Industry committee  I'm just going to echo what she said, really. I think it's what you see in, again, the White House executive order context. There is a reason that everything is being designed around disclosure—tell us what tests you've run and what the results were—and setting up infrastructure to reveal that and to train regulators to understand the results of those audits.

Industry committee  Yes. I strongly agree. I think that's essential. One lens that I'd bring to this is national security. You know, you build these systems and you don't have to deploy them for them to be tempting targets for theft and exfiltration. I can say from first-hand experience in certain contexts that the labs that are leading the way here are not resourced to withstand a sustained exfiltration campaign from nation-state attackers.

Industry committee  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.

Industry committee  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.

Industry committee  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.

Industry committee  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.

Industry committee  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.

Industry committee  To land the plane and give you concrete timelines here, when you talk to folks in the frontier labs, the median, totally reasonable supposition they'll give is that they think it could be anywhere from two to five years to reach human-level AI across the board. Let's say they're completely off....

Industry committee  I don't want to be flippant about this, but I think it's clearly bigger than that. Again, the one thing that distinguishes human beings from other species on this planet is the thing that we are right now trying to figure out how to bottle up on a bunch of servers. Yes, there's huge potential, but great power...great responsibility.

Industry committee  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.

Industry committee  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.

Industry committee  You're quite right. That's part of the paradox here. We're talking about intelligence, as was highlighted earlier. This is the most general purpose thing humans have. It may be the most general purpose tool ever.

Industry committee  I just want to circle back to this notion of whether you regulate the model or the end applications. That is pretty central here. We're going to have to walk and chew gum at the same time. There are risks that, irreducibly, come from the model. Look at OpenAI's ChatGPT, for example.

Industry committee  It was indeed me.

Industry committee  I think those are excellent questions. I think, fortunately, we're not without tools for dealing with them. To piggyback off the testimony that Jennifer just gave, I think it's actually quite right to ask, “How can we massage this into a form that fits within our legal frameworks?”