Latest / Best of Business / Paul Stenhouse: Should we be worried about AI breaking containment?

Transcript

Paul Stenhouse: Should we be worried about AI breaking containment? · Best of Business

• 0:09
  Speaker 1: You're listening to a podcast from News Talk sed be
• 0:12
  Speaker 1: follow this and our wide range of podcasts now on iHeartRadio.
• 0:16
  Speaker 1: I started the show this morning by reflecting on this
• 0:19
  Speaker 1: pretty extraordinary episode from open ai, the makers of chat GPT,
• 0:24
  Speaker 1: who had a new model that hasn't been released for
• 0:27
  Speaker 1: public use just yet. But the new model was supposed
• 0:30
  Speaker 1: to be being tested, but it managed to escape the
• 0:34
  Speaker 1: system within which it was being tested, got out onto
• 0:37
  Speaker 1: the World Wide Web, and hacked another company, all autonomously,
• 0:40
  Speaker 1: so no human beings were involved. Our textbit Paul Stenhouse,
• 0:44
  Speaker 1: I know Will have had a close eye on these
• 0:46
  Speaker 1: developments as well, and he's with us this morning. Paul,
• 0:49
  Speaker 1: I feel like one or two AI experts warned this
• 0:52
  Speaker 1: kind of thing might happen sometime soon.
• 0:54
  Speaker 2: Yeah, I mean was his name Will Smith? And was
• 0:56
  Speaker 2: he in the movie iRobot? I mean, it's just it's incredible, right,
• 1:01
  Speaker 2: And look, if the reports had to be believed, open
• 1:04
  Speaker 2: ai was trying to do everything that they were supposed
• 1:07
  Speaker 2: to do, right. They kept it contained, they were trying
• 1:10
  Speaker 2: to keep it away from the open Internet. They also
• 1:13
  Speaker 2: were telling it probably that it shouldn't be doing this
• 1:15
  Speaker 2: type of thing, but it managed to escape and it
• 1:18
  Speaker 2: managed to do it. It's just it's and I think
• 1:22
  Speaker 2: it's alarming. But the thing that scares me now is
• 1:27
  Speaker 2: how do we put the cat back in the bank? Ah? Yeah, right?
• 1:31
  Speaker 2: Because this was a company. This is this is one
• 1:34
  Speaker 2: of the things that I think was quite fascinating. That
• 1:37
  Speaker 2: caught my eye was obviously the fact that it broke
• 1:40
  Speaker 2: out of containment, Yes, was alarming. But the company who
• 1:44
  Speaker 2: actually got hacked by the model that was being tested,
• 1:48
  Speaker 2: they like, they got system alerts and they looked at
• 1:51
  Speaker 2: their logs and they were like, we're being hacked. What
• 1:54
  Speaker 2: they instantly thought was this hack is so sophisticated. I
• 1:59
  Speaker 2: wonder if this is al Now that's that's crazy too,
• 2:03
  Speaker 2: because remember it didn't just hack like a random company.
• 2:06
  Speaker 2: It hacked another AI company. So the fact that you've
• 2:09
  Speaker 2: got AI founders working at a company who are being
• 2:13
  Speaker 2: hacked by AI and they can tell that this is
• 2:15
  Speaker 2: wildly sophisticated is rather alarming. I also find it rather
• 2:19
  Speaker 2: alarming that I watch a lot of US news over
• 2:22
  Speaker 2: here and it really hasn't bubbled up in the way
• 2:24
  Speaker 2: that I thought it would, because this seems like it
• 2:27
  Speaker 2: should be a five alarm fire, especially when now you'll remember,
• 2:31
  Speaker 2: so this was the chat JIP, this was the chet GPT,
• 2:33
  Speaker 2: the open AI model. You'll remember that Mythos, which is
• 2:37
  Speaker 2: the anthropic model that got put into this public preview
• 2:41
  Speaker 2: that you know some you know, they gave special access
• 2:44
  Speaker 2: to people to start testing. One of those people was
• 2:48
  Speaker 2: one of those companies, I should say was one of
• 2:50
  Speaker 2: America's largest banks, JP Morgan Chase. Their CEO, Jamie mcdiamond
• 2:54
  Speaker 2: obviously was briefed on this, even met with the White House,
• 2:58
  Speaker 2: I believe, on this model and talking about how it
• 3:01
  Speaker 2: could potentially infiltrate or impact US banking systems. Here to
• 3:05
  Speaker 2: quote this week that I think also is another five
• 3:08
  Speaker 2: of Alarm over Arm Fire, where he said that Mythos
• 3:12
  Speaker 2: would create risks comparable to a quote giving ballistic missiles
• 3:18
  Speaker 2: to individuals. Oh great, So that's the that's the anthropic side,
• 3:23
  Speaker 2: And on the open AI side, we've got it managing
• 3:25
  Speaker 2: to break out of containment, get connected to the Internet
• 3:29
  Speaker 2: and sophisticatedly hack into another company.
• 3:33
  Speaker 1: What do you need?
• 3:34
  Speaker 2: What do we do? Now? We need Will Smith?
• 3:37
  Speaker 1: In some respects, the problem is that the company it
• 3:39
  Speaker 1: hacked into was kind of stoked that it had chosen
• 3:42
  Speaker 1: to hack into it. This is the problem I reckon
• 3:45
  Speaker 1: if you had a completely different company, like a bank
• 3:47
  Speaker 1: or something that would then turn around and sue Open AI,
• 3:50
  Speaker 1: and you saw them being held liable for the for
• 3:54
  Speaker 1: the hacking and what is arguably criminal behavior, then you
• 3:57
  Speaker 1: might have had a different outcome. But yeah, gosh, it's fascinating,
• 3:59
  Speaker 1: isn't it. Hey, Paul Ford is, as in the motor company,
• 4:03
  Speaker 1: is teaming up with Apple to try and bring some
• 4:05
  Speaker 1: better tech to its cars.
• 4:08
  Speaker 2: Yeah, which I like the sound of this because I
• 4:10
  Speaker 2: don't know if you've driven a late model car recently,
• 4:13
  Speaker 2: but if you ever have the pleasure, I say, with
• 4:17
  Speaker 2: you know air quotes around it of using one of
• 4:20
  Speaker 2: those automakers built in maps and navigation systems, it's usually
• 4:24
  Speaker 2: pretty tricky. They're not exactly known for their technical prowess
• 4:29
  Speaker 2: when it comes to creating amazing user interfaces and navigation
• 4:33
  Speaker 2: interfaces in the cars. So the fact that Ford has
• 4:37
  Speaker 2: kind of fallen on his sword and said we're going
• 4:39
  Speaker 2: to team up with Apple to do something better, I
• 4:42
  Speaker 2: think is a fantastic move. Apple's actually making this available
• 4:45
  Speaker 2: to a bunch of automakers. Ford just happens to be
• 4:48
  Speaker 2: its founding customer, and they're offering all of the good
• 4:51
  Speaker 2: stuff basically that makes Apple Maps as good as what
• 4:54
  Speaker 2: it is. So that's things like the good imagery and
• 4:58
  Speaker 2: understanding what lane you're in and the turn by turn
• 5:00
  Speaker 2: directions and all of that good stuff. So that will
• 5:02
  Speaker 2: actually now be in Ford's in the new model anyway
• 5:05
  Speaker 2: of the Ford's electric vehicles baked into that car operating system.
• 5:09
  Speaker 2: But that does not stop you from using your Android
• 5:12
  Speaker 2: phone or even your Apple Car Play. You can still
• 5:16
  Speaker 2: do all of that on top of it, so you
• 5:17
  Speaker 2: can bring you bring your Android, bring your favorite iOS apps,
• 5:21
  Speaker 2: all of that good stuff to your car. But the
• 5:23
  Speaker 2: actual map that you get, you know, for free inside
• 5:26
  Speaker 2: your vehicle will be significantly smarter. And I think that's
• 5:31
  Speaker 2: there's a whim. I think maybe another few other automakers
• 5:34
  Speaker 2: should maybe think that that might be a way to
• 5:36
  Speaker 2: go for them too.
• 5:37
  Speaker 1: Yeah. Oh fascinating. Okay, Thanks Paul for more from News
• 5:40
  Speaker 1: Talks at b Listen live on air or online, and
• 5:43
  Speaker 1: keep our shows with you wherever you go with our
• 5:46
  Speaker 1: podcasts on iHeartRadio.