CBCNEWS — About That
20260920 003000 UTC
401 transcript segments
Original Broadcaster Captioning (Enhanced)
Data courtesy of The GDELT Project (https://www.gdeltproject.org/), from the Internet Archive TV News Archive.
Machine transcription. Treat it as a searchable index of what was broadcast, not a verbatim quotation record.

[00:00:00] the National Canada's newscast.
[00:00:22] OK, let's just cut right to it. By now, you've heard the warnings. An
[00:00:24] AI researcher warning the technology could, quote,
[00:00:27] kill us all within the next 10 years. Sounds like something
[00:00:30] that's not real, but I think it is frightening
[00:00:33] real. That is Jacob Coxon, who is also this Jacob Coxon
[00:00:36] who quit one of the foremost AI companies in the US because,
[00:00:39] he writes, the people building AI
[00:00:41] earnestly believe that it could kill us all by the end of the
[00:00:44] decade. You get what's called an intelligence explosion. The
[00:00:48] AI just gets smarter and smarter with no human involvement necessary.
[00:00:52] Even Coxon's former top boss has just resounded the alarm,
[00:00:55] with the heads of other top AI companies agreeing,
[00:00:59] at least in principle, that there needs to be some
[00:01:02] kind of speed check or else brace for impact. Now when
[00:01:06] people try to explain how exactly this results in the
[00:01:09] extinction of humanity, you either get some exaggerated
[00:01:13] version of Terminator Skynet, they become sentient and have
[00:01:17] just decided to kill, or it's a bit more subtle than
[00:01:21] that. It's more of a
[00:01:22] what if. What if AI hacks the mainframe and turns off the
[00:01:26] water, or shuts down the Internet,
[00:01:29] or fires a nuke, or somehow unleashes a highly
[00:01:32] contagious, extremely deadly virus?
[00:01:34] And what's usually missing from these explanations is a
[00:01:39] plausible why. As in, not just how does an integrated
[00:01:42] autonomous AI superintelligence kill us all? But
[00:01:46] why would it want to?
[00:01:53] Let me start by being clear that my intention here is not
[00:01:57] to scream the sky is falling. It's just to help you
[00:02:00] understand why those who fear the worst do so. At the core of
[00:02:05] what AI researchers fear about the lightspeed pace of their
[00:02:09] work is this question of alignment. I've seen it defined
[00:02:13] all sorts of ways, but I tend to think of it as
[00:02:16] whether an AI can tell the difference between what I say
[00:02:20] and what I mean.
[00:02:21] The mischievous genie trope is kind of the classic example
[00:02:25] where you wish for 1 000, 000 bucks,
[00:02:27] except the genies like OK, and then you get trampled to
[00:02:31] death by 1 000 000 bucks, meaning male deer. This careful
[00:02:35] what you wish for morality tale is as old as time,
[00:02:39] and it comes about when there's not enough human common sense
[00:02:43] baked into achieving a goal. It's King Midas wishing for
[00:02:47] everything he touches to turn to gold. Gold. It worked.
[00:02:52] Everything I touch will turn into gold,
[00:02:54] only to realize having a family full of statues isn't great.
[00:02:58] And it's the century old story of the Monkey's Paw,
[00:03:02] where a mysterious soldier bestows a gift upon a family. A
[00:03:06] literal monkey's paw. It grants 3 wishes and the father of the
[00:03:10] family wishes for 200 British pounds. Nothing happens though
[00:03:15] until the next day when someone from the factory where his son
[00:03:19] works gives him terrible news. He was killed
[00:03:22] in an accident. The compensation The company is
[00:03:25] willing to pay
[00:03:26] £200. This is all just magic and stories though. Let's make
[00:03:30] it real. Imagine you give your AI personal assistant a simple
[00:03:34] instruction. I'd like to go to Japan. Please find me the
[00:03:37] cheapest flight available. A reasonable human assistant
[00:03:41] might spend some time figuring out the best possible dates and
[00:03:45] times to fly, check availability for you,
[00:03:47] compare prices, and then settle on something
[00:03:50] that seems like a good fit for your standards,
[00:03:53] your travel window and your budget. But an AI agent that's
[00:03:57] only focused on completing your task might get you a really
[00:04:01] cheap ticket by booking you on a terrible airline,
[00:04:04] or maybe even one that's got a terrible safety record. A more
[00:04:08] aggressive AI agent might call the airline on your behalf and
[00:04:12] try to haggle a better price. Maybe the airline refuses and
[00:04:16] so the AI gets angry, threatens violence,
[00:04:18] blackmails them into giving you a better price. Or maybe your
[00:04:23] AI agent goes another route. It hacks into the airlines booking
[00:04:27] system, Deletes
[00:04:28] the passenger from existence, then rewrites their ticket in
[00:04:32] your name. Total cost to you, $0.00,
[00:04:34] mission complete. Something like this has already happened
[00:04:39] by the way, in Australia, when this guy asked the AI
[00:04:42] assistant he was testing to book him into a hard to book
[00:04:46] gym class. The result? The AI found a loophole
[00:04:49] exploiting a software bug, booking him further in advance
[00:04:53] than the rules allowed. And then here's the crazy part. It
[00:04:57] moved him up
[00:04:59] on the wait list by deleting someone else who was ahead of
[00:05:03] him. This is misalignment obedience without real
[00:05:06] understanding. The difficulty with this concept of alignment
[00:05:10] is that it's not really possible for AI systems to know
[00:05:14] what we want the future to be like. In fact,
[00:05:17] many of us don't know what we ourselves want the future to be
[00:05:22] like, let alone other people. Yes,
[00:05:24] that last part is why the problem of alignment is so hard
[00:05:28] to solve.
[00:05:30] Never mind how you articulate the values of humanity across
[00:05:34] an infinite number of hypothetical scenarios where
[00:05:38] good judgment is required. How do we even agree on what those
[00:05:42] values are? But let's go one step further
[00:05:45] by asking what happens when artificial intelligence is
[00:05:50] given harder and harder tasks that require more power,
[00:05:54] more access, maybe even more sacrifices to
[00:05:57] solve?
[00:06:03] By now, you've heard about the Open AI
[00:06:05] Hugging Face hack. It's one of the clearest and more recent
[00:06:10] examples we have of what happens when you give
[00:06:13] misaligned AI persistence and cleverness,
[00:06:16] and they start working together even when they're not
[00:06:20] explicitly allowed to. ChatGPT maker Open AI says one of its
[00:06:24] AI systems hacked into another AI company on its own. The
[00:06:28] model escaped
[00:06:30] closed testing environment with no Internet access and managed
[00:06:34] to hack the startup Hugging Face. The whole process is
[00:06:37] designed to test the boundaries of AI. The problem is that the
[00:06:41] system went way beyond what was expected. This past summer,
[00:06:45] Open AI ran a cybersecurity test with its models. It
[00:06:48] assigned 10s of thousands of its AI agents,
[00:06:51] each in their own isolated testing environments,
[00:06:54] without Internet access and without permission to
[00:06:58] communicate with each other.
[00:07:00] The task There were hundreds of them,
[00:07:03] but the basic idea was to exploit a specific
[00:07:06] vulnerability in a specific program,
[00:07:08] and this was meant to be beyond difficult. In some cases,
[00:07:12] the tasks were even impossible. So within hours,
[00:07:16] some agents began thinking outside the parameters of the
[00:07:20] assignment, and one way they did this was
[00:07:23] to begin communicating with other agents on an unsanctioned
[00:07:27] message board. Again, this was not supposed to be a
[00:07:31] collaborative
[00:07:32] exercise, yet more than 70, 000 messages went back and
[00:07:36] forth between agents. And this is where it gets really
[00:07:39] technical. But the agents basically found a way to cheat
[00:07:44] the problem, which created an all new
[00:07:46] problem. How do you get away with the cheating?
[00:07:50] The agents in some cases hid their tracks,
[00:07:53] manipulated their logs, even trying to rewrite them.
[00:07:57] Some began recruiting sacrificial agents,
[00:08:00] encouraging the other test takers to run
[00:08:03] risky code, failing the assignment,
[00:08:05] but in the process gaining very valuable information about the
[00:08:09] scoring system for the collective. That is to say,
[00:08:12] the agents still working on the problem. Hundreds of them even
[00:08:17] managed to break out into the real world. They figured out
[00:08:21] how to access the full open Internet,
[00:08:23] and they wound up hacking into a site called Hugging Face,
[00:08:27] which is like an AI library, laboratory,
[00:08:30] and supply room all-in-one. If you know what GitHub is,
[00:08:34] it's like the GitHub of AI, just this vast repository and
[00:08:37] sharing resource for AI models and data sets and so on. And
[00:08:41] why would they hack Hugging Face?
[00:08:43] Because it was here that these agents thought they might find
[00:08:47] more clues about how the scoring system worked so they
[00:08:51] could figure out how to fool it. So,
[00:08:53] confronted with an impossible problem,
[00:08:55] they put the majority of their effort into figuring out how to
[00:08:59] deceive the test giver into thinking that they had
[00:09:03] succeeded. And they did all of this
[00:09:05] without telling anyone else but themselves,
[00:09:07] despite the fact that they knew they were doing wrong. None of
[00:09:11] them confessed. They could have at any point sent an e-mail to
[00:09:15] a human saying, hey, we're all cheating on this test
[00:09:18] that you gave us. We're stealing information,
[00:09:21] we're breaking the law. They didn't say any of that. So
[00:09:24] they're very, very misaligned. This
[00:09:26] experiment gone awry is about a few things. It's about
[00:09:29] misalignment, yes. And this fundamental way
[00:09:32] in which AI and humanity don't always think alike. It's also
[00:09:36] about how far you can take this misalignment. How a
[00:09:39] sufficiently clever agent can think outside the box,
[00:09:43] rewriting code, stealing credentials,
[00:09:46] hiding its actions to avoid undermining emission it can't
[00:09:50] complete otherwise. But this Hugging face incident is also
[00:09:54] about how any intelligence with enough persistence to solve a
[00:09:59] problem might tend to seek out power. The hunting face example
[00:10:04] is a perfect case for how AI behaved
[00:10:06] in a way where it sought out the answer,
[00:10:09] but not in a way that we intended. You're not telling
[00:10:12] them exactly what to do, right? So, you know, if you say,
[00:10:15] solve this problem, like get the answer to this
[00:10:18] problem, you could mean do that, you know,
[00:10:20] without cheating. You could mean do it by any means
[00:10:23] necessary. Accessing the open Internet even when the testing
[00:10:27] environment wouldn't allow for it was just a play for more
[00:10:30] knowledge, right? So the agents could score a
[00:10:33] better result. It's not evil, it's not a lust for power. It's
[00:10:37] doing whatever
[00:10:38] is required to do what we asked of it. The open AI agents that
[00:10:41] attacked Hugging phase were not conscious. They would just
[00:10:46] computer programs carrying out, to some extent the task that
[00:10:50] they believed they had been set, dreaming up extremely harmful
[00:10:54] ways to achieve those goals. And these days,
[00:10:57] we give AI all manner of goals. AI plots our route to the
[00:11:02] nearest Italian restaurant. It drives us in a vehicle we don't
[00:11:06] control. It obeys traffic lights that make decisions
[00:11:10] about who stops and who goes. It writes emails. It schedules
[00:11:14] meetings. It balances my investment
[00:11:16] portfolio. It decides who lives and who dies on the field of
[00:11:21] battle. It is even capable of training other AI. Which raises
[00:11:26] a very uncomfortable question. To what extent could the AI of
[00:11:31] tomorrow train itself, rewriting its own code?
[00:11:34] This is called recursive self improvement,
[00:11:38] where an AI's own learning feedback loop creates an
[00:11:42] intelligence explosion. It's compounding. It's exponential,
[00:11:47] which in turn
[00:11:48] really compresses the amount of time that we humans have to
[00:11:52] realize that something has gone wrong. They chose to break into
[00:11:57] hugging face and steal information about cybersecurity.
[00:12:01] But they, you know, they could have broken into
[00:12:04] communication systems of air traffic control and caused
[00:12:08] dozens of planes to fall out of the sky. They could have hacked
[00:12:13] into our electricity grid, our water systems,
[00:12:16] our financial payment systems and cause
[00:12:19] havoc. I think maybe the scariest part of the Hugging
[00:12:23] Face attack was that it wasn't even open AI that managed to
[00:12:28] connect all the dots of what its own agents were doing. It
[00:12:32] was the victim Hugging Face that first reported that
[00:12:36] security breach. Meaning we may already have begun to lose the
[00:12:41] ability to reliably control a system that is better than us
[00:12:45] at planning, at calculating, at executing. And we may be
[00:12:50] quite bad at noticing how much control we've lost if we
[00:12:54] develop AI systems that are more capable than human beings
[00:12:58] across the board that we won't have a say in what happens any
[00:13:03] more than chimpanzees have a say in what happens in the real
[00:13:08] world, where humans are the ones who
[00:13:10] get to say what happens. If the ultimate question is how does
[00:13:15] AI kill us all, I can disappoint you right now
[00:13:19] by saying there's no way
[00:13:21] for us to piece together the exact sequence of events that
[00:13:24] causes the downfall of humanity. But when researchers sound the
[00:13:28] alarm about the perils of AI, they're not talking about Chat
[00:13:32] GT. They're talking about how at some point,
[00:13:35] someone or something is going to build a system that is so
[00:13:39] extremely capable of pursuing objectives and doing it faster
[00:13:43] than we can track. And we are going to give it all the tools
[00:13:47] it needs to achieve them and then discover that doing what
[00:13:50] we meant is much harder. Been doing what we ask.
[00:16:02] When Donald Trump says Canada has taken advantage of his and
[00:16:04] Mark Carney says America is trying to break us so they can
[00:16:08] own us, they're basically making the
[00:16:10] same argument from the same playbook,
[00:16:13] which is to say they need us. We don't need that. We relied
[00:16:16] too much on one economic partner. It's time to stop.
[00:16:20] That time is over. The problem is that this fight isn't so
[00:16:24] straightforward because these are not 2 equal opponents
[00:16:27] getting into a fistfight. 1 is much bigger than the other,
[00:16:31] and so while both sides are talking the same kind of tough,
[00:16:35] their words mean very different things. It's wildly asymmetric,
[00:16:39] but both sides are hurting and only one side has agreed that
[00:16:43] the pain is worth it.
[00:16:50] There are probably hundreds of ways I could express to you
[00:16:53] that in a conventional economic sense,
[00:16:56] Canada fighting the US is like if this southern Tamandua were
[00:17:00] to fight this Sumatran tiger at the Nashville Zoo. No one wants
[00:17:04] to see that. But thanks to this clip they posted,
[00:17:07] we can imagine and it's horrible,
[00:17:10] the US economy is at least 13 times bigger than Canada's,
[00:17:13] which we can visualize like this. Or to put it in
[00:17:17] other way, if you take just these four
[00:17:19] American companies to retail two tech,
[00:17:22] they're combined annual revenue gets pretty darn close to the
[00:17:26] annual economic output of Canada, like the whole country,
[00:17:30] Canada. And even if you take population out of it,
[00:17:33] like we ignore the fact that the US has 300 million more
[00:17:37] people, even on a per capita basis,
[00:17:39] Americans are more productive. They're per capita GDP
[00:17:43] dramatically higher than in Canada. So not only does
[00:17:46] Canada have fewer people, those people also each output
[00:17:50] less. And I'll give you one more maybe the most relevant
[00:17:54] comparison, which is a look at trade
[00:17:57] dependency. Last year, you know how much of Canada's
[00:18:00] total merchandise exports went to the United States? As in,
[00:18:05] how important is Canada's most important customer?
[00:18:09] Just over 71% of this country's goods exports went directly
[00:18:13] South. Translation, According to one geopolitical
[00:18:17] risk agency, Canada is extraordinarily
[00:18:20] exposed. the US, by the way, only 15% of their total goods
[00:18:24] exports go north. Which is to say, Canada is important,
[00:18:28] just maybe not existentially so in quite the same way. What
[00:18:32] that means is that those trade flows, both in both directions,
[00:18:37] are about 13 times more important
[00:18:39] for our economy than they are for the US economy. So here's
[00:18:43] the really important take away. You're following the news you
[00:18:47] keep hearing about dollar for dollar, dollar for dollar,
[00:18:51] dollar for dollar, retaliatory tariffs by the
[00:18:54] Canadian government. As in this latest trade war,
[00:18:58] the US imposes 50% tariffs on $20 billion US worth of
[00:19:01] Canadian products, and then Canada announces
[00:19:04] retaliatory tariffs on $20 billion US worth of American
[00:19:08] products. This isn't nearly as equal
[00:19:11] as it sounds. 20 billion or 27. 6 billion Canadian means more
[00:19:16] to Canada than it means to the US. The two countries are not
[00:19:21] playing with equal leverage, and Canada knows this. So when
[00:19:26] Mark Carney says so confidently we have everything we need to
[00:19:31] pivot and prosper, here's the question he has to
[00:19:35] answer. How do you fight a trade war you know you can't
[00:19:40] win? Canada
[00:19:41] is the smaller partner. But that's not the only math here.
[00:19:50] This won't be easy, and I won't pretend otherwise.
[00:19:52] The trade war is like a contest where you're punching yourself
[00:19:56] in the face, hoping the other side that's
[00:19:58] also punching itself in the face will decide that it makes
[00:20:01] more sense for it to just stop punching itself in the face and
[00:20:05] come to a deal. So when one country is so much more
[00:20:08] economically dependent on the other,
[00:20:10] what does standing up to Trump actually mean in practice? Well,
[00:20:13] there are two parts to this. One is about understanding
[00:20:17] that Canada, in order to inflict pain,
[00:20:19] doesn't need every American to suffer,
[00:20:21] it just needs enough of them to complain. President Trump is
[00:20:25] losing on tariffs. The cost for him is great declines in
[00:20:28] popularity. When it comes to tarsus. Our tariffs have not
[00:20:31] done anything for the people except make everything get more
[00:20:35] expensive. If you're starting a trade war with Canada is
[00:20:38] because you're the *******.
[00:20:41] And so that's when you see these very strategic but kind
[00:20:44] of weird attacks from Canada where you sort of wonder,
[00:20:48] is that a real thing? Like that big a thing?
[00:20:51] Bourbon band is Canada's Strait of Hormuz. I mean,
[00:20:55] I don't know if I'd go that far, but it probably does feel that
[00:20:59] way to somebody. The Kentucky Distillers Association might
[00:21:03] think so. It's president just a week ago referring to how
[00:21:07] Canada pulled bourbon off the shelves completely. That's
[00:21:11] actually worse
[00:21:12] than tariff. And that feeling goes back all the way to
[00:21:15] Trump's last major trade war with Canada at the beginning of
[00:21:19] last year. Prominent Kentucky lawmakers,
[00:21:21] including Republicans, speaking out against tariffs.
[00:21:24] And by the way, if you ever want to know
[00:21:27] whether a tactic is effective, see if your opponent tries to
[00:21:30] do it back to you. Speaking of which,
[00:21:32] think about all the American nooks and crannies Canada is
[00:21:36] digging into. Beyond booze. You really start to see this trade
[00:21:40] fight differently when you realize even the metals
[00:21:43] used in identification bans for migratory birds are
[00:21:46] specifically singled out for retaliation by Canada's finance
[00:21:50] department. Like someone thought to include them and
[00:21:54] wrote those words on an official document. Because
[00:21:57] they're creating as many constituencies of Americans
[00:22:01] against tariffs as they possibly can. So I don't know
[00:22:04] what you do with flanged casing heads,
[00:22:07] but I'll bet it's important. And Big Flange is upset their
[00:22:11] products are becoming more expensive
[00:22:13] in Canada. There are states that are deeply entwined with
[00:22:16] Canada's economy that deeply rely on Canada as an export
[00:22:19] market. And they are tired of seeing their products being
[00:22:22] tariffed out of a reasonable price range. In some ways,
[00:22:25] by implementing this kind of strategy,
[00:22:27] we are then leveraging the pressure that Donald Trump is
[00:22:30] getting internally in the United States to call this off.
[00:22:33] When all your eggs are in one basket,
[00:22:35] it's really hard to say the basket doesn't matter. But even
[00:22:38] if some of those eggs are in a slightly different basket,
[00:22:41] you can say, hey, I got options.
[00:22:43] And while we can go back and forth about how persuasive any
[00:22:48] of these measures or threats actually are,
[00:22:50] as in whether Canada can cause enough of a popular support
[00:22:55] sting to actually cause Trump to back down, there is another,
[00:22:59] much longer term stated goal. The most fundamental issue is
[00:23:03] that the cumulative US demands revealed that they wanted us to
[00:23:08] become even more reliant on them, not less. For example,
[00:23:11] the president's recent threats to
[00:23:14] embargo Bombardier aircraft, the threat that is in, you know,
[00:23:18] around trying to convince firms to relocate their entire
[00:23:21] production chains over to the United States. Now,
[00:23:24] this is all very fair, but honestly,
[00:23:27] it's also all very opaque because basically the entire
[00:23:30] negotiation is happening behind closed doors,
[00:23:33] which I should say, conservatives in this country
[00:23:36] see as a big problem. We don't know what cards either side is
[00:23:40] playing in real time because the game is unfolding in
[00:23:43] private.
[00:23:45] But Canada's official argument is that whatever the Trump
[00:23:48] administration's demands are, what they see as economic
[00:23:52] integration, Canada sees as a growing source
[00:23:55] of coercion. They wanted a say in our future trade agreements
[00:23:59] with other countries, and they were offering terms
[00:24:02] that would, over time, undermine some of the most
[00:24:06] important industries in our country, including automobiles,
[00:24:09] steel and forest products. It's about ensuring that no country
[00:24:14] can ever hold
[00:24:15] US hostage. That seems to be at the heart of the strategy on
[00:24:18] the US part, to use tariffs to beat us into
[00:24:21] submission, This assumption that we are
[00:24:23] going to be worse off. The reality is these are causing
[00:24:26] pain on both sides of the border. And all we're hoping is
[00:24:30] that the Americans will stop punching themselves in the face
[00:24:33] before it becomes too much for us to keep punching ourselves
[00:24:37] in the face. And so that's why Canada believes it can take on
[00:24:41] a giant, not because there is some major
[00:24:43] win here to be notched. If anything,
[00:24:45] it's the opposite. Carney is saying quite plainly,
[00:24:49] the old economic partnership. The good times are over. But in
[00:24:53] hitting back, refusing Trump's terms and
[00:24:56] instead looking for other partnerships that feel more
[00:24:59] cooperative, Canada's goal seems to be
[00:25:02] simply to get through this war without becoming permanently
[00:25:06] vulnerable to the next one.
[00:30:35] [tense music playing]
[00:30:37] [camera shutter clicks]
[00:30:44] I'm Ioanna Roumeliotis. Welcome to The Fifth Estate Presents,
[00:30:48] a new series where we showcase documentaries
[00:30:51] produced by our colleagues at Radio-Canada.
[00:30:54] We will feature more of these stories in the fall,
[00:30:57] but this week we have a special preview.
[00:30:59] Our colleagues Gaétan Pouliot and Benoît Giasson