With the emergence of a video game center of international stature, this is one of Quebec’s greatest successes. Based in Montreal, the Mila Institute of Artificial Intelligence (AI) welcomes many doctoral students from all over the world every year. Researchers specializing in algorithms, which big companies like Google, Samsung and Facebook are crazy about. With a pleasant smile, bushy gray eyebrows, Yoshua Bengio, founder of the place and winner of the Turing Prize of the Association for Computing Machinery – the equivalent of the Nobel Prize in computing -, evokes the road traveled in thirty years: “In the 1990s , the institute did not yet exist. In my lab attached to the city university, I was one of the few researchers working on machine learning. ”
Today, the Mila brings together around 600 students and has around 80 professors. Its premises house many start-ups and companies. A critical size that allows it to more easily attract funding. “In this area, we are rather well off: beyond the money provided by the chairs, the institute takes care of seeking partnerships and national grants. As a result, the professors have a base at their disposal. comfortable enough to have a small group of doctoral students, ”says Yoshua Bengio.
The work of these researchers is already finding applications in everyday life. For example, Mila has developed a conversational agent that collects health data from people in need of care over the phone, before putting them in contact with a real doctor. He also designed an algorithm capable of detecting human trafficking and sexual exploitation activities online. Info Shield accurately identifies suspicious content, going so far as to identify the culprits, but also the victims. A future tool for the police? Some criminologists are convinced of this. According to them, with such algorithms, the chances of successful police investigations would increase.
Limited offer. 2 months for 1 € without commitment
Drawing inspiration from the brain to improve algorithms
However, the institute’s primary goal is not to fight crime, but to revolutionize machine learning. “The current AI systems are not as good as humans. This is a major issue for research. For example, if I decide to drive a vehicle in England, it will require concentration. after a few hours, I will know without too much trouble driving on the left lane. The algorithms themselves do not have this adaptability “, details Yoshua Bengio. Even the biggest systems, like the famous GPT-3, the natural language processing model with 175 billion parameters, have weaknesses. “They give the impression of understanding. But sometimes even a 2-year-old child does better”, laments the researcher.
To try to correct this problem, the Mila scientists are therefore drawing inspiration from our brain, in particular our ability to focus only on certain elements – the most useful – to learn. This approach is already allowing significant progress. “We are only at the beginning. It may take another decade to develop more advanced systems. But we hope that eventually large groups like Google, Facebook or Microsoft will use models based on our work “, explains the professor.
To follow the analysis and decryption wherever you are
Download the app
Of course, it will be necessary to regulate these new tools intended to take an increasingly important place in our society. “In Quebec, we were among the pioneers with the Montreal Declaration, a set of principles established in 2018 intended for businesses and researchers as well as for the political class,” recalls Yoshua Bengio. This document states, for example, that AI systems must be able to explain the decisions they make and that they must be subject to democratic scrutiny, debate and control. “Some see these constraints as a brake on innovation. But if we do not regulate properly, there will be abuses, negative effects on society. The population could even reject this technology.” And, for this world leader who has dedicated his life as a researcher to machine learning, that would be the worst-case scenario.