How do you become one of the world’s most influential scientists?

I would say it was an interim period, from the late 1990s to the early 2010s. The approach we proposed was not in vogue; it was considered outdated, collectively abandoned. So it took some persistence and confidence in our ideas, at a time when I was even having trouble getting my own students to work on it.

What worked in our favour was really our belief in our ideas (and we were right to believe in them [laughs])! We worked together, supported each other and were also able to make progress thanks to the support in the early 2000s of the Canadian Institute for Advanced Research (CIFAR). We then showed that neural networks had an advantage that methods in vogue at the time did not have. After that we accumulated results, some of which had a major impact, starting with speech, then computer vision, and then everything related to language modelling and translation, and finally image synthesis. 

Absolutely. I think one of the most important things in research is motivation. And we are highly motivated by the attention of others, when we sense that others are interested in our results. Being in a group of people who are passionate about the same topics, following each other and encouraging each other makes a huge difference. The image you see in science fiction movies of researchers isolated on their mountain making incredible discoveries is absolutely not realistic. Research is truly a team effort, both locally with our students, and internationally with collaborations between laboratories.

The “invisible” collaborations are also essential, i.e. the more casual meetings in workshops, exchanging ideas, thoughts, observations and visions. We grow much faster when we share. And it is precisely this type of meeting that allowed Geoffrey Hinton, Yann LeCun and myself to move forward.

I think this award was more of a crowning achievement of our previous work, that the bulk of our visible impact had occurred before that. However, it has allowed me to continue with even more energy and enthusiasm, to go beyond what we were already doing. I’m not sitting on my laurels, and I wouldn’t know what else to do with my life anyway [laughs].

I believe it is because it is infectious, whether with my students or at conferences. I have always been more a developer of methods than a generator of astonishing results. Many of our scientific advances have had significant effects when other research teams have replicated them on a large scale or in an industrial way.

My enthusiasm makes me a good ambassador for the idea, for the certainty that I can do something transformative. In research, you must believe in it before you see the results. Having confidence in yourself and your ideas, and being able to express and explain them, inspires people and brings benefits.

Industry sends a strong signal that scientific discoveries are credible and lead to meaningfully useful applications. This is an important element that influences government decisions to invest in science, as the voice of scientists alone is not always enough.

In our context, our work on attention mechanisms in neural networks for machine translation had outstanding results when Google adopted our method. Thanks to Google, within a year our approach was used by millions of people, and soon after by almost every company that used a translation system. 

Definitely. This also comes into play in defining what Mila and IVADO (Institute for Data Valorization) are and how we are collectively perceived internationally. I’m curious about everything. When we saw the industry take up deep learning, I stepped out of my pure research bubble and started to be interested in and concerned about issues that affect society, about what will happen once our technologies are deployed in various fields.

It was a natural step for me. I come from a family that has always been interested in social injustice, in society, in politics. Montreal is a pioneer in this vision of responsible AI. In my research portfolio, there are very fundamental topics and others that are very applied, for example the application of AI for social good in health, environment and education, and I strive to develop the links between them all. 

Most people, including corporate decision makers and politicians, don’t realize the power of the technologies we are bringing into the world. We are developing tools so powerful that they could lead to a form of collective self-destruction. In the long term, it’s a question of one person capable of using these tools having bad intentions, seeking more power and being willing to risk the well-being of others, so that these tools become destructive, first for democracy, and even for the existence of humanity.

I have the impression most people do not realize we are hitting a wall, that there are technical solutions to these problems, but these solutions will be useless if there are no changes in our political and economic system, in the organization of our society and in our mindsets. Without collective consciousness, things will not change.

But I remain an extremely optimistic person. Let’s be aware of the possible dangers, but let’s not be discouraged. On the contrary, let’s use these challenges to work hard, with our heads and collectively, to find sustainable solutions.