Different perspectives within AI

Within Artificial Intelligence field there are two main fields that have been trying to create a computer as similar as the human brain. In scientific words, a computer able to pass the Turing test (it will be discussed in the following section). The first method was exploited between 1950 and 1970 and it was oriented to discover the knowledge rules. It was called the symbolic IA. After 1970 a second field appeared due to the bad results of the first one. This new method was called sub-symbolic and it was based on the probability. These are the main features of these fields but they are much more complicated than this. Thus, this is a more deeply explanation of their features.

In the earliest 50’s a new field in computation appeared, Artificial Intelligence and a lot of researchers concerned about this new field. At the beginning they tried to develop the knowledge, in the computer, as a collection of basic rules. For instance, if we know that the birds can fly and a pitch is a bird, the pitch can fly. This is a very simple deductive reasoning. Therefore, the researchers only had to store in the computer a lot of premises in the computer analysed the inputs with these premises. However, this method has a problem, the ambiguity of a normal conversation caused a lot of difficulties to the computer. Therefore, these premises must be enunciated rigorously. Furthermore, all the rules have an exception, for instance, in the case of the birds, a penguin is a bird but it cannot fly, however the computer would deduce that it can fly.

Because of these obstacles this field based on an easy deduction method was abandoned. In the 70’s the researchers focused the IA in the probability. They wanted to create a program able to learn with a lot of examples. For instance, the researchers showed to the program a lot of pictures with birds and it had to distinguish what they have in common. This new method worked quite well but the problem was that with complicated concepts as maternity or grammar this software got confused.

This is the most popular field in IA nowadays, also called Computational Intelligence, and there are different techniques as Evolutive Computation, Swarm Intelligence or Neural networks. The last one is maybe the most famous because it tries to copy the human brain operation. It is a system in which all the neurones are interconnected in order to produce and output. Nowadays the problem of this technique is the capacity of the computers. A human brain has 100000 neuronal connexions and a powerful computer only has 10000. Nevertheless, the Internet could help in this problem because it give us the opportunity of having connected computers and it makes possible to copy the human brain.

In the last year a researcher called Goodman has developed a program language, Church, that includes basic rules as the first IA methods but they are based on probability. Therefore it mixes both methods. For instance, if we establish that a pitch is a bird, the program Church will give a probability to its possibilities of flying, If we give the program some extra information about the pitch (its age, its sicknesses), the program will modify its initial probability estimation and it concludes that maybe the pitch can not fly. This method is still a theory but it seems a huge step in the IA, which last years would seem stagnant, because we give the opportunity to the computer to learn by itself using a probabilistic calculation. Indeed, this is the way the humans use to learn.

These are a few simple examples in which we can see how these technique works. They are all related with Neuronal networks and it is quite visible how the computer by itself learns how to work when we give it a certain problem.

http://www.youtube.com/watch?v=lmPJeKRs8gE

http://www.youtube.com/watch?v=oU9r64tc7yE

http://www.youtube.com/watch?v=MbtJ-Y4-T0Y&feature=related

http://www.youtube.com/watch?v=rFMBTIPLUFg&feature=related

http://www.youtube.com/watch?v=ytRi4rvnBsc&feature=related

Perhaps the most amazing is the last because we can see literally how a machine is learning to face a problem and to solve it. As It was said before these are simples examples with just a hundred neuronal connexion, imagine what we are able to do with millions of them, this is just the beginning.

Furthermore, this field is not only in the computational labs. There are some commercial products in the streets. For instance, a famous videogame called Quake II (you have a gun your mission is to kill as many zombies as you can) has one of the enemies based on neuronal Networks. It is called Neuralbot and it learns from the movements that we do during the game. Here you can find some extra information about it:

http://homepages.paradise.net.nz/nickamy/neuralbot/nb_about.htm#about

This is a simple video to explain how this enemy works:

http://www.youtube.com/watch?v=_StkH25eulg&feature=related

In one hand, this is an IA approach, to develop a technology able to create machines with reasoning capacities similar to the human intelligence.

On the other hand, some researchers have focus on another approach. This approach uses the computer as a simulation tool in order to validate theories. It doesn’t want to obtain intelligent programs but discover what is the intelligence. Because the intelligence activity come up from the animals, a lot of researchers have focused their attention on what is the life. This field, which is called Artificial Life, tries to create something alive using a combination of data and programs. The first assumption that is made in AL is that the intelligence comes up from the life. Furthermore, if we are able to create life we will understand better how it works and what it needs to exist. Nevertheless, a lot of people have decided to separate this field from the IA because of its huge dimension.

Within IL, the Celular Automatons are the best example of life generator. They don’t seem intelligent but they present several fundamental aspects of life. The definition of Cellular Automaton is:

It is an ensemble of cellules that are interconnected each other. The state of one of them depends on the state of its neighbours and its own previous state. So if we give to the Automaton several inputs it will produce an output depending on the transition function that we have gave to the automaton at the beginning.

The most famous program is the game of Life created by John Conway. It is quite simple because you have a screen with a lot of cellules (in this case the bits of the computer screen), which could be white or dark. If we give the program some initial instructions (for instance, if one white bit has three black bits neighbours it will turn into black) and we let it run, the bits will change its colours and the screen evolves with time. This program became famous because with particular initial instructions the screen seem evolve as something alive and there are a lot of researchers interesting in this. It is impossible to talk about this deeply in this article because it would fill hundreds of pages but I will put some example of how this program works.

http://www.collidoscope.com/cgolve/map.html

http://pentadecathlon.com/lifeNews/index.php

I put here the real program if someone is interested on it:

http://www.bitstorm.org/gameoflife/standalone/

As you can see in the example it seem alive and the question is, is there life inside the box. My answer would be affirmative, artificial life, an artificial universe. Maybe we are living in a Game of Life.