Industry Committee on Dec. 5th, 2023

[Voices]

Oh, oh!

[Jérémie Harris]

This is the nature of the technology. The challenge is that, yes, we're going to be facing the strongest economic temptation to take risks with this and to swing for the fences right when the risk is most acute. That's what we keep hearing from these labs and we see that dynamic play out internally as they compete with each other in the race to the bottom.

[Brad Vis, Conservative, Mission—Matsqui—Fraser Canyon, BC]

Finally, I have one more quick question.

I, like many of us here around this table, have children. What do we need to consider for children with respect to AI?

Is there anything specific we can be doing on the AI aspect of Bill C-27 to ensure that we do whatever we possibly can to protect the innocence of kids?

[Jérémie Harris]

I'm not an expert on child welfare or child psychology, but what I will say is that the persuasive abilities of these systems are ratcheting up at insane rates.

Look at OpenAI's GPT-4. They ran evaluations on it. They found out it was able to persuade human beings to solve captchas for it—those annoying tests that prove that you're not a robot. Well, it was able to persuade people to do those for it.

Think of the applications for marketing. I think adults and kids are going to belong to the same equivalence class of entities relative to these things. I think it's bad for kids, but I think adults start to look an awful lot more kid-like in the face of highly persuasive reasoning machines like this, essentially.

[Dr. Jennifer Quaid]

It's a reason to have an agency, and we're already thinking about these questions in the competition context, if I can tell you, in terms of the persuasive use of digital technologies and so on. These things are already happening, so I fully agree.

The pitch I'm going to make is that the Competition Bureau and the commissioner have a mandate to try to strike this balance between risks associated with concentration, misleading advertising and so on, and dominance and the benefits of economic activity, dynamism and innovation. I think you are certainly going to see that same tension in certain parts of artificial intelligence.

You need a specialized agency that has the expertise to look at those things and to make proposals. Ultimately, of course, you in government will have to tell us what the values are that are important, but I think it's very analogous. We have models we can use. It's not rocket science, believe it or not.

[Jean-François Gagné]

I'd like to add that I'm very concerned about the content recommendation tools for social media, particularly for content recommended for children. I'm worried about the compulsive behaviour and dependence that these tools might engender. I'm very concerned about this as a parent.

[The Chair, Liberal, Joël Lightbound]

Thank you. That's extremely interesting.

Ms. Lapointe, before giving you the floor for five minutes, I'd like, on behalf of the committee, to wish you a very happy birthday.

[Viviane LaPointe, Liberal, Sudbury, ON]

My question is for Ms. Quaid.

In May 2022 you appeared before the public safety and national security committee. At that time, they were performing a study on assessing Canada's security position in relation to Russia.

At that time, you—

[Dr. Jennifer Quaid]

Are you sure you have the right testimony? I have never appeared before that committee, but anyway I will listen to your question.

[Viviane LaPointe, Liberal, Sudbury, ON]

Okay, maybe it's a different one.

You said that cyber-threats are becoming more sophisticated and are increasingly pervasive.

Is that associated to you?

[Dr. Jennifer Quaid]

Maybe that's artificial intelligence imitating me.

I am not a cybersecurity expert at all, and I would not venture an opinion on something I don't feel I have the expertise to comment on, so it's a little surprising.

I have been before the industry committee, but no other committee. I have been before the Senate banking committee, but not national security. It must be someone else.

[Viviane LaPointe, Liberal, Sudbury, ON]

I must have a different one.

I will ask you a different question. I would be very interested in your thoughts on what role you see for public awareness and education in building a resilient nation against AI-related threats in Canada.

[Dr. Jennifer Quaid]

I think it's crucial. There's no question that education is a fundamental, especially when we're talking about children. I think it's going to be challenging, and I don't want to understate it. Once again, I defer to the technical experts who know what is in the technology, but there is no question that education helps.

I'm going to beat my horse this entire meeting just using the example of the Competition Bureau, which does a lot of education proactively—a lot to do with misleading advertising, and so do all the consumer protection agencies of the provinces. There's quite a good collaboration there, and that's because fraud, and particularly digital manipulation, is going through the roof.

People need to be informed. We're playing catch-up. You know that has been true of the criminal law and the criminal justice system forever. That's not going to change, but we still have to try and, as best we can, keep up with what's going on.

I think the worst thing we can do is say that because it's too hard, we do nothing. That's why I'm here, and my colleague agrees 100% with me. We have to do something. It's going to be imperfect. We're going to play catch-up, but it's important.

You know what? There are a lot of people who can contribute their expertise to developing the tools. I really do believe this is not an impossible task— hard but not impossible.

[Viviane LaPointe, Liberal, Sudbury, ON]

Mr. Harris, I would like to ask you the same question.

[Jérémie Harris]

Yes, it's a pleasure to answer.

This is close to one of the areas I work in a lot. One subset of the work I do is training for U.S. officials, especially more senior ones, in the defence and national security universe. One of the challenges with training is.... It's been commented on many times. The space is moving so fast that the training has to somehow be relevant and fresh.

There are a few core things the public should understand about the drivers of this technology—about capabilities. This idea, for example, of scaling up AI systems so we can have a rough sense.... If I tell you roughly how many computations went into building a system, you can have a rough sense. “Okay, that's a ChatGPT-level system.” Immediately, you have a comparable that you can establish. We can basic things like that.

There are other things. I think this is a solvable problem. We've had a lot of success finding scalable ways of doing this.

Anyway, there are a lot of partners to collaborate with, in terms of the point that was just made here. I'm optimistic on that front.

[Viviane LaPointe, Liberal, Sudbury, ON]

Mr. Harris, earlier we heard Ms. Quaid talk about that need to strike a balance. She believes a good, viable solution around that is the creation of an agency.

I would like to ask you how Canada can strike a balance between harnessing the benefits of AI for security while minimizing those associated risks.

[Jérémie Harris]

One of the frameworks I like is, again, computing power as a kind of barometer we use to determine the level of the general capability of systems. There are asterisks galore on that. We heard that, yes, you absolutely can do—in technical terms—inference time augmentations. You can do all kinds of stuff, but the fundamental capabilities of a base model are limited by the amount of computing power you put into it.

In that sense, look at what's being done in the executive order. They're pulling on that thread. They're starting to build institutional capacity for using that as a yardstick. I think that's the best yardstick we have. It's imperfect and I wish it were not, but it is the best yardstick we have at the moment.

There's a lot of stuff we can do around evaluations and audits, depending on what level you are at on that computing-power hierarchy. The more computing power you spend to build a model, the more it costs. GPT-4 is costing, by our estimates, anywhere from $40 million to $150 million to train, just in computing power alone. I'm sorry, but if you can afford to train GPT-4, you can afford a little auditing.

That's the nice thing about this yardstick. It maps onto resourcing, as well, and we can use that to calibrate the trade-off between risk and reward.

[Viviane LaPointe, Liberal, Sudbury, ON]

From your perspective, how important is international collaboration in addressing the global security implications of AI?

[Jérémie Harris]

Hugely.

I think, ultimately, partly because of what's going on in the open source world right now, the problems that Canada generates, the rest of the world gets to eat. The problems the rest of the world generates, Canada gets to eat. We live in one Internet ecosystem. If we drop the ball, we're letting the world down. If the world drops the ball, they're letting us down. We can't have a hypocritical system where we turn around to other, adversary countries and say, “Hey, you ought to do this”, if we're not doing it ourselves.

I think there's a certain universalism to the situation we find ourselves in.

[Viviane LaPointe, Liberal, Sudbury, ON]

Ms. Quaid, would you like to add to that?

[Dr. Jennifer Quaid]

I think this is the central challenge. You need to figure out....

We're having this same conversation in competition, namely, of how do we adjust to the digital economy? That's a word I don't like. It's the new economy, which has a lot of digital artifacts. There's a lot of experimentation happening internationally. People are trying different things. Part of that is because we have different legal structures, institutional structures and cultures.

I think the balance that has to be struck is this: There are probably a small number of things—and you are the elected representatives who have to make that assessment for the country—that really matter to us, as Canadians, and that might be unique to us. Monsieur Lemire evoked our linguistic identity and the cultural specificity of Quebec, but there are other things that might be very important to us. Think of our indigenous communities. If those are very important, we have to bake them into our system. Then, internationally, what we try to do is make sure we're aligned on most of the big things. We can have a couple of things that are very important to us, and maybe we have special rules about these, but we need to have general alignment, because otherwise it doesn't work.

The challenge is making sure we take those general-agreement principles and translate them into operational legal rules. I guess I'm a bit of a nuts-and-bolts lawyer for that kind of thing. We have to be cognizant of the structural and legal limitations of our system.

We exist in a federation. I want to make one point about regulating everything. There is a division of powers, and a lot of regulation has to come from the provinces. Let's be very clear: This bill is centred on interprovincial and international trade, and on criminal law power that doesn't cover everything. Co-operative federalism is going to be essential.

International co-operation is important, but we also have to agree in the federation.

[Viviane LaPointe, Liberal, Sudbury, ON]

Where have we seen that done well?

[Dr. Jennifer Quaid]

I guess you could ask yourself whether the Europeans have done it. Maybe Céline wants to say something about how well they have navigated the necessity of integrating these considerations.

I think the European system already lends itself to this necessity of dialogue. This trialogue is the three entities that represent the three basic political powers in the European Union—and I'm mangling this—and they have to get together and agree.

I think that we might have to imagine something like that. I know no one likes that idea, but I think a conversation has to occur among the provinces and the federal government. It also probably has to involve local communities. It's all hands on deck.

I take the point that we can't regulate everything with one general framework, but we do need a general framework to set things up.

There are some out-of-bounds things. Let's say, "Kids—out of bounds." Just period, right? We can do blanket prohibitions. We do it, right? It can be done, but you have to target those things, and then for other things we need to make sure that the patchwork fits together.

I'll share one concern I have and then I'll stop. My concern is that if we are not co-operative in the federation, what is going to happen? There will be a set of litigation, founded on the division of powers, like we had 25 years ago in environmental law where large economic interests who have the money to do it will say, “this isn't federal jurisdiction”, and then provinces will want to say, “this isn't provincial jurisdiction", and it will take years to sort out.

If there's agreement, you can make sure there are no holes and that Canadians are protected.

[The Chair, Liberal, Joël Lightbound]

Thank you very much.

Go ahead, Mr. Lemire.

[Sébastien Lemire, Bloc, Abitibi—Témiscamingue, QC]

Thank you, Mr. Chair.

Ms. Castets-Renard, I heard you yesterday on Radio-Canada as I was headed to Ottawa, and the topic was really interesting. You were talking about the things that could go wrong with artificial intelligence as a result of its use by law enforcement authorities, particularly in connection with facial recognition. What I understood from the case that occurred in Ireland was that the use of artificial intelligence could, for instance, place the presumption of innocence at risk.

Are current Canadian laws sufficiently advanced to protect against potential social problems? Bill C‑27 may not be the solution. How can we plan for or protect ourselves from these problems, which are probably imminent?

Not only that, but the use of artificial intelligence in political face-saving endeavours might well lead to other restrictions. That's what happened, I understand. Is that right?