Computers make giant leap

Lee Se-Dol had never felt so helpless before. His opponent was getting better with each game, and coolly countering all of his best moves. It was a surreal experience.

Lee is a legend of his time, the best player on the planet, and one of the best players in recorded history.

His playing style was described as creative, intuitive, wild and unorthodox. But he had never faced an opponent like this before.

He was sure he would win at least four games – after all, the opponent was just two years old, while he had been playing the game for pretty much his 33 years.

But here they were, three games later, and Lee had lost all. How had it come to this?

Go is one of the most complex strategy games in the world. The game is played on a 19×19 grid with black and white stones, and the basic rules are fairly simple: choose a colour, move your stones across the board, and try to corner all your opponent’s stones until he has no moves left. But beyond that, the game is complex, largely due to the fact that there are more possible legal moves in a game of Go than there are atoms in the known universe.

Although logic plays a part, the game relies heavily on creativity and intuition to succeed. Unlike in chess, there are no proven strategies to win a game. Each game is different, and players have to rely on their “gut feeling” to succeed. This made it impossible to programme a computer to play Go.

In a game such as chess, programmers would simply input the rules for the game, as well as strategies, and the computer would use this information to play. In a typical game, the computer would play by mapping out every possible move from the current state of the board, and every subsequent move based on that one.

But that was impossible in a game like Go, where there are an incomprehensible number of possible moves that are beyond even the most powerful computers to calculate.

Clearly, traditional programming techniques are out of the question. Yet, the temptation to create a Go-playing algorithm was too great to resist.

Because Go is so much more complex than chess, if such an algorithm could be created, it would be a huge step forward for the field of artificial intelligence.

In 2014, scientists at the Google DeepMind, a company focused on artificial intelligence, began work on a Go-playing algorithm called AlphaGo. The approach they used was what is known as “unsupervised machine learning”.

What that meant was that they would not programme the computer at all, nor teach it how to play. Instead, they would show the computer a large number of Go games, and let it figure the rules and techniques out by itself.

The computer was shown around 30million Go games and it learned to play the game. Over the next few months, it continued learning, and even began playing against itself.

Although AlphaGo was making progress, all predictions were that it would be some time before it be- came good enough to play against a human.

In an article on Wired, Professor Alan Levinovitz predicted that it would take at least a decade before a computer Go champion emerged.

By late 2015, the DeepMind team decided it was time to test AlphaGo against a human. The player chosen was Fan Hui, the European Go champion. A match was arranged, and AlphaGo beat him five games to nil. This was a momentous occasion: for the first time in history, a machine had beaten a human in a game that relied not just on logic, but on creativity and intuition.

When the news broke, the world was abuzz with excitement.

It wasn’t just that a computer had beaten a human at a game: the long-term implications were astounding.

Were computers now becoming creative and intuitive?

AlphaGo had proven to be too good for Fan Hui, but people wanted to know how good it really was.

Fan may have been the European champion, but he was rated only 633 worldwide.

How would AlphaGo perform against the world champion?

In March 2016, a game was arranged between AlphaGo and the world’s number one player, Lee Se-Dol.

By the third game, things were not looking good for Lee: he had lost all three games. He, nonetheless, gathered his wits and came back to win the fourth game. It wasn’t enough to win the overall contest, but at least it was a win. But AlphaGo wouldn’t stop there. It won the fifth game, giving it a 4-1 victory against the world champion.

In essence, the new world champion Go player is not a human, but a machine.

Predictions are that we will see more technological advancements in the next decade than we saw in the previous century.

People wonder how this accelerated pace of advancement will be possible, until they realise that it will be driven by machines. A few years ago, this idea would have been met with scepticism because innovation and techno-logical advancement require higher-order thinking abilities, such as creativity, intuition and problem-solving abilities – things that are strictly human traits.

AlphaGo has blasted this notion out of the water, and has shown us that computers can learn, and are capable, to some degree, of creativity and intuition.

Share This Post
Have your say!

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

This site uses Akismet to reduce spam. Learn how your comment data is processed.