This article is from a special issue of Science et Avenir n ° 199 dated October-November 2019.
Hervé Bourlard is the director of the Idiap Research Institute (Switzerland), a foundation dedicated to artificial intelligence.
Science et Avenir: Your organization is developing AI in the service of humans. what do you mean ?
Herve Borlard: AI should improve lives, not slaves, as is often the case – see how people relate to their smartphones! For example, we are working on improving videoconferencing systems to give participants a better sense of human interaction. or on voice banking, A great technique for patients who need surgery for throat cancer, with risks to their vocal cords. These people record the sentences to make a synthetic voice so that after the operation, they can find their voice through this computer system. This application has a direct impact on the quality of life!
What level of empowerment are we at in robotics?
We end up making robots that we no longer need to program. They are shown several times how to perform a task and manage to repeat these movements as long as the environment is controlled. At Idiop, we have developed a robot with the aim of helping people with disabilities eat and dress; For example, it grips your jacket so you can put your arms through the sleeves. It’s complicated, but there it gets! Another example – we are in Switzerland: we are working on “Scraper”, the first robot in mankind to scrape cheese! He is not a humanoid, just an arm above a table, “trained” by a racquet world champion.
Will computers soon be able to replace humans in some areas?
Progress is steady … but slow, which may be contrary to what the media is promoting. Alan Turing, the first scientist to do AI in the 1940s, said: “Identity not understood”. To recognize is not to understand. Today, the autonomous car is capable of recognizing almost everything around it: signs, obstacles… but to know how to avoid a stone on the road, I have to program it. Which key size should I choose? I get information from different drivers. They answer me: “You have to see, it depends, it’s instinctual.” Me, instinct, I can’t program it! … So, autonomous vehicles, this won’t happen in my lifetime!
Will we be able to make AI understand its choices?
Systems learn from large databases. As a result, they do not know how they reached their conclusion. In the past ten years, we have made substantial progress with expert systems to revisit what we were doing in the 1970s and 1980s: following the rules laid down by a human, a doctor, a lawyer, an engineer, etc. Trying to make intelligent system by presenting. . The great scientists were doing what is called symbolic AI, to no avail. But I believe that we will quickly reach the limits of machine learning, and that the future will be a combination of these two…
Analyst. Amateur problem solver. Wannabe internet expert. Coffee geek. Tv guru. Award-winning communicator. Food nerd.