In the first part of the 20th century, science fictions have familiarised the world with Artificial Intelligence, where they would often be represented as evil metallic creatures that would take over the world. So today we shall cover subjects about AI, such as a brief history of it, how they work, what types they are, and many others.
First of all, what is AI? AI stands for Artificial Intelligence, they are machines that process information without help from humans and synthesise or output a result. They can be used in many different domains.
Now we shall cover a brief history of these computers. There are many debates on when the first AI was invented, but most people say that it was in 1943 by Alan Turing, during WW2 called the Bombe. It is estimated that without it, there would have been almost 15 million more deaths and the war would have been almost 2 years longer. It did so by helping the Allies crack the enigma code, a German code that was extremely complex to break as it changed every day. Before the 1960’s, the development of AI had huge setbacks: those were that for a computer, you needed a lot of money: maintaining one at the time cost up to 200 000£, and they were only able to execute commands, and not store them. But from 1957 to 1974, research in AI flourished, as computers could store more information and became much more accessible, and in 1960 the first chatbot ever ELIZA was developed by MIT professor Joseph Weizenbaum. In the 1980’s, deep learning was invented, a technique allowing a computer to actually learn. During the 1990’s and 2000’s, an enormous amount of landmarks were achieved: in 1997, Gary Kasparov, chess Grand Master, was beaten by “Deep Blue”, a chess playing computer program created with a deep learning algorithm.
Now that we had a very brief overview of the history of Artificial Intelligence, we shall look at what types they are, as some are more adapted for certain situations than others. The 3 most general types of AI are ANI, AGI, and ASI. ANI is an AI that has a narrow range of abilities, it is designed for doing only a specific task, for example facial recognition or driving an automatic car, it is learning and in reason of this, this concept is often called machine learning. An AGI is a “general” AI, meaning that it has generalised human cognitive abilities (for example memory, visual processing, etc…), so that faced with unfamiliar task or problem, it can solve it. It is as smart or almost as smart as a human and this is why it is called machine intelligence. ASI is an AI that is much smarter than the best human brains in almost any field. Quite obviously, ANI is the most common type of AI out there. And now, we shall explore the different types of learning in ANI AI’s. The 4 most common types are Unsupervised Learning, Supervised Learning, Semi-Supervised Learning (which we will not look at today), and Reinforced Learning. Unsupervised Learning is when the machine has a limited amount of things to do (in most games there is a limited amount of moves to play) and the machines see which moves are best or winning by trying it out randomly or with a certain pattern. An example of an unsupervised learning machine is Deep Blue, which we talked about previously. Supervised Learning is when the programmers of the machine feed it images or others and tell it what it corresponds to, and then they give it images that are not labelled and the AI tries to say what it corresponds to. A famous example of a machine that uses this kind of learning is Dall • E. Reinforced Learning is the art of making a decision. It attributes each decision it can take a certain number of points (a reward) and chooses the path that gives it the most points. Tesla uses this type of learning in their cars.
Now we shall attempt to create a well known AI that can be built with or without supervision called “Hexapawn”. Hexapawn was introduced by Martin Gardner in March 1962 in the journal Scientific American. Hexapawn is a game based on chess: each player has 3 pawns on a board of 3x3 (as you can see in the image below). Here are its rules:
A pawn in Hexapawn moves just like in chess: it moves forward except if there is already a pawn in front of it. A
pawn can move diagonally if it eats an opponent pawn. In hexapawn, the human always plays white and white always begins.
There are 3 ways to win:
You win if you get one of your pawns to the other side of the board
You win if you eat all of your opponent’s pawns
You win if you leave your opponent without a possible move.
Obviously in all of these cases your opponent is the AI we are going to build. First of all we shall do so that the AI knows all the possible moves, and each time it will play, it will choose a random move to make among all the possible moves it can perform. If it wins, it does nothing, but if it loses, it eliminates the last move it made among all the moves it can perform. If this sounds confusing right now, it will become clearer with an example.
And here are all the possible moves the AI will be able to make:
If as a first move we advanced our left pawn, and then the AI, choosing a random move, advanced its middle pawn (this is the first possibility you can see on the image). Well then we would just advance our pawn at the right and would win as the AI would not be able to perform any possible move. Seeing it has lost the AI will eliminate the move it made from its possible moves. This means that if the next game we advance our left pawn just as in the previous game, it would not be able to advance its middle pawn as it knows it is a losing move. In this way, by playing over and over with the AI, it will be left with no losing moves, and as there are no stale mates possible, it would win, and you, the human, would be unable to win another game. To build, it would be preferable but not obligatory to be in the possession of 24 match boxes. Cut out the image of the possible moves on the image above, and glue each one of the possible moves on each one of your matchboxes. Then, in each matchbox, put the number of moves the AI can perform in beads of different colours: if the AI can perform 3 moves, put 3 different colored beads, with each bead corresponding to a certain move the AI can make. When you start playing, put away the beads that are losing moves for the AI as we described above. After about 14 games, you should not be able to win anymore.
Comments