Enjoying a wonderful Milken event, the first Toronto Global Dialogues here, spearheaded by the wonderful Milken Institute. And I'm sure you've heard about the headlines about AI this weekend. So this is a great time to have this conversation. And I have to say, from Toronto, a girl that grew up in Toronto, I am so, so enthused, and it's just so wonderful to see the vibrancy and the capitalism and the, I would say, enthusiasm to build out Canada and this wonderful, wonderful city. So this panel, I don't really need to introduce these individuals because I think everybody in the country, really around the world, know who these individuals are. Of course, the brainchild of Cohere, the man that built the architecture for Google Brain, which turned into virtually every other LLM on the planet. We have Aidan Gomez of Cohere, the CEO of the largest company in Canada, RBC, a former programmer himself, Dave McKay, and I believe they're still trying to clean up your code from those days. Also at the end, the Minister of AI, a man who was really wonderful to me at the CBC when we worked together, Minister Evan Solomon.
So, I think we have to start off with this conversation talking about what's happening in the world of AI. You've seen the stock markets today. We're looking at a sell-off because people are concerned about maybe a slowdown taking place when it comes to artificial intelligence. So, I'll start with Aidan. Of course, you heard Dario Amodei calling for a slowdown, need safety guardrails. But then you heard China saying we're not so sure. Trump says, 'Well, if China's not going to slow down, we're not going to slow down.' Are you concerned?
Well, I'm definitely concerned. The article from Dario basically called for a temporary and focused on his company and a couple others alleviation of antitrust laws, the laws that were set up to prevent the formation of cartels. So that should raise a huge red flag in terms of what some of the incentives behind this might be. Of course, I do think that the technology is incredibly potent, very powerful. It's unequivocally the most powerful cyber weapon that has been created, and we've seen this in examples like the Hugging Face hacking incident. So I don't want to downplay the risks, but the idea that we would need to weaken antitrust law to try and solve that, I think that's really a mistake. What we need is a process by which to come up with the right regulation, and that should include things like testing of these models in advance, an independent third party that isn't isolated to any one country and isn't dominated by any one industry player. So I'm concerned both on the level of the technology is powerful, we need oversight and we need to get ahead of any potential risks, but also these big players trying to entrench their power and use regulation as a moat to keep competitors from rising up.
So basically, you called it a cartel this morning.
Yeah, yeah. I mean, like essentially that's regulatory.
That was what that regulation, like the antitrust law that is in place, is to prevent cartels forming, and they're asking, please give us a temporary alleviation so that we can coordinate amongst ourselves to keep you all safe.
I think that's a really, really dangerous proposition.
Dave, since you are building out in artificial intelligence, investing more and more, announcing a $1.44 billion fund, given that you believe innovation is in front of us, do you feel that we have enough guardrails in place to keep this technology safe?
No, I don't think we have enough guardrails. In fact, you know, in real time, we're spending, you know, as much effort building out new capabilities, and the product development cycles are accelerating rapidly. At the same time, we're scanning our software for vulnerabilities. We're using the most advanced tools, and these most advanced tools, whether it's Mythos and others, are being selectively released outside of the United States. So there is, you know, a sovereign issue around accessing the most powerful talent to make sure you're not cyber hacked or powerful capabilities. So, you know, as Aidan says, as we look at what is safe AI over time, it's uniform access to the most powerful tools to make sure you can defend your organization at the same time you can leverage the tools going forward. So, we see differentiated access. US firms get, you know, selective access to the most powerful tools, whether it's Miso, Mythos 51 now. And therefore that level playing field can create cyber risk. So I think the world is catching up, and, you know, it needs to think about, you know, sovereign frontier capabilities and a safe AI ecosystem.
So you're saying it's an unfair playing advantage then for the US side?
Currently, yes, absolutely. There's selective access to the most powerful tools, the tools that create the incidents like you mentioned and Hugging Face and others. And therefore, you know, we combine tools, and, you know, every company around the world is trying to make sure their software estate is as secure as possible. And therefore, you know, you need access to the most advanced tools to not only build product but also make sure you're cyber safe as a society and as an economy. So, it's both.
Yeah. Well, Minister, I mean, what exactly is Canada's, I guess, position when it comes to AI safety?
So, safety is the number one concern, full stop. Like, you got to keep, whether you're a bank, keeping your customers safe, for us, our citizens. And, you know, when we launched our AI strategy, AI for All, it was trust, opportunity, sovereign control. Those are the three pillars. First one's trust. Trust is safety. Can't trust this stuff, you run into a social license whether you're building AI infrastructure and digital infrastructure like data centers, which is why we launched our data center build framework with five key pillars to protect people, which we launched this summer with the Federation of Canadian Municipalities, because you need communities to lead. And then we got now over 40 companies signed on. So you got to protect, whether you're building data centers, you got to lead with safety. And this is frontier model. Look, AI, there's lots of different types of AI. AI in healthcare, there's going to be safety. AI in transport, there's going to be safety. But on the frontier models, there's real concerns.
We all share the same concern. In the AI world, this is not a new concern. When Dario came out and said that recursive self-improvement is concerning in terms of the Hugging Face incident, this was not a new issue that anybody in the...
Canada on the side of the US or are you on the side...?
Well, we're on the side of safety.
Anthropic asking for a slowdown.
No, hold on. We're on the side of keeping people safe. Like, that's the side.
And but what is the position?
Okay. Well, first you got to have legislation. So, we passed legislation to keep our privacy, our kids, and our data safe. We've got that legislation. Two, we're setting up a regulator because if you really want to regulate, you need someone to regulate. So, this is the first time we're going to have a regulator with real power. So, that's also tabled. And when it comes to frontier legislation, we've invested in our AI safety institute, another 50 million. On Wednesday, we will have a major announcement on AI safety on the technical side with Yoshua Bengio, who is the leading scientist with the Germans at our largest AI conference in Montreal. We have a major, major announcement on one aspect. And finally, there's how do you regulate this? I was on the phone with Sam Altman on Sunday to hear from them. The job of the people building is not going to be to regulate. They're going to suggest we are going to work with our counterparts. I was on the phone with the Brits. I just got off the phone with the French. We will be having a big meeting on Friday, and you heard the prime minister today talk about a technology stability board, which is like in finance where banks will know, you know, in the world of finance there are regulators that are in place inside financial institutions to make sure there's compliance. So look, this is evolving very quickly, but first and foremost, know that we got to protect our own sovereign environments, and we are.
Dave's made a really interesting point. The models are evolving. And they're spending a huge amount of money on patching. And this is why I will say Aidan's on stage. This is not an accident. We happen to be one of four countries that have frontier, like, large language models. The US has a bunch, China's got a bunch, France's got one, and Canada has one, and it is Cohere. We declared it when we did an MOU with them a strategic asset for us. We contracted with them. RBC has contracted with them, and the question is, and we have said we are going to champion our champions. We are going to invest...
For Aidan and Cohere then?
Well, no, but it's important that we have a Canadian model at the frontier, and we're going to make sure that happens.
Okay. Well, so Aidan, you heard Dario this weekend, and he says in his warning is that AI could take over the internet in 6 to 12 months. And did you see that? You're looking at me like...
Yeah, yeah. No, but I just think it's really bizarre.
Why do you think it's bizarre?
I think it's bizarre because he's also made the claim that there's going to be like 90% of software engineers without a job within 12 months, and he made that claim two years ago.
A year ago, almost a year ago on our network. That's right. He said that 50% of white collar jobs in 12 months.
Yeah, so I don't think you can take that seriously. But certainly there are risks around this technology. Like I was saying, the cyber threat is a real one. Contribution to bioweapons is another one, and these need to be mitigated. And there is a decision point that we're at in terms of how do we regulate these. I just hope it doesn't come from a very tiny collection of companies. You could make the argument the proposal that Dario put forward and that Sam seconded and Elon thirded benefits Cohere because for any entrenched player you're inherently in an advantaged position now. You can set up the rules and basically set the bar for competitors who might want to try and come in. But it's unethical and it's the wrong way about going about actually making the future safer for everyone.
So I think it's an important topic, and I'm glad that we're debating it in public. I just hope it doesn't...
Go down that path.
Okay. But then you heard other Anthropic former engineers say that, and that went viral across X, I'm sure everyone has seen it, that there's a 10% chance it could wipe out humanity. So your thoughts? I mean, that is not a mathematical conclusion there.
Well, there's a number, and what do you...
There is a number. You're right. But okay, you know...
Yeah, yeah. Anyone can throw out a number, but there's nothing behind it.
Yeah, I think that incident is really interesting. I suggest folks look into it. He worked at Anthropic for six weeks, and his Twitter account was created a week before, the day before the announcement, and he had exclusives with the press lined up.
Fascinating, fascinating event.
Oh yes. Dave, you were shaking your head at some of the comments. Do you have concerns that AI might wipe out humanity? 10% probability.
AI can be used for good and it can be used for bad. I don't know how deep it can go before we catch it and stop someone who had a malicious intent to that degree. So I think Aidan would have a much greater knowledge whether that's real and we couldn't detect that. What I see on the capability side is we are developing new products and our development cycles are now a fraction of what they used to be. Things that took us two years to build are taking us two months, three months. I mean, it's not just coding. We went from coding and getting a 20 to 30% efficiency in generating new code, which it's true, my nickname was Infinite Loop and I used to produce [laughter] a software code. Now I can't go back to that, but it's not just coding. We got the coding efficiency, it's the entire product design life cycle is being compressed significantly. And we just sat down and improved a completely new product from scratch last week, and I was expecting to hear it was going to take two years before we're in market end of 27, 28, we're in market in early 27 with a complete new build with everything done. That's just unheard of, and that cycle is going to continue to compress. So when you think about bringing society with us, where I worry more, will we change more jobs than we create? If you start every industry from sciences, pharmaceuticals, industrial engineering, finance, if you start creating new products and new cycles that fast and you launch these, you're going to need to pivot people from here to here to manage those. The job creation cycle can accelerate. So you've got both sides of this trying to manage the human effort through this and making sure that society comes with us on this journey, because if society opts out, you see in the United States, you know, communities are opting out of the infrastructure around AI and the power consumption, the water usage around these large factories. Well, we don't want society to opt out of all the benefits we can get in life cycle design. So how we bring everyone together to me is where a lot of the risk lies.
Job security is obviously a big risk that we highlight day in, day out, is probably the most talked about topic when it comes to AI. So from the RBC perspective, you know, they say that 50% of white collar jobs could be erased in a year or two. What are your thoughts?
Definitely not in a year or two. They're going to be changed, and you're going to adapt these tools to be a lot more productive in your role. So as we look at deploying this in every aspect of a business, we're thinking about the tasks we can do with an agentic workforce. We're thinking about value chains that we can decompose and recombine around the value chain. So there's different orders. We have kind of three pillars that we're going after this. At the end of the day, we think we can make our teams much more productive. We can create new products. Yep, the workforce is going to have to be super flexible about moving from here to here to here and leveraging these tools into your role. But when we do that, we're going to have a much more efficient society with more options, and we're going to move a lot faster. It's a, I call a huge period of pivoting and reimagination, but I don't think it's a destruction.
So it's a, I guess maybe...
We're reimagining our economy. We're going to grow our economy. We're going to produce new drugs. We're going to teach people differently. We're going to move differently in society. But it's a big reposition where you need more infrastructure to build along different lines. It's a massive pivot. And with these pivots that we've seen in history, they can be disruptive. And if you don't do them well, they lead to conflict. But if you do them well, and maybe slowing down some of this is important to make, okay, we're getting out to a front faster. We got to bring an element of our society with us. Let's create a safety net to do that so we can pivot. Each country is going to have to judge at what pace is this going, and are we bringing a large part of our constituency with us or not?
New iteration. So from the government side, are there going to be social guardrails when there's disruption to the workforce? Evan?
Yeah, let me just say on the workforce and on the safety. No, no, it's good.
The framework is how do we prepare for this change, right? I mean, government's job is not to figure out what predictive models, right? Oh, you know, we got to be here and set up a society that no matter what the change is, we can take advantage of the benefits. We take these safety concerns seriously. Some people say, do you believe it's 10%? These are real concerns. But so for us, and this whole day today at this summit, this is about not to build or not to build, and we should, you know, should we stop building AI because of the danger or not. It's not to build or not to build, it's build right or build wrong. That's it. If you build wrong, you can't get the social license for a data center. If you build wrong, you might get dangers, right, and the risks go up. We've seen this in every tech phase. You build automobiles wrong, people get hurt on the roads, and then you have Unsafe at Any Speed and Ralph Nader, and then suddenly you say, 'Yeah, you got to put seat belts and airbags,' and cars get safer and people want to buy them. You don't stop building, you build them safer. And we are at that moment in AI. We are at that moment in technology where you got to either build right or build wrong. We have seen the build wrong. We have now been warned that building wrong has risks. So the question is, how do we build right? So how do you do it? One, we talked about the data center. You got to build right community benefits, water, environment, transparency, making sure that ratepayers don't get dinged on someone else's innovation. That's really important. On jobs, how do you prepare a workforce? So last week in Edmonton at Amii, at our national AI institute, we launched Canada's first national AI literacy strategy, a free AI literacy course to every student in the country, every educator. And later this fall, every single Canadian will have access to a free AI literacy course. You need the best educated workforce. You need people to have literacy.
We will also have 90,000 jobs for young people to be in small companies that we're paying for those job placements. That not only helps adoption, but it creates jobs. Three, we are going to announce later this fall, as we said we will, an AI agent for every post-secondary student in the country, four or five years, free, protecting their data, protect. So they have the tools to be the most educated workforce. Okay, education, literacy, investing in students, investing in teachers, investing in Canadian sovereign technology. That's working with partners. So like, you got to build right now. Now, what's the future of jobs? You're going to have 10 experts tell you 10 different things. Okay? You want to take the fear... and someone says, 'Well, that didn't turn out, so dismiss it.' We don't know. Every technology has had a disruptive cycle, and there's been new... you can't predict what the next new jobs are.
