Aided by humans, they mutate and improving. The theory of evolution will it one day apply to the PLC?
They had to revolutionize the world, invade industry, to interfere in the slightest interstices of our daily lives, before turning into higher beings … There is still nothing. When, in October 1950, Alan Turing published in the journal Mind an article in which he raises the question-fetched: “The machines can they think?”, It stirs the wildest predictions. Test to support, the famous mathematician replied in the affirmative and Their appearance around … 2000. Today, the intelligent robot passion always the same. On 27 June, the symposium organized by the Group of prospective Senate, and sponsored by Express, show the abundance of research and hopes will point the most serious. Long, robotics has sailed into troubled waters because, until the mid 80’s, inspired by Turing, the proponents of artificial intelligence called “traditional” have ceased to copy the human brain, trying to reproduce complex cognitive systems (language, numeric, and so on.) while trying to give a maximum of knowing their creations. There were thus very ambitious undertakings, as the project CYC, the American Douglas Lenat, who wanted to teach his computer knowledge of a child 5 years. Or Elisa, an attempt to computer psychiatrist.
A formidable computing power
But the most positive symbol of traditional artificial intelligence remains without a doubt the famous Deep Blue chess king enthroned, an evening of May 1997, after defeating world champion Garry Kasparov. But the player IBM had only luggage for a formidable computing power (200 million positions a second) and a good big memory … But at no time Deep Blue has understood what he was doing. Simple running, he could still less read the emotions of his opponent. Because all these machines had one thing in common: they were desperately stupid. Today, robots are no longer just computers. They have taken shape. Some have arms (industrial robots, for example), legs (Asimo, the last humanoid Honda), Caterpillars (U.S. Mocasin of the University of North Carolina). Others have invested the infinitely small (Taro, the robot cockroach-Mitsubishi), took a look animal (the dog Aibo, Sony) or play football. These creations are crawling in steep terrain to search for victims under the rubble of an earthquake, inspect the piping of nuclear power plants operate on a patient with unmatched skill, swimming, etc.. But often they are hardly more advanced than Deep Blue. Most remain unable to adapt to a changing environment. Can we then talk about intelligence? “In the same way it was obstinate in vain to copy the bird to fly, man is come to nothing in trying to trace the brain,” says Philippe Coiffed, a researcher at the Robotics Laboratory at Paris ( CNRS). In 1985, a handful of scientists, including Rodney Brooks, Massachusetts Institute of Technology, near Boston, are secession: before trying to give the robots a conscience, it is better to enable them to behave as the most primitive animals. And if the intelligence came with life? “It was revolutionary, remembers Jean-Arcady Meyer, director of the Laboratory AnimatLab informatique de Paris (Paris VI mixed-Jussieu and CNRS). Rather imiter the human brain, whose mysteries elude us, it is better to copy the living, inspired by nature so robots adapt and learn based on their experiences. “Artificial intelligence” evolutionary “is born. This is Darwin at home robots. From natural selection becomes artificial. In a given environment, scientists confront generations of robots. The most successful are then allowed to “reproduce”, ie to see their programs order – artificial neurons – crossed with those of their descendants. Over time, they mutate and improving. “We have developed a hexapod robot equipped with sensors and sensors that has learned to identify and move, said Agnes Guillot, the Anima Lab. The first generation had difficulty moving, the second was small jumps jerky, and the most advanced eventually reinvent walking insects. ” But it is also possible to compare several vehicles in the same generation. “One of our recent experiments was to put two robots Khepera – small mobile robots barely bigger than a piece of 5 F and wheeled – in the same arena,” said Dario Floreano, the Institute of robotic systems the Ecole Polytechnique Federal de Lausanne (Switzerland). The first, the predator, the addition of a camera to see the second, the prey has no vision, but it moves faster. The hunter quickly understood that he could not catch up with his opponent, he then sought to anticipate its movements to catch it. At another stage of development, the prey has also evolved by deciding not to move and wait predator. After a few days of fratricidal struggle, strategies have sharpened on both sides and experiences gained are used to other generations. For the moment, evolutionary robotics does not exceed the boundaries of laboratories, but its promoters hope will soon lead to industrial applications in domestic and playful. It is not a panacea, since it covers actions primitives (walk, avoid obstacles, pursue an object, etc.).. Will it be applied to conduct more sophisticated, such as those involving memory or emotions? Nothing is less certain. Hugo de Garis, Starlab, Brussels is trying to create an electronic brain consists of artificial neural networks that grandirait according to the principles of Darwinian selection. Technoprophète or sorcerer’s apprentice, he has often drawn the wrath of a portion of the scientific community, which accuses him his lack of results. The disappointment could last: last week, Garis a net stop its work, a lack of funding. Fortunately, perhaps, because the oracle of doom is not very reassuring: “In twenty years, machines will be more intelligent than humans. This could then no longer be the dominant species on Earth. “