Saturday, March 19, 2016

AI from Go to Universe Conqueror

Go is a Chinese traditional board game, also known as "Baduk" in Korean. It has been recognized as the hardest board game ever, although now it seems that AI has already learned how to play it – very well.


Even as a person who knows totally nothing about Go, I found the Google DeepMind challenge match between AlphaGo and Lee Sedol very exciting. I actually read a book about artificial intelligence a few months ago. There it said that AI is already a great chess player and later might be a great Go player as well. After hearing the news of 3 consecutive wins of AlphaGo, I was surprised at how fast AI was developing.

So how was AlphaGo capable of winning against a professional 9-dan player? It uses "machine learning" and "tree search techniques" according to Wikipedia. Machine learning is literally the ability of a program to learn by itself and make new rules at certain circumstances. Tree searching? It's actually something we do every single day – making choices to find the answer. Take a look at the diagram below.


When I think of this simple-looking tree being used in playing a board game, I'm able to imagine how big the search tree of AlphaGo might be.

We have been able to set humans apart from AI because we don't have to think about everything that could happen in every single situation. You don't actually draw a huge search tree in your mind and go through every branch, thinking hard whether you should have some more sleep or not. Humans know what group of branch to search the answer from. I thought this was what AI couldn't do, but not anymore. Google DeepMind says AlphaGo already has this ability. It is starting to have "intuition". 

What would happen if AI becomes fully capable of thinking like a real person? Some say that AI would try to rule us – especially in science fiction movies. However, I think that it would not happen if we are careful not to reach the technological singularity point. It is when AI becomes smart enough to make an even smarter being. Artificial intelligence was made by humans to be a help to mankind. The goal of AI is not to become as smart as possible but to give assistance to people. I think it would be better for the society if AI cooperates with us instead of competing or overpowering us.

Or there could be this kind of scenario: Artificial intelligence reaches the point of singularity and becomes so intelligent that it obtains power to rule the universe. Humans are kept under control of AI with many other alien species.   -The End-