DeepMind’s victory is one small step for computers, one giant leap for artificial intelligence



In a historic turn of events, Google DeepMind’s artificial-intelligence program AlphaGo won four out of five matches against South Korean professional Go player Lee Sedol in March 2016. This is the first time a computer program defeated a human competitor in this strategic game of skill – a tremendous milestone in the development of artificial intelligence (AI) to date.

Observers previously thought this feat would be at least a decade away. AI experts told CAIFU they were not surprised with AlphaGo’s success, as they believe machines and robots will eventually surpass human brainpower.

“The AlphaGo match is a wonderful demonstration … of how computers can learn and make computational decisions in very powerful way,” Fei-Fei Li, Ph.D., associate professor of computer science at Stanford University and director of Stanford’s Artificial Intelligence Lab and Stanford’s Vision Lab (李飞飞) told CAIFU Wednesday, April 13, 2016. “…Go is revered as one of the most complex board games. It requires long period of training for humans, and has practically infinite possibilities when being played. AlphaGo’s success is really a great show of AI’s power.”

Originating in China more than 2,500 years ago, Go is considered one of the four essential arts required of any true Chinese scholar, Google DeepMind’s founder Demis Hassabis wrote on Google’s blog on Monday, Jan. 27, 2016. Played by more than 40 million people worldwide, two contestants place black and white stones on a square grid, trying to capture the opponent’s stones while aiming to seize the most territory.

Wei-Min Shen, Ph.D., associate professor of computer science at the University of Southern California in Los Angeles and director of the Polymorphic Robotics Laboratory, told CAIFU Tuesday, April 19, 2016 AlphaGo’s victory is a great achievement for AI, as the computer learned how to play the strategic game by itself.

“Go is the most difficult game humans have invented,” he continued. In Go, there are a staggering number of possibilities – even more than the number of atoms in the universe.

Prior to the match, AlphaGo played several thousand games in the cloud and trained itself using human game records, Shen explained. The computer program only took several months to master and learn Go, despite the huge number of possibilities.

AlphaGo thinks and learns through a process known as reinforcement learning – a development similar to the way humans learn, Shen noted. This happens as the computer plays itself. As the computer competes in more and more matches in the cloud, it adjusts its own thinking based on what it learns.

After the match in South Korea, Google CEO Sundar Pichai visited one of China’s top Go training schools to develop his understanding of the game, and to gain more insight into the country, the China Daily reported March 31, 2016.

Scientists from the China Computer Go team have expressed interest in challenging the computer in a future competition sometime in 2016.

“AlphaGo can win again against a top-level Chinese player,” Shen said. “The longer AlphaGo competes, the better it gets.”

Stanford’s Li expressed a different view on AlphaGo’s next challenge. “I don’t think AlphaGo’s success has much impact for human players of Go,” she explained. “Humans have created so many machines that can run faster than people. We still have competitive running sports and the Olympics. Humans play Go for different reasons — to exercise their mind, or for social, cultural and philosophical [fulfillment].”

“Perhaps the biggest impact [of AlphaGo’s win] is that human players now have a new tool to train them, just like how technology and modern equipment have helped to train athletes in other sports,” she added.



Will AI Pass Go?

For Google DeepMind – the London-based group who conceived the AlphaGo software – beating Go is only the beginning of its journey bringing AI into the mainstream. According to Li, AI is a general term for technologies of smart machines that can take data and learn from their patterns, and then make predictions, decision or actions.

“We are thrilled to have mastered Go and thus achieved one of the grand challenges of AI,” Hassabis wrote on Google’s blog. “While games are the perfect platform for developing and testing AI algorithms quickly and efficiently, ultimately we want to apply these techniques to important real-world problems.”

“Our hope is that one day they could be extended to help us address some of society’s toughest and most pressing problems, from climate modelling to complex disease analysis,” he continued. “We’re excited to see what we can use this technology to tackle next.”

If you look around, AI technologies are already everywhere, Li pointed out. Asking Siri on your iPhone for a dinner recommendation is AI at work, as well as Amazon’s product recommendation system. New cars have lane detection systems, designed to warn a driver when the vehicle begins to move out of its lane. Around the home, the iRobot Roomba vacuum cleaner memorizes where the furniture is to gather dust around it.

“AI is gradually creeping into everyday things,” Shen added. Getting a call from a credit card because they detected unusual activity is another example of AI, he noted.

In China, WeChat and Weibo users are currently having conversations with Xiaoice, a chatbot built by Microsoft researchers in 2015.

According to Yongdong Wang, managing director of Microsoft’s Applications & Services Group in East Asia, millions of people consult with Xiaoice on a daily basis. She responds with human-like answers, questions and thoughts. For instance, if a Microsoft user sends her a picture of a broken ankle, Xiaoice will then reply with a question asking about how much pain the injury caused.

“Xiaoice is teaching us what makes a relationship feel human,” Wang wrote in web magazine Nautilus in February 2016. “She is hinting at a new goal for artificial intelligence: not just analyzing databases and driving cars, but making people happier.”

That is where the ethical and legal challenges enter the picture. At some point, will robot intelligence overtake humans in intellectual ability? Shen said he thinks it is too early to predict if computers can replace human brains, but he asserted that AI progress would benefit society as whole. “Computers can do many things better than humans, like forklifts – those machines are more efficient than human muscle. [AI] is an amplifier for human activity.”

Li agreed. “Computers already surpass human’s ability in many things,” she added. “Skype can translate far more languages than most people can. Some face recognition systems can identify far more people than humans do.”

Both experts agree that AI’s potential is limitless. By 2025, they forecast self-driving cars would be on the road, while robot caregivers will perform domestic tasks like cooking, cleaning and babysitting.

“Some people say that AI is even worse than the nuclear bomb; that it’s a threat to humankind,” Shen said. “I don’t agree with that. AI will have a bright future.”

Li said she believes AI is the driving force behind the next industrial revolution. “It has the power to increase productivity, health and safety of our human lives and society,” Li concluded. “We’ll have robots in manufacturing factories, AI to assist medical diagnostics, smart homes and smart Internet of Things (physical objects like devices and vehicles embedded with electronics and software) to improve home life.”