If Google Has "Solved" Go, Then What's Next for Artificial Intelligence?
Larry Alton
MAR 09, 2016 13:44 PM
A+ A A-

If Google Has "Solved" Go, Then What's Next for Artificial Intelligence?

by Larry Alton

The results are in. Google, one of the leading authorities in the modern world of artificial intelligence development, has created an algorithm capable of defeating a human expert in the game of Go. If you’re not familiar with Go, or the logistical hurdles of overcoming a human player with a machine, this may not seem very impressive to you. Suffice it to say, this is a massive breakthrough, and it’s opening up new doors for the development of artificial intelligence.

How It Happened

Go has long been considered one of the toughest challenges for an AI program to overcome, and thus hasn’t been seriously attempted until recent years. In fact, Google and Facebook were in the middle of a friendly competition to see who could get there first—though neither competitor expected such significant results so soon in the race.

Google’s project AlphaGo, a program based in AI company DeepMind, was put to the test last month. The modern European Go champion, Fan Hui, played against the program and lost five out of five games. AlphaGo was also able to defeat competing programs in 99.8 percent of all games played. The fact that the victory was so pronounced—not 3 out of 5, but 5 out of 5—makes the achievement all the more extraordinary. AlphaGo hasn’t yet played against a world champion, presumably one small step in ability higher than the European champion, but that match is scheduled in March, and programmers are optimistic about the results.

Why Go Is So Hard to Solve

Beating a human Go player may not seem significant or advanced, especially since IBM’s famous Deep Blue computer was able to beat chess grandmaster Garry Kasparov back in 1997. But chess and Go are very different games, and Go is far more challenging to program and understand.

First, Go has far more possible moves; played with two players laying stones on intersections of a 19x19 grid, there are far more possible board states in Go than in chess—in fact, there are more possible Go games than there are atoms in the universe. Second, it’s almost impossible to predict a series of responses to the end of a game, the way chess programs are developed. Most professional players have developed a kind of subjective “instinct” for the game, playing moves because they “feel right.” Games tend to last about 150 moves, with 250 viable choices per move, making each game extraordinarily complex.

A “brute force” algorithm, one that maps all potential choices then systematically chooses the best one, is therefore an impossible challenge. It would use up too much computing power and would take an inordinate amount of time to calculate.

How They Solved It

The solution was to avoid pre-programming the system to achieve victory, the way Deep Blue and other chess programs had been solved. Instead, engineers developed a system that could viably mimic human intuition and imagination—recognizing patterns, limiting possible choices only to the most viable, and only looking a handful of moves ahead in the game. This greatly reduces the burden on processors and enables the machine to “think like a human.”

Moreover, AlphaGo is capable of learning from experience. Programmed with a component of machine learning, the system could feasibly get better with every game it plays—and it runs through thousands to millions of simulations every day.

Why Does This Matter?

This is an historic, landmark breakthrough in the world of artificial intelligence, and has been identified as a theoretical impossibility by computer programmers in the past. But aside from that symbolic victory, there are some very practical benefits to solving the game of Go.

Go involves complex pattern recognition, sophisticated learning techniques, and an almost “intuitive” decision engine. Because of this, the AlphaGo system is theoretically not limited to solving the game of Go. Its algorithms can be adapted to serve other purposes with similarly complex parameters, such as diagnosing individual patients based on their histories and current symptoms, or working on city planning problems.

Deep learning is only going to get more advanced, especially in the next few years as momentum continues to build. We’re already solving mathematical hurdles that were thought to be impossible, so only time will tell how far we can go over the course of the next few decades. It’s possible that AlphaGo will end up losing to the world champion in March, but even if it does—this is still a monumental leap forward for AI.

[%= name %]
[%= createDate %]
[%= comment %]
Share this:
Please login to enter a comment:

Computing Now Blogs
Business Intelligence
by Keith Peterson
Cloud Computing
A Cloud Blog: by Irena Bojanova
The Clear Cloud: by STC Cloud Computing
Computing Careers: by Lori Cameron
Display Technologies
Enterprise Solutions
Enterprise Thinking: by Josh Greenbaum
Healthcare Technologies
The Doctor Is In: Dr. Keith W. Vrbicky
Heterogeneous Systems
Hot Topics
NealNotes: by Neal Leavitt
Industry Trends
The Robotics Report: by Jeff Debrosse
Internet Of Things
Sensing IoT: by Irena Bojanova