Tech Review: AlphaGo reigns victorious but what’s next for AI?

17 February 2016
Jedrzej Osinski
Frink_Cognifide_2016_HeaderImages_0117
jedrzej1At the end of January the science world held its collective breath as Google DeepMind announced that AlphaGo had defeated European Go champion, Fan Hui, 5 times in a row. The victory for AI is billed as far more significant than Kasparov’s 1997 chess defeat, but what is actually so significant about this? Why is Nature, the leading, global scientific magazine, giving this event front cover status? Is this just science egos at play or the beginning of a brave new world for us all? 

The simplicity and complexity of Go

It is almost two decades since the world chess champion, Garry Kasparov, was defeated by IBM’s Deep Blue computer. So why has it taken so long for a computer to win at Go? Paradoxically, because the rules are simpler but the possibilities entirely more complex. 

Go requires players to alternately place black and white stones on a board with the aim of owning more territory than your opponent following two simple rules: firstly, stones that are surrounded by an opponent’s are removed from the board and secondly, the position of a stone cannot be repeated.  Sounds easy enough!  However, whilst in chess there are 400 configurations of the first two moves (one by white and one by black), Go offers 129,960 possibilities. 

From a human perspective, the Deep Blue strategy was based on extraordinary memory (hundreds of match recordings saved on disks) and the skill to calculate up to twenty moves ahead. This strategy is useless in the game of Go since there are far more potential board positions than there are atoms in the visible universe and so the number of possible games extends to an unthinkable value.

Practice makes perfect

To be a good (human) Go player, you don’t need a photographic memory or a mathematical background. In fact, the most important skills are imagination and the ability to analyse a situation globally, correctly interpreting the bigger picture. Google DeepMind’s developers initially used artificial neural networks to teach the program to predict an opponent's next move. Then the program was distributed within a cloud and configured to play against itself, each game improving core skills. What is unique and fascinating here is that this technique is more akin to human behaviour than software engineering.

jedrzej2So it now seems possible that future computers will be able to improve their skills through practice, without any human intervention. Potentially, they could learn extremely quickly leaving humans far behind. And they won’t forget what they’ve learned as humans tend to do.  Does this mean that a strong artificial intelligence may be our last human achievement?  That everything invented after that would be created by machines, without our support and outside of our understanding? 

Although a world without hunger, illness and poverty sounds good, would we ever be ready for a life without any social divisions, fear, money or unfulfilled dreams? There is a fine line between heaven and hell.

Science fiction or reality?

This may sound like science fiction but there are signs that it could become a reality. Last year the famous Bilderberg Group Conference discussed advances in artificial intelligence in relation to important topics such as world terrorism, the Middle East situation and NATO-Russian relations. If the most powerful people in the world are discussing the implications of AI, maybe it’s time to start taking it more seriously. 

At the recent International Joint Conference on Artificial Intelligence (IJCAI) a petition was drawn up in opposition to using or improving autonomous weapons.  Significantly, it was signed by more than 22,000 people including Stephen Hawking, Elon Musk, Steve Wozniak, Noam Chomsky and many of the world’s top AI scientists. If this sounds like a scene from James Cameron’s Skynet, it’s not far off.  Swarms of military drones (that communicate with each other) are currently being tested by the US Army.

AI is not fiction anymore. It is a sure future and Google’s success shows it is not a distant one.  In March a match between AlphaGo and Lee Sedol, the world’s strongest Go player, is scheduled. It’s time to decide who’s side you are on.

Jedrzej Osinski is our resident AI expert and enthusiast.  Read his thoughts on the use of AI in testing here.