I, Terminator

Artificial intelligence is very popular nowadays. And for very good reasons. There is a never seen before advancement in AI technologies and even our tiny cells start to use artificial intelligence to better serve us. The British company DeepMind, now part of Google, created a program capable of self learning. It doesn’t seem much, we heard this kind of claim many times before. But I believe this time it is different... I’m passionate about board games, chess and especially Go, an ancient asian game using black and white round pieces on a 19 by 19 table. Simple, but very hard to play. It is very strategic, somewhat like 5 interrelated chess matches at the same time, 4 on corners and the fifth in the center. People use to say that probably a 21 by 21 table surpasses the human brain capacity to control the board. When I started playing, chess programs were already better than humans, but the general opinion was that this will be impossible for Go. Why? Simply because chess programs were following algorithms and were calculating thousands and thousands of variations. There was some input such as known openings and previously played matches, but the rest was something we call in technology brute force - try everything possible. And everything possible in Go represent actually more than the number of atoms in the Universe. DeepMind created something different. An intelligence. Yes there are still algorithms, but mostly algorithms allowing the program to self-learn and improve. A Go match was organised by Google between the program DeepMind created and Lee Sedol, a top professional player, not currently number one in the world but a legend at the top for many year. He’s compared with Roger Federer in tennis. The result was shocking: 4 to 1 for the Google DeepMind program, called AlphaGo. Everybody started to talk about the match and soon after DeepMind created a new version, even more autonomous, called Master, that won all matches against top professional Go players. And then they created yet another version, called AlphaZero, that started from nothing. The game rules were input into the program and than AlphaZero played millions of games against itself and learned what to do and how to win. Without any input such as known variations and techniques. AlphaZero won all matches again its predecessor, Master. Later on, a version of the program called AlphaZero was presented the rules of chess, and then trained against itself for only four hours, in order to compete against the strongest chess program, Stockfish. Chess players are rated using a point system called ELO. The best humain players are rated around 2800 ELO, the best program around 3400. AlphaZero finished the 100 games match against Stockfish with a score of 28 wins, 72 draws and 0 losses. So I guess Zero is a good name for the AI. All this with only 4 hours of self-training and no other input. This is a spectacular achievement, but maybe also frightening. It reminds me of Terminators that decided at some point that humans are causing problems, being almost like computer bugs, and need to be eliminated. I think advanced AI can become a risk for the humanity. Does this mean I think we should stop advancing AI? No, impossible, not a solution. But maybe we should achieve a new level of AI, less artificial, more human, almost having feelings like the robot from the famous Asimov book, I, Robot. We may tend to state that this will never be possible, but what we have today was qualified as impossible yesterday, wasn’t it? How are we going to apply this to different areas, such as intelligent communications powered by AI? I don’t know, but it is something I’m very passionate about. I call it creative AI and it might just happen in the near future.