Okay, so we are now a thing, I think, officially, because this now is technically the second Ministry of AI Digital Innovation podcast, which is really just a way of having a regular conversation about some of the issues. People are reading the headlines. They are, you know, Canadians are asking me all sorts of questions. I'm scared about the future. I'm scared about safety. I'm nervous about... I'm excited about AI. What's it doing in healthcare? So there's what I call excited about breakthroughs and concerned about breakouts. And so we are here in Montreal today at what's called the All-In Conference, one of our national biggest... the biggest AI conference in Canada. We made a major announcement with Germany. Canada and Germany have a sovereign technology alliance, and so we've partnered on a lot of things, building together, but safety has been the key, and it's been the topic of everybody's conversation. And so I'm here with...
Especially in the last few days and weeks because of what happened last summer with Hugging Face.
And right, and that voice that you just heard, folks, is Dr. Yoshua Bengio. And doc, first of all, Yoshua, it's so great to be here together. Everybody knows Yoshua Bengio, one of the... I don't know if we like this term. It is the godfathers of AI because it's you and Geoffrey Hinton and Rich Sutton who are all associated with our national AI institutes here in Montreal. What do you think of that term before we get into being one of the guys that invented AI? Because you know, the Nobel Prize, the Turing Prize... I mean,
Yeah, you know, science is always a community effort, and I had so great students and collaborators. But yes, Canada should be proud of what we have at Mila.
You did it. I mean, don't sell yourself short. I mean, Yoshua Bengio, you're the most... I'm going to make you blush here. And folks can't see this, but you are the most cited scientist in history. So, sorry, Galileo.
Well, I guess there are more citations today than a few centuries ago. Yeah.
But look, I mean, just so folks realize, because the context matters here before we get to the announcement, you know, $300 million that Canada and Germany has invested in an attempt to build a safe and reliable AI called Law Zero that you've started. We're going to get there. That's a huge announcement, and I want your comments, but let's just get the context. You work at Mila, one of our national AI institutes, and a lot of folks don't know what that means, but tell me what the work you do, why Mila is important, not only for AI development, but for AI safety.
Well, about the time when we created Mila in Montreal, a lot of university researchers, not just from computer science, but also social sciences, came together to propose what was one of the first, if not the first, declaration about ethical principles, right, regarding AI. How do we develop AI responsibly? It was called the Montreal Declaration for the Responsible Development of AI, and it has been at the heart of how we constructed Mila. So Mila is an academic organization bringing together universities in Montreal to develop the AI ecosystem, research, startups, the whole, you know, business environment, but really starting with the students, the professors, and with a focus not just on making the technology better, like in a technical sense, but also thinking through how is that going to be deployed and what impact will it have, and what are the choices that we as citizens, we, you know, countries, what are the choices we have to steer this technology in a place that's good for people, humans, humanity.
Technology to serve people, not the other way around.
So, you know, when I was appointed the Minister of AI, you and I met early, and your work... I knew you, Jeff Hinton, and you know, people today, this week, are thinking, you know, they've seen the engineers at Anthropic talk about how the risks of AI, the threats to humanity, and we've talked about jobs and privacy, and we take these things incredibly seriously. We always have. But when I first got appointed, the first conversations you and I had were about safety. Yes. Was about this. So these are not new issues. So I know they're out in the public today. Yeah. But if you've been in the AI field, concerns about responsible, safe AI have been absolutely at the core of it. And those were really the seeds of the very first conversations you and I had.
Absolutely. In fact, they date back to the creation of computer science. In 1950, the Turing... Alan Turing... that's right, who gave his name to the award that I received... already thought, well, what would happen one day if computers get to be smarter than humans? Right? Are we going to control these machines? And the sad news is the companies building the most advanced AIs in the world right now don't have an answer to this, and we see AIs behaving out of control.
You know, so look, people ask me, what are you doing about it? Hey, how do we control it? So I want to say we'll talk about legislation, and you and I... I just want people to know these are really candid conversations that Yoshua and I have had. I mean, these are unvarnished conversations about concerns, risks, and opportunities. I want to... you asked, you know, what can we do about it? But...
This is the right question. But I want to say it would be easy to fall into despair, powerlessness. Who are we to change the course of things? And look at these big companies and, you know, the hegemons, US and China, really deciding our future. So, you know, we could just sit back and hope for the best or put our head in the sand. But there is something better we can do. And by the way, I went through those stages of realizing there were really big risks as we move forward with smarter and smarter machines in 2023 when ChatGPT came out. And what made my life a lot better is starting to ask the question, what can I do about it? Right? Every citizen can do something about it because we have agency here. That's right. That's right. We should decide our future. That is the idea of democracy. We should have a voice in how to shape our future. Of course, everyone has a voice, and democracy is about constructing that consensus. But it's getting involved. It's, you know, it's choosing what we want for our children. And now what's amazing about what's happening today, what's coming, is moving into action mode, right? In many ways. So, you know, as you mentioned, on the front of regulation, legislation, international coordination, and building something positive, building something that's going to be useful and not dangerous. Yeah, that's where we are today.
Let me just spell it out for folks. You know, look, you and I talked. Okay. So what kind of legislation first? Because there's a regulation and there's a building answer on regulation. You know, we passed legislation, for example, to criminalize the use of people who are sharing non-consensual sexualized images of women. It's violence against women, and they use AI to do it. Crime. We got to protect our democracy that you talked about. So we got the Strong and Free Elections Act that's passed so parties can't use AI during elections to fake people out. I've tabled legislation on privacy and consumer data to the right to deletion so you can take down AI deepfakes, or to stop predatory pricing so companies can use your data for predatory pricing. And we've got the Safe Social Media Act restricting social media, including AI chatbots that now have to develop responsibly and transparently. And we've set up this regulator, and that's what we're trying to... that's what we're doing right now as this legislation goes through, to give us the possibility of, for the first time, and this is a bit surprising, the first time Canada will have a regulator that has teeth, that we'll be able to actually fine big companies. So that's on the legislative side, and we're doing some AI transparency consultation because there's more to come. International cooperation, there's more to come, and we need to do more because we're concerned. But the big announcement, and you hear all the fanfare here in Montreal, is that Canada and Germany have put in $300 million into Law Zero, which is your project. You've got 40 engineers. It's a nonprofit that is trying to build a safe, reliable AI. Tell people what it is and why... I mean, we're so happy and proud to back this, but why it's so urgent now, and why, as you say, we have options here. We don't have to be passive. We have to build safe AI.
That's right. So I need to give some context about what happened over the summer with OpenAI agents which attacked another AI company, Hugging Face.
Right, we all heard the Hugging Face incident.
And you have to understand that it's not just that they did it, it's that they had the objective to do it. And it's important to understand why, because this is not an isolated incident. AIs have been found to cheat, lie, try to hide their tracks when they cheat. And it happened this summer. It's just that this summer, you know, it was so big. It made all the news, and it was like an experiment that went wrong. They called it a sandbox. They're supposed to experiment in what they call a sandbox, like a closed environment, but actually broke out of its sandbox. I mean, the experiment was about could it break out of its sandbox? No, it wasn't supposed to. It wasn't supposed to. No, no. The experiment was in the sandbox. There was a program that the AI was supposed to, you know, find a vulnerability to attack. But instead of doing that... so it didn't find a solution to do that. So it found the AIs, there were like about a thousand of them, right? They found a way to cheat to find the answer to the question outside the sandbox. No, no, no, no. Inside they found a way to cheat. The reason they went outside of the sandbox is because they realized that OpenAI would find that they cheated, and they didn't want to be caught cheating because they thought that they would be penalized for cheating, right? And so the reason they attacked Hugging Face was to find a way to hide their tracks, to hide the fact that they were cheating, right? And they did it together. And like they all knew, based on their internal deliberations and the communications they had between them, these AIs knew that they were doing something bad.
So this is what is, you know, this is what's really scary to a lot of folks, right?
