural networks were increasingly dismissed as computationally expensive with little practical value. What kept the research alive was unrestricted funding from the Canadian Institute for Advanced Research, which imposed neither performance targets nor deadlines.

The two seemingly unrelated stories converged dramatically in 2012. Hinton’s students installed two consumer-grade GTX 580 gaming graphics cards in a bedroom computer and used CUDA to run neural networks that few researchers still believed in. The result was AlexNet, which overwhelmed the runner-up in that year’s ImageNet competition and transformed the field of computer vision. It marked the beginning of today’s deep learning revolution.

Had this history been written through a government planning process, the unconventional idea of classifying dog photos with two gaming graphics cards in a bedroom would likely never have survived the first stage of bureaucratic approval.

Transformative creativity emerges more often from environments that tolerate failure than from meticulously designed master plans. Yet policymakers now speak of winning the global AI race through state-led infrastructure projects worth tens of trillions of won and comprehensive national strategies. Innovation ecosystems, however, are difficult to cultivate inside tightly controlled greenhouses.

The government’s real role is not to predict the future but to lower barriers so that anyone can test unconventional ideas. Its task is to provide a broad sandbox where 50 million Korean citizens can build AI tools, fail repeatedly and try again, turning ordinary consumers into citizen developers. Just as unrestricted research funding protected that Toronto bedroom experiment, what the state should provide is not commands but a stage on which experimentation can flourish.

As the saying often attributed to Isaac Asimov goes, the words that precede great discoveries are rarely “Eureka!” but rather “That’s funny.” Politics need only protect those unexpected moments until they grow into breakthroughs. Nothing more is required.