The Basics of Artificial Intelligence (AI) for Dummies
We hear a lot about Artificial Intelligence (AI). The acronym AI is very fashionable and consequently arouses many passions. But what exactly is AI?
John McCarthy’s Definition
Let us begin by devoting ourselves to the definition pure and simple in order to avoid confusion. And to do this, it seems that turning to the very inventor of this definition is a wise decision.
John McCarthy, in 1955, at the Dartmouth Conference, giving precisely the following terms:
“All aspects of learning and all other constitutive elements of intelligence can in principle be so precisely described that it seems that a machine can simulate it. Tests will be conducted to find out how machines can use language, forms of abstraction and concepts, solve problems hitherto reserved for humans and progress by themselves.”
In other words, Artificial Intelligence is a machine built to solve problems generally solved by men and women through their natural intelligence. The machine will demonstrate such faculties when it will be able to progress on its own to solve these problems.
The Seven Areas of AI
This same original summit of Dartmouth defined the seven fields of application of the AI. There are a few more today but here are the first seven:
- Simulate the main functions of the human brain
- Programming a computer to process natural language
- Arrange “hypothetical” neurons so that they form together concepts
- Determine and measure the complexity of a problem
- Auto Improvement
- To develop the faculty of abstraction (at the level of ideas and not of facts)
- Creativity and Chance
And in fifty years, we can say that progress has been phenomenal in the treatment of language, measuring the complexity of problems, and self-improvement at least to some extent. That said, we are now starting to tackle the simulation of creativity as you can see on this page detailing how Artificial Intelligence puts itself at the service of art.
But What is the Intelligence in AI?
Yes, since in the expression Artificial Intelligence, it is not only “artificial”, we must already take an interest in what really is “intelligence”.
According to Jack Copeland, author of several books on AI, intelligence can be described through its properties namely:
- The ability to generalize from cases (as does the recognition of images in AI), that is to say the ability to react according to his/her past experience even though we have not yet been confronted with the new situation.
- The reasoning, that is, the ability to draw appropriate conclusions for each situation
- Problem solving, ie putting an equation (not necessarily mathematical) and finding the “x” of this equation.
- Perception, that is, the ability to scan and analyze an environment and create relationships between observed objects (as in autonomous cars)
- The understanding of language, that is, the ability to understand language as a human would do.
Types of Artificial Intelligence
Since the bases are now based on the original definition of AI and even on human intelligence, it is necessary to distinguish two distinct major families.
- Strong Artificial Intelligence – First there is the strong Artificial Intelligence which simulates the human brain in the construction of systems of thought. In short, people would be able to do the same thing as the human or even better.
- Low Artificial Intelligence – Low Artificial Intelligence is a system that “behaves as” a human but does not model the way the human brain works. Deep Blue of IBM who in the 90s won a victory that made a big noise against the chess champion Gasparov is an illustration: the machine made the calculation of all its possible displacements on the chessboard before playing each shot.
- Middle Artificial Intelligence – When the system is inspired by human reasoning but does not remain fixed on its model. This is the case of IBM Watson, who, like a human, reads information, recognizes key patterns, collects evidence and says “hey, given the elements I analyzed, I am sure 67% solution is this “. This is how he won the game Geopardy in 2011.
Deep learning also falls into this category since it reproduces in a certain way the functioning of the brain by relying on a network of neurons without exactly following the same pattern.
Since deep learning is only a sub-family of machine learning, it is necessary to take a second to describe what machine learning is.
The Machine Learning
First of all, it must be remembered that this is only one of the techniques of AI, even if the media generally speak only of this because it is very fashionable and, it must be admitted, created a small revolution in concrete applications.
By Machine Learning, we mean a sum of algorithms capable of improving the performance of the machine as it obtains data. This is an input-output principle: input information is entered, and output information is expected. If they are correct, they say it to the machine (it is supervised) and if they are false it is also told. And gradually, she learns to have more and more correct conclusions in output.
In short, we do not know how to create a program that allows the machine to recognize images of dogs. But we know how to program a machine that will be able to learn, because we will show thousands of examples, draw conclusions from it and offer interpretations of these images that we will validate or not and, by force, will be deceiving less and less.
But be careful: the machine will not know what a dog is, what it serves, for example, will have just learned to say “0001” when the entry is “11100”, neither more nor less.
In conclusion, you now have the basics to be able to understand what is said about the AI, to shine in dinner, or even to have a minimum of retreat both on the uses but also on the articles that you can read here and there. These technologies give us a glimpse of a revolutionary future, even if we are only at the beginning: you will know at least what we are talking about and what lies behind these promises.