Humans must get smarterWe were all naïve to think Korean Go master Lee Se-dol could easily beat AlphaGo, Google DeepMind’s artificial intelligence (AI) program devoted to master one of mankind’s oldest strategy games that has always been considered too subtle for a machine. But human intelligence cannot win against a self-evolving machine, observed Moon Byung-ro, a professor of computer science at Seoul National University.
AlphaGo runs on a system called deep neural networks — an expansive army of hardware and software that work in the same fashion as the web of neurons in the human brain. When playing against Lee, it was connected to a network of machines each equipped with graphics processing units with self-learning capacity on top of standard processors. The supercomputer even developed intuition, which has always been quintessentially human. If it had not been Lee, known to be one of the best and most cool-headed Go players, no other professional would have been able to chalk up a single win against a computer neuron network.
AlphaGo is unlikely to take up Chinese Go grandmaster Ke Jie as its next challenge. DeepMind’s co-founder Demis Hassabis indicated that computer situational game StarCraft could be a potentially next challenge if it can serve as a useful test of algorithmic research. The next level could be weather forecasting, according to AI scientists. At the end of the day, AlphaGo and its creators will be out to make money from their invention. Since Google has succeeded in getting the publicity it desired from the human-versus-machine tournament, the next track for its AI will probably be all about the money.
After building credentials through mastering complex games and even forecasting the weather, Google and DeepMind would be able to reap fruit from their investment by applying AI technology to various industries and services, from medical service and traffic to financial investment strategies. Tesla chief executive Elon Musk predicted the day would arrive when human drivers could be prohibited because self-driving cars will be deemed safer and more reliable than mere human intelligence behind the wheel.
Lee Jeong-dong, a Seoul National University professor and author of the best-selling book “Time of Accumulation,” found some of the behind-the-scenes moments as impressive as the match between AlphaGo and Lee. Lee, jubilant after his first win, proposed playing with black stones in the final match of the five-game series — instead of the white stones he played with in his winning fourth match — to be fair. Hassabis happily obliged.
Hassabis is the paid CEO of DeepMind, the company he helped create and then sold to Google. He had full authority to change the rules of the game even with Google co-founder Sergey Brin and Eric Schmidt, executive chairman of Google’s parent company Alphabet, present at the hotel for the event. The AI experiment is dubbed a moonshot project as it is as daring and complex as exploration of the moon, and the engineering chief should be in full command of the testbed, Professor Lee said.
Another factor that helps fuel an intangible future in the United States is the state of financial markets. In the innovation hotbed of Silicon Valley, there is a so-called 15-minute rule that can determine the fate of a start-up. Even the most bizarre idea is allowed 15 minutes in an open cafe to pitch to angel capitalists. Google bought the British AI company DeepMind with a staff of 50 and no revenue record for $575 million in 2014 in this way. Google’s market capitalization expanded by 58 trillion won ($50 billion) last week when its AI innovation was proven at the Go match in Seoul.
If AlphaGo had been an innovation by a Korean company, the scene would have been entirely different. Hassabis would have had to report to people above him in a hierarchy to decide whether to accept Lee’s proposal of tweaking the rule for the final match. Shareholders would have opposed the acquisition of DeepMind, suspecting a shady business deal. The local financial authority would have probed any stock price surge. The government cannot outright promote a single sector like AI because a failure could invite a parliamentary probe in the next administration.
The board game of Go, which has been neglected with the rise of online games, is gaining popularity following the machine-versus-human tournament. But a bigger push should be given to the science and technology field. President Park Geun-hye ordered the launch of a joint government and corporate task force on AI. But it needs cooperation from the entire country.
Super-smart software like AlphaGo cannot be created in a society where most of the talents prefer steady careers like law, medicine and working for the government. We must encourage the young to be adventurous, bold and fearless of failing.
We were not always this complacent. Even as a latecomer, Samsung invested heavily in semiconductors and the government in code vision multiple access (CDMA) technology. Patient and persistent investment for nearly a decade despite astronomical losses turned Korea into a powerhouse in memory chips and high-speed wireless and Internet infrastructure.
We cannot expect the same success with AI. If Korea’s elite schools and research institutes all join forces, they probably would be able to turn out a competitive AI. But we cannot win because we simply cannot breed creative minds like Hassabis or visionary companies like Google. What we should fear is not human-like Al, but its creators Hassabis and Google.
JoongAng Ilbo, March 21, Page 34
*The author is a senior editorial writer of the JoongAng Ilbo.