On Oct. 11, Facebook Artificial Intelligence Research (FAIR) Director Yann Lecun, in collaboration with McGill’s Integrated Management Symposium Series, gave a talk on how artificial intelligence (AI) impacts the modern workplace at Centre Mont-Royal. Lecun discussed the development of artificial intelligence in the 21st century.
Lecun began with a description of deep learning, a procedure in which machines are taught to perform tasks through trial-and-error rather than task-specific programming. For example, to teach a machine to identify airplanes in an image, the program is fed thousands of images with labels describing whether the image contains an airplane. The machine will then—over thousands of repetitions—sets its own parameters for performing the task and eventually be able to distinguish whether an image contains an airplane or not without any labelling.
“When you think about processes that we do very fast, like identifying objects in our field of vision […] many of these processes happen subconsciously,” Lecun said. “They are things we can’t necessarily explain. So it’s easier to teach by providing data, and allowing the program to figure it out on its own.”
Deep learning technology is becoming integrated into today’s business, medicine, and data analysis sectors at an exponential rate. Much of the talk was focused on the potentially negative impacts AI could have on the workplace and employment. Matissa Hollister, assistant professor of organizational behaviour at McGill’s Desautels Faculty of Management, considered the conflicts that may arise as AI becomes more widely implemented.
“I’m concerned that artificial intelligence startups are not consulting directly with the employees in the fields they are affecting, but [are instead] saying ‘Oh, we can do that!’ and designing the technology,” Hollister said.
Hollister elaborated concerns that in certain cases, AI may internalize errors or biases present in the data being used to condition the program.
“If you have a company that is using artificial intelligence for hiring, and it is using past hiring data, the AI may reflect the biases in the data,” Hollister said. “So AI will reflect the best and the worst of reality.”
Lecun, by contrast, noted the positive aspects he saw in the culture of AI research, and said he especially likes open research, whereby major AI players like Facebook and Google make their findings and code available to the rest of the research community. However, he acknowledged the need for large-scale solutions to the problems posed by machine learning in the workplace.
“Technological advancement like this will lead to wealth and income inequality, which makes AI a political issue,” Lecun said. “How do we deal with the losers [from AI]? How do we deal with those who lose their jobs at age 50? These are serious questions governments must be prepared to answer.”
Lecun declined to give a specific estimate for when so-called “general AI,” machines with the intellectual capabilities of humans would appear, citing a trend of underestimation of how long AI-based technologies take to develop.
“In the ‘60s, researchers were saying that fully intelligent machines could appear within 10 years, that they’d give the problem of vision to students as a summer project,” Lecun said. “These are problems we are only now beginning to solve.”
When prompted by the audience, Lecun offered career advice for the AI age as well. To Lecun, the burgeoning AI industry will require contributions from individuals with backgrounds in numerous fields in order to develop.
“The important thing today is learning basic knowledge that has a long shelf life, things that need to be learned while you are young,” Lecun said. “Things like math, physics, critical thinking and even philosophy, basic methods that teach you how to think. These things will continue to increase in value.”
Darlene Hnatchuk, director of McGill’s Career Planning Service, also attended the event. She found the talk to be encouraging and believes that AI will integrate positively with the job market.
“By automating certain narrow tasks and processes and activities, [AI] will allow for the development of new roles, jobs, and opportunities that harness the complex thinking that people do (that cannot yet be done by AI, and not likely anytime soon),” Hnatchuk wrote in an email to The McGill Tribune. “Most of us will be using AI to support our work [and] employers will rely more on AI/applications to support their recruitment and selection.”