Industry Committee on Feb. 7th, 2024

[John Weigelt, National Technology Officer, Microsoft Canada Inc.]

We are in active conversations with governments across North America regarding how we make sure our elections are safe and how we ensure that we stamp out disinformation. We also have strong connections with defence here. We see they are putting in place AI programs that have responsible safeguards and tools to ensure that there's proper human oversight in this new world of conflict.

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

Is there a possibility of overlap between the roles of National Defence, personal security, the application of AI and the threats we face? Does this need to be studied further?

[John Weigelt]

I believe absolutely it needs to be studied further.

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

Thank you.

[The Chair, Liberal, Joël Lightbound]

Thank you, Mr. Vis.

Mr. Turnbull, you have the floor.

[Ryan Turnbull, Liberal, Whitby, ON]

Thanks, Chair.

Thanks to all the witnesses for being here today. I'm really finding the exchanges valuable. The insights and expertise you're sharing are making a big contribution to this conversation today, so thanks for being here, all of you.

I think your companies are all very large, very profitable and very successful, in part because of what my colleague Mr. Sorbara said, which is that you're generating a lot of value for your customers. At the same time, I think all of you are trying to demonstrate leadership in responsible AI and the use of responsible AI. That's great.

As legislators, of course, we have a big role to play, and we have to make decisions based on what's in the public interest. I think we're partners in that conversation, so I appreciate our ability to work together and your candour in the comments you've made.

Earlier this week, we heard from Yoshua Bengio, who is sometimes referred to as the godfather of AI. He was quite expressive of the exponential benefits and risks that are growing as AI evolves. This highlighted, at least from my perspective, the need for speed in getting this legislation through Parliament.

Notwithstanding that there may need to be some amendments and some changes, do all of you agree that as the Canadian government, we need to act with speed to make sure this legislation gets done?

Can I ask each of you for a quick answer?

Go ahead, Ms. Foster.

[Nicole Foster]

To be honest, I don't know that the legislation changes much about how we'll be approaching responsible AI. It will change the level of compliance and the complexity we have to comply with. It may, in one way, divert resources towards compliance and away from responsible AI development, but that would not be a reason not to legislate. I don't think any of our companies are slowing down our efforts to ensure the responsible deployment of this technology, and we continue to rapidly innovate and invest to ensure that we're doing the right things.

I think some of the existential risk is very theoretical, and I think we're very focused on some of the real risks that need to be mitigated in how AI is deployed today. We continue to invest in determining what the next iteration of AI requires from AI developers. How do we manage some of those emerging risks around hallucination and toxicity, and how do we develop appropriate red teaming and safety testing for these generative AI models?

I don't think it will change how we approach responsible AI, but it might change—

[Ryan Turnbull, Liberal, Whitby, ON]

I'm trying not to interrupt you. That wasn't a short answer, but I appreciate that I'm asking you to answer a tough question with a short answer. We politicians do that a lot, I find, but I didn't do it intentionally.

I'm just trying to get a sense of whether you agree that speed is necessary. I take it from what you're saying that you almost don't need government legislation because you're already responsible in AI development and usage. We could probably beg to differ, and you have said already that that doesn't mean governments shouldn't legislate.

I'll move along the line to Ms. Curran.

[Rachel Curran]

Yes, I do think speed is necessary, if for no other reason than to maintain public confidence in AI and in our products and services. I think it's important to get it right and make sure that you don't step on the work that's being done by the Canadian AI ecosystem, but I think speed is a good idea. Passing something is a good idea.

[Ryan Turnbull, Liberal, Whitby, ON]

Thank you.

Go ahead, Ms. Craig.

[Amanda Craig]

We agree that there's a need to move quickly with amendments but also deliberately. My colleague mentioned earlier the importance of consulting with other industry sectors. That is one thing we think would be incredibly valuable because of the breadth of impact, especially to lower-risk systems and how they will impact Canadian businesses across sectors.

[Ryan Turnbull, Liberal, Whitby, ON]

Thank you.

Ms. Patell, I see your hand up.

[Jeanette Patell]

We would go to the proverb “If you want to go fast, go alone; if you want to go far, go together.” We see this as a real global opportunity, and we want to go far together, so I think this is one of those areas where we need to collaborate with international partners to get it right.

[Ryan Turnbull, Liberal, Whitby, ON]

Great. Thank you for your comments there.

I'm going to ask another general question. How many of you have adopted the voluntary code of conduct on the responsible development and management of advanced generative AI systems?

Ms. Foster, please give a short answer.

[Nicole Foster]

Which code of conduct is this? Is it the Canadian one or the G7 one?

[Ryan Turnbull, Liberal, Whitby, ON]

It's the code of conduct that the Government of Canada announced and produced after consultation.

[Nicole Foster]

We have aligned to the White House commitments and look to see those internationalized through an international consensus.

[Ryan Turnbull, Liberal, Whitby, ON]

That's not the Canadian code of conduct but the American one. Got it.

Go ahead, Ms. Curran.

[Rachel Curran]

We asked to be included in the Canadian code. We were told it was for Canadian companies only. We were one of the first to sign on to the White House commitments. We are signed on to other voluntary codes that are similar.

[Ryan Turnbull, Liberal, Whitby, ON]

Thank you.

Go ahead, Ms. Craig.

[Amanda Craig]

We're also focused on the effort to define a global standard or global code of conduct through the G7.

[Ryan Turnbull, Liberal, Whitby, ON]

Okay.

Ms. Patell, is it the same for you?

[Jeanette Patell]

Similarly, we're focused on international entities like the G7. That's an area where we can work with Canada, which is also active in the development of that code of conduct.

[Ryan Turnbull, Liberal, Whitby, ON]

Thank you.

I'm going to switch tracks here a bit. All of you have made comments on the definition of harm and defining material harm, but all of you have described slightly different ways that you determine that yourselves internally. I think that's what I heard. I can repeat back some of the things I jotted down, but it sounded like there was a slight variation in how you assess that internally. Most of you have said that a lot of what you consider to be a high-impact system depends upon the use case.

There are two questions here. Maybe I'll start with the use case question, because I think it's probably the most difficult one. My feeling is that if we were to try to predict all of the various use cases....

Ms. Curran, you said that we should identify the use cases we're concerned about and then, I think you said, identify the threshold of harm, if I'm not mistaken. I find that as regulators and legislators, it would be very difficult to determine all of the various use cases. I'm sure you can't predict use cases either. What I'm struggling with is how that is a real approach for legislators to take. Could you respond to that?

[Rachel Curran]

I think you can do both things at the same time. You can set a materiality threshold that's broad and potentially applicable to an infinite variety of use cases and also outline specific use cases that you are concerned about, including election disinformation and the provision of health care services or employment services. You can have an overriding threshold of materiality that applies to a broad range, an infinite range, of potential use cases.