In two previous articles we saw how the work of Alan Turing, Claude Shannon and John von Neuman made possible the creation of computers capable of hosting artificial intelligence. However, allprograms still had to be created capable of doing the task. That is why in this post we describe the path of the software from the first conversation simulations to the current language models.
Alan Turing was the first to define a way to determine the success of this type of program. The problem is that the Turing test only demonstrated programming skill, not that we were facing what could be called an artificial intelligence.
The path of software
While Marvin Minsky's definition requires that to be considered artificial intelligence a machine must perform the same task as a human being does it requires the ability to think, the Turing test only asks that a human being not be able to determine if his interlocutor is or not.
Although ELIZA, a computer program written in the mid-60s, was not intended to pass the Turing test, it laid the foundations for a number of programs that would attempt to do so. The show assumed the role of a psychiatrist asking a series of questions to a patient about common topics such as family, friends, or mood. According to the answers, she continued following a pre-established line.
Neither ELIZA (and indeed any other software) would have been possible without the works of a self-taught math teacher named George Boole who in the XNUMXth century began to study the translation into mathematical terms of the human reasoning process. For this, he analyzed the way to group objects into classes and what happened when these classes were combined with others. He then assigned symbols for each of those relationships.
From formalization to feedback
If the objects of a set are changed by affirmations and we establish three possible relations between them (AND, OR and NOT) we already have a way to classify them into one of two groups (True or False).
However, Boole's works were not good for all kinds of claims. A way was needed to describe general concepts. That is, they could be true or false depending on the circumstances.
So that it is understood. Boolean jobs allow you to work with the assertion
Diego is Argentine and writes in Linux Adictos
But you can't do anything with:
X is... and writes in Z.
For this, we had to wait until the 70s, when a German professor named Frege introduced the concept of predicates. A predicate is a statement that can be described as true or false depending on the circumstances.
Diego, Argentine y Linux Addicts are entities that are neither true nor false, but depending on how they are combined in a predicate they will make it so.
Freje added two expressions with their corresponding symbols:
For all (Indicates that all the values of a variable meet a condition)
There is a… such that (Implies that there is at least one element that will satisfy a condition.
feedback theory
The next great contribution to the creation of Artificial Intelligence was not mathematical, it came from Biology. Norbert Wiener, founder of Cybernetics, was interested in the common points between Engineering and Biology. It was that interest that led him to analyze how warm-blooded animals kept their temperature constant despite changing ambient temperatures. Wiener postulated that in this and other cases feedback mechanisms were at work. In other words, when receiving information, a response was produced to adapt to it.
Going further, he came to affirm that intelligent behaviors are nothing more than the result of feedback mechanisms. In other words, we can conclude that Intelligence (natural or artificial) is about gathering information, processing it, acting on the result, and repeating the process.