Industry Committee on Dec. 5th, 2023

[Madam Céline Castets-Renard]

In Canada and the provinces, the use of facial recognition, generally speaking, and in particular by law enforcement agencies, is not circumscribed. Of course, without a legal framework, it becomes a matter of trial and error. As was demonstrated in the Clearview AI case, we know from a reliable source that facial recognition was used by several law enforcement agencies in Canada, including the Royal Canadian Mounted Police.

When there is no legal framework, things become problematic. Practices develop without any restrictions. That's why people might, on the one hand, fear the legal framework because its existence means the technology has been accepted and recognized, while on the other hand, it would be naïve to imagine that the technology will not be used and can't be stopped, and possibly has many advantages for use in police investigations.

It's always a matter of striking the right balance between the benefits of AI while avoiding the risks. More specifically, a law on the use of facial recognition should ideally anticipate the principles of necessity and proportionality. For example, limits could be placed on when and where the technology can be used for specific purposes or certain types of big investigations. The use of the technology would have to be permitted by a judicial or administrative authority. Legal frameworks are possible. There are examples elsewhere and in other fields. It is certainly among the things that need to be dealt with.

I would add that Bill C‑27 is not directly related to this subject, because what we are dealing with here is regulating international and interprovincial trade. It has nothing to do with the use of AI in the public sector. We can, in due course, regulate companies that sell these facial recognition AI products and systems to the police, but not their use by the police. It's also important to ask about the scope of the regulation that is to be adopted for AI, which will no doubt extend beyond Bill C‑27.

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

Ms. Ifill, I'd like to hear your point of view on potential uses of artificial intelligence by the police, along with the problems that might follow. As you made some rather astute preliminary comments on this topic, I'd like to hear what you have to say.

[Erica Ifill]

One of the problems is that it does tend to misidentify not only racialized people but also non-binary people. There are cases such as self-driving cars having a problem recognizing women. When these technologies start to affect larger proportions, or a significant proportion, of the population without some sort of accountability measure, we're looking at a very bad fragmentation of society on an economic level, a social level, and in ways that would fracture our politics. I think that can be minimized, to be honest.

One of the things I would like to see with AIDA is that it be its own bill. I personally think it should be spun off so that we can look at these things more clearly, because, as it stands right now, there is nothing to.... For example, if you go for a loan and AI predicts that your loan should be rejected because of a variety of factors, or maybe factors that aren't attributed to you because of race, gender, class, geographical location, religion, language, all the things.... If we're going to build these systems, we have to protect people from the negative impacts of those systems, especially when they happen at scale and especially when they happen with government agencies.

I think one of the problems with this bill is that a lot of government agencies, especially in national security and law enforcement, will be exempt. Those are some of the areas—you think of immigration too—where you will see large uses of AI.

I would say about education that a lot of the education over time should have come from journalists and journalism. We should have had a more robust journalistic tech field that could inform all of us and look into these issues with AI and tech writ large.

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

Thank you very much.

[The Chair, Liberal, Joël Lightbound]

Thank you.

Mr. Masse, the floor is yours.

[Brian Masse, NDP, Windsor West, ON]

Thank you, Mr. Chair.

Ms. Quaid, if this bill doesn't get passed, we'll go to the new year. Then it has to get through the House of Commons. Then it has to go to the Senate. Then if the Senate has any amendments, it has to come back to the House of Commons with them.

Give us some of your concerns about the delay in the process. Is there anything else we can do in-between to deal with this? Even if we have political consensus in the chamber to move this as quickly as possible, our schedule is such that it won't be until the new year—and probably clause-by-clause and so forth will take some time. Then we have to send it to the other place. The other place sometimes can take some time. Then, again, an amendment would have to come back us in our chamber.

[Dr. Jennifer Quaid]

You are the experts of your own procedures. I'm not going to lecture you on how the legislative process should happen.

I will allow myself this small editorial comment. It's funny, on the one hand, that the AI bill has been moving slowly and has sat for a while. There was an original bill that was proposed, but died on the Order Paper, and there was a new one. I contrast that with some things, like in competition, where we're moving at lightning speed. It seems like it's also a question of establishing priorities. That said, I'm not recommending the budget bill process. Please do not quote me as saying that. I think it's a terrible thing.

On how you could move forward, here are some suggestions without my having any knowledge of what your procedures are, so I might be saying things that are wrong.

[Brian Masse, NDP, Windsor West, ON]

Yes, that's fair. That's fine. I appreciate that.

[Dr. Jennifer Quaid]

I am a planner, like many busy working mothers. I'm thinking that maybe you can already start checking out.... If you like my idea of creating an agency, for example—just to pick a random one—maybe you would start already looking at some possible structures and what's being done elsewhere. Maybe a study can be commissioned and ISED can already start looking at that. What kinds of skill profiles would you need to populate that? Maybe you can already start looking at some of the challenges that might be involved in the criminal and regulatory enforcement part.

It's been said before and beaten over the heads of many, but I think there needs to be more work done on getting an idea of what's going to be in those regulations, and thinking about how we are going to create a system where we can have this iterative process of updating the regulations. There are models out there. There are smart people in Canada writing about agile regulation. I think we can already start lining up what the feasibility of certain solutions is before the bill is enacted. Yes, it might not happen, and it might be effort wasted, but I think there are lots of researchers who'd be happy to look at these things and provide you with options. That would be my suggestion.

[Brian Masse, NDP, Windsor West, ON]

Thank you very much.

Thank you, Mr. Chair.

[The Chair, Liberal, Joël Lightbound]

Thank you, Mr. Masse.

Speaking of parliamentary procedures, the bells are ringing. We'll need to have unanimous consent, if you will, just to proceed until 5:50, which would be the time our committee would adjourn for this part of the meeting.

Do I have unanimous consent to continue?

[Some hon. members]

Agreed.

[The Chair, Liberal, Joël Lightbound]

That's amazing.

Mr. Fast, the floor is yours.

[Ed Fast, Conservative, Abbotsford, BC]

Thank you very much.

Mr. Chair, I'd like to get back to the bill before us.

Without prejudging the outcome of Mr. Champagne's proposed amendments, we will assume for the time being that those amendments would be passed and incorporated into AIDA. I'd like to know from the three of you—Mr. Harris, Ms. Quaid and Mr. Gagné—if you support AIDA's going ahead as it is right now. It's my understanding that in the future it is almost certainly going to be amended, re-evaluated, recrafted; and it may come back in a different form.

We have to make a decision on the bill before us right now. You're here giving us advice.

Is it your advice for us to go ahead with this, or are there substantive amendments that you would propose?

[Jérémie Harris]

I can speak for myself, to begin with.

I think the bill right now is significantly better than nothing. One of the key factors for me in evaluating this is just the timeline. Do we want to be confronted in the year 2024, 2025 or 2026 with nothing on the books? My strong impulse is to say no, we must have something.

Given the timeline, as has been explained to me by folks who are working on this bill, it seems unlikely otherwise that we would have something on the books by then. That's my understanding—it may be wrong.

If that is the case, then the bill in its current form is better than nothing. That's literally how I'm approaching this. There are things that are actually very good. I think the general purpose AI system stuff and the cessation-of-operations components to the bill are really good.

Overall for me, given the current landscape and the timelines, I would be in favour of the bill's going ahead. However, I see significant issues with it, which I highlighted, including the computational power thresholds and all of that stuff, in my testimony.

[Dr. Jennifer Quaid]

I can speak for Professor Castets-Renard as well on this point. We are of the view that it is better to proceed with this bill. It is improvable. We have identified the things that we think should be improved. I do think that making sure there are sufficient resources and an institutional framework to support the actual implementation of the bill is important. We have things that we suggest could improve and strengthen it—although those are probably at the level of regulations, or could be achieved through regulations. I think we need to move forward.

I would urge this committee to continue to solicit opinions from, or to pay attention to, people who are analyzing for the potential problems. That's not because they should defeat the bill, but we should go in with our eyes open to what the potential challenges are. I definitely come down in favour of the need to move forward. It's already too late.

[Jean-François Gagné]

I will provide a list of recommendations from my perspective on this. I think there are lots of aspects that need improvement. Some are too broad and others are too harsh in terms of the consequences around certain actions. Personally, my perspective would be to keep working at it.

[Ed Fast, Conservative, Abbotsford, BC]

And not go ahead with the current form of the bill?

[Jean-François Gagné]

No.

[Ed Fast, Conservative, Abbotsford, BC]

That's helpful.

Were any of the three of you or Ms. Ifill consulted in the lead-up to this bill's being drafted? All four of you are coming in after the fact and providing advice.

Go ahead, Mrs. Ifill.

[Erica Ifill]

This bill didn't have public consultation. It seems like a big step that the things we're talking about now would have been mitigated with proper consultation. To me, that's a big red flag about this bill.

Second, it really has no protection for the public. I know that there's this framework, but even the framework is insufficient.

I would have to agree with the last person who spoke and asked if you really want to push through a bill that has no protection for the public.

[Ed Fast, Conservative, Abbotsford, BC]

Thank you.

Mr. Gagné, in your opening remarks, you suggested that Canada is falling behind the rest of the world—certainly the developed world—in addressing the challenges of AI. You also went on to talk about Canada's losing investment in this space.

I'd invite you to expand on both of those. They may be related. Maybe we're losing investment because Canada hasn't had a regulatory framework in place, or because we have taxation challenges or other challenges that scare away investment.

What are your comments, sir?

[Jean-François Gagné]

There are a multitude of aspects that come into play when you're going to evaluate. If you're going to, for instance, train one of these very large models, one that has access to data, there are currently certain issues around the copyright law that prevent companies from using text or images, whereas, in the U.S. or Japan, they can freely use them.

This has been pointed out in the past. This prevents anyone who's training these types of models from training them or operating them from Canada. What you're seeing is that Canadian companies are going to the U.S. and driving runs of hundreds of millions of dollars of training into the U.S.—not in Canada, because that prevents them from being able to innovate. I'm giving an example. My concern was that by having a blanket approach, there will be pockets of situations like that.

What I'm pointing out here is a small thing. It's just text for this particular situation, and then, boom, Canada is not playing in the large language model game. There will not be any large machine-learning data centres built in Canada, not a single one. These are multi-billion dollar investments.

[Ed Fast, Conservative, Abbotsford, BC]

Go ahead, Mr. Harris.

[Jérémie Harris]

To share a perspective on it, why is it that OpenAI, Google DeepMind and Anthropic are all based in Silicon Valley and there is no Canadian equivalent?

I'm a start-up founder veteran. I built all my start-ups in Silicon Valley; I didn't build them in Canada. I was born and raised here, and I've lived here basically the whole time. I moved to Mountain View to build my start-ups early on, and then I moved back, but I still based them there.

There are some regulatory factors. It's nice to have a Delaware C corporation, but that's not the fundamental reason. The fundamental reason these companies are based in the Silicon Valley area is just that the best investors in the world are based there. That's it. That is literally the single most important factor by far.

When I go to Y Combinator, I hear the best advice on start-up building on planet Earth. There is no equivalent to Y Combinator in Canada. This is the world's best start-up accelerator, full stop.

The best investors are in Silicon Valley, the Vinod Khoslas and the Sam Altmans. That is why this is happening there.

There are, at the margins, regulatory things going on here, but as a start-up founder who has done this multiple times and has been faced with this exact decision many times, whether it's with AI or other things, there's a kind of talent delta there in terms of the best VCs, the best angel investors. That's the ecosystem.

Tobi Lutke from Shopify started his company here in Ottawa, but there's a reason that their cap table is filled with Silicon Valley money. It's because that's where the best investors are.

At the end of the day, it's the same story over and over. I think we're just seeing it replicated in AI. I don't think there's anything too different there.