Talking to machines in a natural manner, or talking thanks to machines (via automatic translation programs); solving complex problems that cannot be programmed a priori with the help of computers; living with service robots; simulating realist environments in order to understand certain mechanisms of nature, educate oneself, and intervene in dangerous environments from a distance; enriching our interaction with the physical world with new information… These objectives, which seem disparate at first, are today being pursued by research and industry, who are mobilising an ensemble of diverse but closely linked disciplines such as artificial intelligence, virtual reality, “artificial life,” augmented reality, and robotics. On a basic level, these have a common aim: an ever-greater intimacy between man and machine, between the physical and virtual worlds.

Artificial Intelligence (AI): simulating reasoning… and even more

Artificial intelligence (AI) aims to “automate reasoning” and to enable programmes or “intelligent” behaviour robots capable of adapting the way they act within unexpected contexts, learning from their experience, managing uncertainty, and making the correct decisions to obtain a given goal…

This relatively ancient discipline is based on a combination of methods that come out of computing (neural networks, recognition of shapes), cognitive science (expert systems), statistics (Bayesian networks), biology (genetic algorithms, swarm intelligence) and mathematics (fluid logic).

In its “strong” incarnation, popularised mainly by science fiction, AI has the fixed goal of reproducing the way in which the human brain functions, and then performing better than it. But this aim is still far off, even, some believe, unattainable (and/or undesirable).

These days, AI is used for much more specific, and sometimes very ambitious, ends: beating the world Chess champion (1997), diagnosing some illnesses as well as a doctor, understanding a human when he or she speaks (still a distant goal apart from a few relatively precise contexts), making an entirely automatic car cover 200km of hilly terrain (DARPA Grand Challenge), or, on a more everyday level, detecting spam, helping us choose investments, aiding an industrial robot to choose the correct piece from a pile, optimising bed occupation in a hospital, detecting a suspicious face entering the subway system, choosing the interesting segments of clientele from a large database (data mining), or automating telephone customer service response.