Basic artificial intelligence

Par Jérémy Perret, . Lien permanent Définitions

A recurrent theme of this blog will be artificial intelligence, AI. I need to define properly what I mean by this term.

Take your toaster. It's a tool, to which you provide bread, and returns you toasted bread a few minutes later. You can set the cooking time, and on most models, that's about all you're going to get. The toaster doesn't make any decision by itself. It's not AI (yet).

Now take a modern washing machine. Why does it spins slowly for such a long time before starting drying clothes ? Well, if the laundry isn't spread properly in the drum, all the weight is going to concentrate on one side, which will damage the machine when it dries at full speed. So the machine spins the drum a little, forward and backward, to change the distribution of the clothes inside. A sensor checks that the weight balance is better, before starting to dry. This is basic artificial intelligence.

You didn't ask the machine to balance the laundry. You just assigned it (by pushing buttons) a goal: to wash the clothes. The thing is, there are multiple ways to do that. The machine "decides", from its perception of its environment (the drum), to delay drying to lengthen its own lifespan.

It doesn't matter whether this behavior, those reactions, are programmed or not. The key concept is that an "intelligent" machine has a goal, and take decisions, given its environment, to better reach that goal. Whether it's a washing machine or a driverless car.

The basic level of AI is a machine with pre-programmed reactions. The level above adds memory, where the machine stores information during its operations, to be reused later. The level above that adds learning, where the machine adjusts its reactions from its observations, and improves little by little. I'll cover this topic in later posts.

If you encounter "smart" objects, keep in mind it's just a label, and that many things can hide behind it. First, the word hides the intensity. Say that a human is intelligent, and that a washing machine can be too, and you will be correct. In the same way that your cousin is tall, but the Eiffel Tower is too.

Second, intelligence adapts to situations, but doesn't necessarily learn. Your washing machine won't create its own original washing programs, even if it could technically perform new ones. Failure against an unknown situation doesn't defeat the performance of a machine in standard conditions.

Third, all intelligent systems are not alike. Artificial intelligence techniques range from speech recognition to real estate price estimation to robot finding its path in a room. Describing an object by "it's an intelligent thingie" doesn't get you very far down the path of understanding how it works.

We can, in any case, perform plenty of task with basic AI. Even though its techniques are well-studied, many domains can still benefit from them!

Ajouter un commentaire

Le code HTML est affiché comme du texte et les adresses web sont automatiquement transformées.

Ajouter un rétrolien

URL de rétrolien : http://www.jrmyp.net/blog/index.php?trackback/8

Haut de page