Why it matters that AI is better than humans at games like Jeopardy
– Mastering the complexities of natural language is one of the biggest benefits of AI playing human games
– Playing games helps AI systems learn strategy, negotiation, and the ability to predict what humans will do – which can be applied to business problems
– Games like Jeopardy and AlphaGo can be very high stakes, just like business negotiations, military strategy, cybersecurity, finance and medical treatment
For many people, the first time they ever heard about artificial intelligence and IBM Watson was when it played Jeopardy! on television against human opponents. And not just any humans, but Jeopardy! champions Brad Rutter and Ken Jennings. While Watson made a few mistakes on its way to victory, it cemented its reputation well enough that many articles about Watson still describe it as the artificial intelligence system that played Jeopardy!
But even before Watson competed on Jeopardy!, AI systems also learned games, ranging from tic-tac-toe to chess. You may recall that in 1997, IBM’s Deep Blue beat the world’s chess champion Garry Kasparov. And since Jeopardy!, AI systems like Watson have continued to learn to play other games, ranging from the ancient game of Go to Texas Hold’em poker.
Why AI playing human games is valuable to us all
It might seem trivial – wasteful, even. Why spend an estimated $1 billion teaching a computer to play Jeopardy or AlphaGo? But teaching an AI system like Watson a game isn’t just child’s play, says Dr. Gerald Tesauro, principal research staff member at IBM’s TJ Watson Research Center and an expert in using AI to play games.
“Games can embody a lot of complexity, if you think about games like chess and AlphaGo,” Tesauro says. “The number of moves is rather enormous. It’s also very clear-cut, extremely well-defined, what the best move is going to be. That was appealing historically when we were getting going with AI.”
But as AI systems started regularly beating humans in that style of game, researchers needed new challenges. And as it turns out, learning how to play games can be even more valuable when it comes to acquiring natural language, or the way that humans put words together, Tesauro says. “A lot has to do with ambiguity or things being ill-defined,” he says. “Mastering the complexities of natural language is one of the big challenges.” That was one of the big challenges in Jeopardy!: the puns, double meanings, and other wordplay involved.
Transfer learning and reinforcement learning
Skills learned by playing games can also be applied to other areas, which is called “transfer learning,” Tesauro says. “In transfer learning, you initially train your learner in some sort of environment. It learns expertise in how to do well in that domain, and takes that learning from the original domain and tackles some other domain that may be more important.”
A similar concept is “reinforcement learning,” where an algorithm learns to reward itself for actions that help it attain a goal, and punish itself for actions that do not, says Chris Nicholson, CEO of the AI company Skymind. “For example, if you have a drone, one goal for the drone is to keep it aloft. So instead of maximizing the number of points it wins, you maximize the number of seconds it stays in the air without hitting anything. The input the algorithm would need could come from cameras attached to the drone, so it can avoid tree branches, as well as an internal gyroscope, so that it can balance itself in the air.”
So playing games can also help AI systems learn strategy, negotiation, and, in particular, predicting what humans will do – all of which can be applied to business problems as well. “If you’re a business, you have a magic number that you’re trying to maximize, too, like profit,” Nicholson says. “So that could be the equivalent of points in a real-life game. And the choices you face might be which products to order for your store, and when to order them, to sell the most and not keep useless inventory. In this case, you would have many more choices to make because a lot of products exist, so it’s a more complex problem.”
Numerous benefits of AI that has strategic reasoning abilities
“These games can be very high stakes, like business-to-business negotiations, military strategy planning, cybersecurity, finance, medical treatment planning of certain kinds,” Tuomas Sandholm, a computer science professor at Carnegie Mellon University who is teaching computers to play poker, tells Time. “These are really for a host of applications – really, any situation that can be modeled theoretically as a game. Now that we’ve shown that the best AI’s ability to do strategic reasoning in an imperfect information setting has surpassed that of the best humans, there’s really a strong reason for companies to start using this kind of AI support in their interactions.”
For example, poker is a great way to teach one skill in particular: Bluffing. “The computer can’t win at poker if it can’t bluff,” Frank Pfenning, head of the Computer Science Department in CMU’s School of Computer Science, says in a statement about CMU’s poker playing system. “Developing an AI that can do that successfully is a tremendous step forward scientifically and has numerous applications. Imagine that your smartphone will someday be able to negotiate the best price on a new car for you. That’s just the beginning.”
Finally, learning how to play games also puts AI systems into a context that typical people can relate to, and helps them both understand what AI systems can do and gets them more comfortable with AI in general. “IBM knew from the Kasparov vs. Deep Blue days that we’re all suckers for the ‘man vs. machine’ trope, going back to John Henry’s mythical race against the steam-powered hammer,” writes The Atlantic. “It certainly makes for a better storyline than, say, ‘Check out the latest incremental innovations that Natural Language Processing researchers are making in the field of question-answering!'”
“Games bring audience to AI,” says Stephen Baker, author of Final Jeopardy: The Story of Watson, the Computer That Will Transform Our World. “This is important for funding, and also to inspire students to pursue careers in the math and science fields that lead into AI. In the case of Watson, it put AI before a national audience, and gave the world a preview of what AI could and will be.”