Next Stage in AI - AlphaGo

Just finished watching a documentary on Netflix but now on YouTube, on how a software developing company in the UK was able to put together a team that that was able to build a software program sponsored by Google that could beat the world's best Go player. On February 10, 1996, Deep Blue by IBM was able to beat Garry Kasparov who was at that time the world's best chess player in the first game of a six-game match—the first time a computer had ever beat a human in a formal chess game. Now, computers beat world champions regularly, that it just formality and they are winning every game if it properly tuned.

Go is regarded as a more complex game than chess,  the number of legal board positions in Go has been calculated to be approximately 2 × 10¹⁷⁰, which is said to be far greater than the total number of atoms in the universe. Go is the oldest continuous board game and described as the most complex game ever designed by man. Chess is reported to have a mere 5 × 10⁵².in comparison. Deep Blue was able to beat Kasparov by the sheer computing power it was able to evaluate 200 million positions per second. In chess, the computer was programmed with professional chess players and had various scenarios calculated and permutations or outcomes. Go is an abstract strategy board game for two players, in which the aim is to surround more territory than the opponent. The playing pieces are called stones. One player uses the white stones and the other, black. The boards are of different types of sizes but the 19 X 19 is the standard international style. Players take turns, placing one of their stones on a vacant point at each turn, with Black playing first. Stones are placed on any unoccupied intersections of the lines rather than in the squares they want to and once played stones are not moved. However, they may be captured, in which case they are removed from the board, and kept by the capturing player as prisoners. At the end of the game, the score is the number of empty points enclosed by a player's stones, plus the total number of prisoners captured by that player

But in the case of AlphaGo, it involved teaching the machine and then having it play better and better versions of itself till the best performance, comes out on top. Each time is collecting data and improving, i.e. learning. By self-learning and reinforced learning millions of times, you will have a better version of yourself. By using neural networks, this is an artificial intelligence-based on how the brain, which consists of neurons operates. It is a form of machine learning and based on neurons, i.e. a thing that can hold a charge or an item, i.e. a number or character. The software has neural layers and uses big data.

Fan Hui

The group decided to test the software by inviting the current European Go Champion Fan Hui who lived in France and was a professional Go player, to come over to London and play AlphaGo. In October 2015, he came over for a series of 5 games and the machine beat him in all five games. The team then decided to up the ante and played the best Go player in the world as at that time, Lee Sedol from South Korea, the European Campion was a 2p while Lee Sedol was a 9p in the ranking. As none of the programmers were professional Go players, they needed expert help in the way the machine was playing and decided to call Fan Hui over to help in its development as a consultant. Hence, the invitation was sent, and a date was agreed upon. Lee Sedol happens to be the greatest player over the latest decade and has won up to 18 international awards and is at the top of the game. After playing a series of games Fan Hui, detected a vulnerability in AlphaGo which could easily be exploited but this would require a significant overhaul in the system. 

Lee Sedol

The games were set in March 2016, from the social media contents it seems that people believed that Lee Sedol would win as it was more complicated than chess and had a lot of permutations and possibilities that it was inconceivable that a computer could deal with. The general idea was that there were so many different types of moves and in different directions that a computer processes could not deal with it and it would just burn out, or it would be too slow on time that the human would win on time. Yes, that is another thing, they decided to use Chinese rules, and it is timed. Fan Hui who was the European champion, which AlphaGo had beaten in 5 games, was thought to have spent so much time in the West and hence not a good player. Even though Fan Hui was a 2p ranking, which is relatively high, he was thought not to be much of a challenge since he was beaten so easily.

The tournament took place at the Four Seasons Hotel in Seoul, South Korea, and there was a press bonanza, As Go is very popular in South Korea and Lee Sedol is an international player with a prize of $1 million to go to the winner. Sixty million people alone were watching it in China, and the first game went to the AlphaGo. The team was able to see that that AlphaGo had calculated 50 - 60 moves ahead mapping out each scenario. Its actions were based on had a policy network (general rules of the game), value network (i.e. probability), and tree search (different variation). By the second win, there was a general sadness in the room, and after the 3rd win by AlphaGo, there was a general depressive mood. During this game, it made a new move that it was not taught or did not have experience of, it was a new approach to the game that after game analysis proved it had calculated a win. An average human player would be trying to get more area, but AlphaGo was a program, programmed only to win and it seems that winning just by one or two points is still consider a win. AlphaGo was learning, the pain of lee Sedol was palpable, and everyone could see that this great player was struggling and it was as he was in deep anguish and distress. The whole room was silent, you could hear a pin drop, and there was a humility suddenly in the space of journalist, the fanfare and arrogance before the first game had disappeared and was replaced by fear and frustration. People began to look at a 5 - 0 result at the end, as Lee Sedol was looking at the person playing the moves of AlphaGo face, as if he was playing the human, from habit to detect the person's mental state. But he was not getting any response, or just a cold reply, an emotionless person was filling in the moves to the computer and giving the output. At this stage, the journalist's questions began to focus on control and restriction of the technology. They saw a terminator end world scenario. But Lee Sedol persisted, and in the middle of the fourth game, AlphaGo's move was out of the ordinary and did sometimes totally abnormal again, from there it was downhill. Lee won the fourth game after the AlphaGo resigned, it had calculated a 95% probability that it would lose.
 
News spread fast, and the press room went wild it was reported that in China, Japan, and South Korea people were dancing in the streets. The press coverage for the final game was enormous, and the hotel lobby was full, there was more press coverage than the previous games. Lee Sobel was smiling more than ever, and the atmosphere had lightened as he had found the bug in the system, its weakness or vulnerability after which he would be a successive win for humans. He was playing for the whole human race against the machines. Unfortunately, the final game was won by the program despite the newly founded enthusiasm, but he left with his head high.

The film did raise essential points about control of technology and raised issues involved uncontrolled use of AI. General protocols were suggested to be presented by the educating bodies. But testing the AI in this format allowed the development team to understand neural AI in different content. The way it solves problems and expands our understanding of complex issues. Both Fan Hui and Lee Sedol became better players as they understood the game Go at a more in-depth content, and things they took for granted prove significant. AI will help us understand the various complexities of our world and could help us solve our problems.