You read today the first article in a series of articles on the term "Artificial Intelligence" and abbreviation (AI). This term can be translated into Arabic in several terms, including artificial intelligence or artificial intelligence. In this series we will rely on the term artificial intelligence.
So what are we going to talk about in this series, or rather what questions will we try to answer? We will try briefly to answer the following questions: How do search engines work like Google? How can machines distinguish sounds and understand natural languages? How can Facebook recognize faces? How is automatic translation? How do self-powered machines work? These are all complex problems in the real world, and the goal of artificial intelligence is to solve these problems across mathematical algorithms and tools.
In this series, we will identify the basic principles of artificial intelligence. These include, for example, topics such as machine learning, neural networks, deep education, the difference between supervised and unmonitored education, and many more.
But before we start, I will tell you a story that briefly explains where the IQ of artificial intelligence has reached, and we will try through this story to describe some of the concepts that we will be exposed to in the next article. The story is the story of the defeat of the world champion in the Go game on the "hands" of artificial intelligence.
How did IQ defeat the world champion in the Go game?
More than a year ago, on March 9, 2016, the confrontation between human beings and industrial intelligence was established to determine who is the real hero of the Go game: the human beings represent world champion Lee Sedol of South Korea. Artificial intelligence is an algorithm developed by Deep Mind Laboratories In Google _ called "Alfaago". The representative of human beings and the world champion is confident of winning the most confident and this is confirmed by his statements before the game, while the representative of artificial intelligence is thrilled win the European champion in the game and his morale is very high (of course artificial intelligence does not have morale, but the need for plot).
The frantic confrontation between the two sides began as expected, and lasted for more than three hours, after which artificial intelligence emerged victorious. A few days later, Alago returned the ball and won again and again before Seydoul won his first victory and then returned to Alago to win the fourth time. The final result ends four to one for industrial intelligence. Of course the game and what happened in it does not matter much, but what matters to us is the result that we came out and it surpasses artificial intelligence on natural intelligence in the unexpected as a game, "Go", and to understand the magnitude of scientific progress that is behind this huge result must explain what is Go, and why the superiority of artificial intelligence is so great.
Why is this a detailed victory for industrial intelligence?
Go is a very ancient Chinese game that was invented 2,500 years ago, in which two players compete for a pot of black and white stones, alternating each one in a 19x19 box. If a stone is placed, it is not permissible to move it after that. The final goal of the game is to surround the stones of each player with the stones of the opponent, so that they are surrounded by all sides. The game has very simple rules but nevertheless has a fantastic number of possibilities for the available movements up to 10365 possible movement. Imagine how large this number is compared to the number of potential movements in the chess game which is estimated at 10123 possible movement and then compared to the number of atoms in the universe as a whole Which is estimated at 1080 atoms only.
A huge number of potential movements makes the idea of building an algorithm capable of being surrounded is impossible, but it happened and the proof is Alfago and his last win. The win, which has been delayed for 20 years since another Deep Blue industrial intelligence won the world chess champion Garry Kasparov in 1996.
Here we have even noticed the importance of what happened, and why this victory took this big fuss, but what remains is more important how Alfago managed to do? But before we answer that question, let's go back together in a little while.
History or how the story of artificial intelligence began
The term artificial intelligence can be applied to all attempts to make computers smart. By smart computers, computers can be able to think, draw conclusions, and reach solutions that have not been pre-programmed. Despite many attempts before that date, artificial intelligence was not officially born until 1956, at a conference at Dartmouth University in New Hampshire, USA. To remind us that artificial intelligence was born, the father, John McCarthy, PhD in Mathematics, Dartmouth University Teacher and four others who can be considered the spiritual parents of industrial intelligence: Marvin Minsky, Oliver Silverridge, Ray Solomonov, and Richard Moore.
Other names such as Alan Newell, Herbert Simon, Claude Shannon, Arthur Samuel and Nathaniel Rochester. These and other personalities gave a variety of lectures on artificial intelligence, including topics such as future models of thinking, networks, learning and research, , Logic, language and cognition, games (especially chess), and human interactions with intelligent machines, such as personal robots.
Marvin Minsky of the Massachusetts Institute of Technology (MIT) and others attending the conference were very optimistic about the future of artificial intelligence, which led Minsky to say:
"Within a generation, the problem of building artificial intelligence will be largely solved."
But what Minsky predicted did not materialize, and artificial intelligence was not so simple. After several reports criticized the progress and development of artificial intelligence, government funding declined and interest in this area, and then resolved the first winter of industrial intelligence between the years 1974-1980.
Artificial intelligence was revived later in the 1980s when the British government began funding research again to compete with the Japanese, but this did not prevent another winter from coming, specifically from 1987 to 1993, coinciding with the collapse of the computer market and the decline in government funding .
A very important historic station and again in 1997, the station where Deep Blue beat chess world champion Gandhi Kasparov, was the first superiority of industrial intelligence on natural intelligence in a highly complex game. Another triumph of industrial intelligence took place 14 years later. A computer with an industrial intelligence called Watson won the 2011 Jeopardy Information Contest. This win required Watson to understand English and then answer many questions and general information. The rest of the story you know.
Let's go back to Alago
Note: If you encounter some strange terms, it is not a final problem. We are still in the forefront and give a glimpse of general topics that we will discuss in the coming articles of this series.
How did Alfajo do it?
From the best upstream to answer this question, I mean the source Deep Deep Mind and his paper published, explaining how Alfajo algorithm works.
The paper tells us that AlfaGo's algorithm relies on two neural networks, the first called the value network, and the second is called the network plan. The plan's network is tasked with knowing the potential moves at each stage of the game, after which the value network evaluates all possible moves according to the probability of defeating the opponent.
The paper also tells us about the game tree, which consists of branches and nodes and includes all available movements in the whole game from any position, each branch in this tree is a course of play that can lead to loss or win, and since we have seen that the movements available number to number For the astronomer, the best way for AlfaGo algorithm is to randomly select some tree branches that seem to win and evaluate them to the end. The value of the different nodes in the game tree is given, and these values are determined so that the contract leads to victory. Repeat the previous steps repeatedly called the tree of research "Monte Carlo"
Neural networks and the Monte Carlo algorithm were not the secret behind Alphago's victory, but rather his ability to learn. The algorithm was trained on 160,000 real games, thus enabling it to play millions of games against itself, the core algorithm of which no biological organism can divide its mind. For two parts and play against the same.
While training the algorithm on real games, the algorithm learned from the actual movement chosen by the professional player, and this technique in education is called supervised education. Finally deep learning mechanisms have been applied, to make the network more likely to choose the best movement next time. The algorithm is therefore able to predict optimal movements in new games that have not been previously trained.
