A brief history of AI

If you are a user of smart speakers and personal digital assistants, you are familiar with the notion of AI. But as often, we pretend to know what we are talking about without really being sure, or even completely missing the mark (but in front of the in-laws, it goes, they do not know how to use their smartphone anyway).

So in the end, what is an AI? And how does it work? What about connected speakers? We will try to synthesize all of this for you to make you shine in society or just for your own pleasure!

What is an AI?

What is artificial intelligence? The Larousse dictionary defines it as a set of theories and techniques implemented in order to produce machines capable of simulating human intelligence. The definition is quite clear in the sense that AIs are multiple, both hardware and programs, hard and soft ... but the final goal remains to get as close as possible to the human psyche, to no longer be able to tell the difference between a man and a superficial intelligence, arrive at a singularity. And overtake her if possible.

Since the existence of computers, man has sought to describe simple tasks with increasing precision so that machines can do them more and more faithfully. And if it works with tasks, it can work with ideas, concepts… right? And this is basically the principle of AI: describe with maximum precision an intellectual activity to simulate it by a computer (according to John McCarthy, one of the founders of AI). And the more powerful the machines, the greater the precision and the more tasks described.

But then AIs are not new, far from it! Well yes, programs dedicated to the resolution of a particular task, this is not new !!… And there you have it! We must not confuse AI and software. Software will be coded to do the same task repeatedly, to process it according to a predetermined way via an algorithm. Thus, it will make it possible to make entries, word processing or even to make calculations in order to reduce the probabilities, to make forecasts but always in the sense that it follows a pre-established way of doing things. While an AI will be edited in order to deal with complex problems but always with a notion of intelligence in the sense that it will have to think about possibilities, anticipations, solve particular cases such as playing chess, recognizing a face or a figure. object or driving an autonomous vehicle. An AI also has in its programming a notion of machine learning, allowing it to learn by itself because it is impossible to code all possible situations. Do you understand the nuance?

A little history !

The universe was born 13,7 billion years ago, and in the 1950s, the first computers began to emerge. No link between the two, except that without the first, the second could not have taken place, hence this shortcut in the timeline. During the second world war, Alan Turing (remember its name) has also succeeded in changing the understanding of computing. He was able to develop one of the first computers (they were not yet "computers", appeared in the 1950's) capable of deciphering the code of the German machine. Enigma and greatly contributed to the victory of the allies. It was only a machine capable of transforming one letter into another once the key had been deduced and thus decipher a message.

But this brave Alan dreamed of AI before the concept even existed. From the 1930s, he details the ideas of programming and programs when computers do not really exist. And in 1950, he even created the Turing test. This test, imaginary at first, consists in making communicate a human and a machine, without him knowing it. If the human is not able to say that his interlocutor is a computer, the test is successful. This concept is very simple and timeless. And there, we touch the artificial intelligences of the finger. At the end of the 1950s, the beginnings became real bases with the creation of the first AI program It was in 1956 that the date for the creation of AI was fixed, during the dartmouth conference. This is only a milestone because before this date work on AI existed, much theoretical and more and more practical. But beware, the early AIs had nothing to do with Deep Thought, the deified Artificial Intelligence of the Galactic Traveler's Guide, which answers the big question about life, the universe and the rest (ask for the answer to life). , the universe and the rest to Alexa). No, we are on improved programs, trying to think like humans to do human mono-tasks, like playing checkers or chess.

It is indeed on games that AIs are calibrated because they respect precise rules and also require deduction, analysis and anticipation capacities. The first programs were more of the improved supercomputers gathering information, existing hits, calculating probabilities but no match for humans. Then they improved, they had more and more the capacity of deduction, anticipation and decision-making until arriving, in 1997, at chess matches between Kasparov and Deep Blue, earned by IBM representative. In one of the matches, Deep Blue got creative by sacrificing a jumper in order to create an opening in the opposing game. At present, AIs are attacking the game of Go, which is even more difficult to master than chess.

The fact that the researchers tackled games as a priority made it possible to focus only on a particular task and to make the AI ​​evolve so that it learns and improves. Before running, you must know how to walk and before that, you must be able to stand. This is what they did. When a step is successfully passed, the knowledge can be duplicated on another type of task. There is a very strong advance in the 1960s and 1970s thanks to more powerful computers, and we tackle algebra, languages, mathematics ... Machines more or less learn a language on their own, they solve problems more and more complex (prime numbers…). But it's still limited and fairly straightforward for humans.

Then material limits arrive and computing capacities stagnate. And this is a problem because solving more complex situations requires far more resources than what is available at the time. It was not until the 1980s that everything started up again (with a further slowdown in the early 1990s). Not only are machines more powerful, but the development of the Internet is also playing a big role, and researchers are developing new approaches. These approaches allow a better learning of the programs and a much faster evolution.

But how does it work?

To make the history of the various approaches to the functioning of AI would be quite complex and soporific. But a few ways of doing it are quite interesting. First of all, know that there are two types of Artificial Intelligence: the weak and the strong, or the descending and the ascending.

Weak AI is the first type of AI to appear because it is dedicated to solving specific problems. It is a program that will imitate humans for the accomplishment of a task and will only do that. She will not or little change, learns more or less autonomously from her mistakes, but always within a defined framework. She can reason based on the information she has in her database. These weak AIs are the ones we encounter on a daily basis. They are used in many software such as text and image recognition, text translation, games, but also search engines, turning on or off a device on demand. Yep, connected speakers are weak AIs… for some.

Alexa is therefore a weak AI? Not really. Because Alexa approaches what is called a hybrid AI; it is at the same time weak (because one feeds its database, it does planned and repetitive tasks) and strong (because it learns certain aspects of its mechanics by itself, by basing itself on its errors thanks to the multitude of cloud data and a touch of Deep Learning).

A weak AI will see its database supplemented with a multitude of information necessary for its task. She will be taught what to look for and how to deal with them. We therefore start from a database to go down to a result. It is necessary to constitute its bank of information and what it must find: photos with a car on it, human face… We show it the correct answer, it classifies and reproduces. It therefore manipulates the information. This is called the machine learning, machine learning or automatic.

But weak AI doesn't mean cheap AI. They are extremely powerful, the networks necessary for their functioning are based on an approach of the human neural network, with connections by stratum making it possible to manage each one a concept, an action and to distill it more and more. And it takes a lot of computation, with tens of CPUs processing trillions of operations per second. But weak AIs remain a simulation of intelligence and don't understand what they're doing. From there, their evolution is limited.

This brings us to strong AIs. In this case, the machine has the same basic analysis capacity as a weak AI, but it will moreover reason by itself, anticipate, learn from its mistakes in an autonomous way. She will create her own solutions to a problem on her own for the good reason that she understands what she is doing. It goes even further because we lend to ascending AI a consciousness of oneself or even emotions ... And here we are on the equivalent of the depressive robot Marvin of H2G2: a "person" with emotions, a consciousness, states of mind. … But this is only anticipation or even utopia at the present time. However, it is possible to push artificial reasoning further and further into specific areas, but not all at once, not yet. In this case, we fall back on the setbacks of the 70s: the lack of computing capacity and data storage.

But the beginnings are here, especially with the Deep Learning ou deep learning. It is a proven self-study program technique. The company DeepMind, owned by Google, specializes in this niche. They apply a deep learning even more advanced (deep reinforcement learning) involving notions of psychology and the notions of cause, effect and consequence. It is also thanks to this that DeepMind succeeded in creating an AI having beaten in the game of Go one of the best players in the world in 2016!

 

Did you like this little digital overview? in any case, it is always interesting to understand the how of things, without going into too much detail. AI is a big topic and one thing is for sure, we are at a crossroads. Strong AIs will become real within a few years. What will be the place of man, the consequences for the future? We refer you to our previous article on the subject. Let us profit as man, with a few exceptions, we all know, is still intellectually superior to the machine!

Did i choose Alexa or did she choose me? Let's say that as a beta tester, it's a bit of both! And I do not regret having met our favorite assistant nor this adventure undertaken with Alexien Modo. Technophile, self-taught and loving popularization, I try to make our common passion easy to access!