Latest / DEEP INSIGHT / Trump calls AI fears 'hoax' but says he'll form 'AI force' & name 'AI czar'

Transcript

Trump calls AI fears 'hoax' but says he'll form 'AI force' & name 'AI czar' · DEEP INSIGHT

• 0:00
  Speaker 1: President Trump is once again lashing out at AI alarmists
• 0:03
  Speaker 1: while simultaneously vowing to keep the technology in check. Today,
• 0:08
  Speaker 1: the president again slamming concerns about AI's potential threat to
• 0:11
  Speaker 1: humanity as a hoax. Over the weekend, the Treasury Secretary
• 0:15
  Speaker 1: Scott Besson met with Chinese officials and pitched new safeguards
• 0:18
  Speaker 1: for both countries. President Trump later announced he would create
• 0:22
  Speaker 1: an AI force appointing an AI czar over safety concerns.
• 0:27
  Speaker 1: Sources say the president is planning to discuss AI with
• 0:29
  Speaker 1: China's President Xi Jinping when the two meet later this week.
• 0:34
  Speaker 1: Let's talk about all of this with Matt Sheehan. He's
• 0:36
  Speaker 1: the co-director for the China AI Initiative at the Carnegie
• 0:40
  Speaker 1: Endowment for International Peace. Thank you so much for being
• 0:43
  Speaker 1: with us, Matt. Are you optimistic that the US and
• 0:46
  Speaker 1: China will come out of this with some kind of
• 0:48
  Speaker 1: agreement on AI, whether on those safety notifications that the
• 0:52
  Speaker 1: Treasury Secretary was talking about, or perhaps even a broader
• 0:55
  Speaker 1: pause on innovation?
• 0:57
  Speaker 2: I'd say that we should have very low expectations for
• 1:00
  Speaker 2: what will come out of this set of meetings. But
• 1:03
  Speaker 2: I say I'm optimistic that it will hit those low expectations.
• 1:06
  Speaker 2: We saw yesterday, Treasury Secretary Besant met with his Chinese counterpart,
• 1:10
  Speaker 2: He Lifeng, After the meetings, Besant said that they had
• 1:14
  Speaker 2: proposed a notification mechanism for when AI is involved in
• 1:17
  Speaker 2: national security incidents. That type of mechanism, some way for
• 1:20
  Speaker 2: the U.S. and China to talk to each other, to
• 1:24
  Speaker 2: exchange information when there are serious safety threats, that's an
• 1:28
  Speaker 2: important baseline. That's a starting point. They also said that
• 1:31
  Speaker 2: they were setting up a sort of a longer-term mechanism
• 1:34
  Speaker 2: for dialogue, and that's also very important, the next meeting
• 1:37
  Speaker 2: likely in about two months from now. So that's a
• 1:39
  Speaker 2: good pace, but we shouldn't expect much more than that.
• 1:43
  Speaker 1: So on one end, broadly, the perception in Beijing is
• 1:47
  Speaker 1: that the U.S. wants to constrain China and Chinese progress
• 1:52
  Speaker 1: through export controls of advanced ships, for example. Conversely, Washington
• 1:56
  Speaker 1: is alleging that Chinese firms are committing industrial scale theft
• 2:01
  Speaker 1: of U.S. trade secrets, right? So where is the middle ground?
• 2:05
  Speaker 1: Do you see a path toward bridging that gap. Obviously,
• 2:09
  Speaker 1: you don't have high expectations for this week, but further
• 2:12
  Speaker 1: down the road?
• 2:13
  Speaker 2: I think that both of those things that is believed
• 2:16
  Speaker 2: in the capitals are true. I mean, the United States
• 2:18
  Speaker 2: has been imposing export controls on advanced chips to China,
• 2:22
  Speaker 2: the advanced chips that are necessary to train these AI models.
• 2:25
  Speaker 2: So the U.S. definitely is trying to constrain China's AI development,
• 2:28
  Speaker 2: even if those controls aren't as tight as they could be.
• 2:31
  Speaker 2: And it's also true that Chinese AI companies are using
• 2:34
  Speaker 2: the leading U.S. models to train their own models to
• 2:38
  Speaker 2: a practice that is seen as theft or industrial scale,
• 2:42
  Speaker 2: not espionage, but industrial scale trade secret theft. It's a
• 2:45
  Speaker 2: bit controversial, but I think both of those things can
• 2:47
  Speaker 2: be true. U.S. trying to constrain China, China trying to
• 2:50
  Speaker 2: steal advances from the U.S., and there's still a lot
• 2:53
  Speaker 2: of room to work together when it comes to the
• 2:55
  Speaker 2: safety of the most advanced models. It doesn't matter if
• 2:58
  Speaker 2: one side is constraining or stealing from the other one.
• 3:01
  Speaker 2: If a Chinese model spins out of control, it escapes
• 3:05
  Speaker 2: the control of the lab that built it, That's a
• 3:07
  Speaker 2: threat not just to China, but also to the United
• 3:09
  Speaker 2: States and vice versa. So there still are mutual interests
• 3:13
  Speaker 2: and things that we can do to try to constrain
• 3:15
  Speaker 2: the development of the models or enhance the safety of
• 3:18
  Speaker 2: the models, even if we don't agree on everything else.
• 3:21
  Speaker 3: No doubt.
• 3:22
  Speaker 1: And your point about superintelligence perhaps going awry and going
• 3:25
  Speaker 1: out of human control, notable because China does have extensive
• 3:30
  Speaker 1: regulations on AI, though up until recently, they've been more
• 3:34
  Speaker 1: focused on what you would call censorship on controlling the
• 3:39
  Speaker 1: way that information gets out on protecting young people from self-harm,
• 3:42
  Speaker 1: for example, they haven't been as focused on the idea
• 3:45
  Speaker 1: that AI may get beyond human control. So I wonder
• 3:49
  Speaker 1: what you think might change, how regulation there might possibly
• 3:54
  Speaker 1: shape rumblings of regulation here in the United States.
• 3:59
  Speaker 2: I think you described it well. China has had the
• 4:01
  Speaker 2: world's most extensive AI regulations, putting real demands on its
• 4:05
  Speaker 2: companies to do tests before they release models, to file
• 4:09
  Speaker 2: those tests with the government.
• 4:10
  Speaker 1: That's all very real.
• 4:11
  Speaker 2: At the beginning, that was primarily about making sure the
• 4:14
  Speaker 2: models didn't say the wrong thing. It didn't violate CCP
• 4:17
  Speaker 2: censorship controls. But they've expanded those to areas that you mentioned,
• 4:21
  Speaker 2: you know, AI companions, sort of emotional relationships with chatbots
• 4:25
  Speaker 2: and trying to protect kids. So they have this extensive
• 4:28
  Speaker 2: apparatus for affecting how models are built. And the question is,
• 4:32
  Speaker 2: are they going to turn that apparatus towards safety concerns
• 4:36
  Speaker 2: that are shared by, say, the U.S. labs, even if
• 4:38
  Speaker 2: not by the U.S. federal government? I think there are
• 4:41
  Speaker 2: some signs that they're moving in that direction. If you
• 4:43
  Speaker 2: look at the sort of the standards documents, the deeply
• 4:46
  Speaker 2: technical documents that come out of China's leading labs, that
• 4:50
  Speaker 2: come out of the government agencies that oversee AI, There
• 4:53
  Speaker 2: is this growing concern about large-scale risks from AI, but
• 4:57
  Speaker 2: it's just not where the U.S.
• 4:58
  Speaker 4: Labs are.
• 4:59
  Speaker 2: There's not that same sense of urgency or the decision
• 5:02
  Speaker 2: that we need to slow this down right now. The
• 5:05
  Speaker 2: question is, are they going to get there? And if
• 5:07
  Speaker 2: they do, are they going to get there fast enough?
• 5:10
  Speaker 1: We'll see what comes of this week. As you put it,
• 5:12
  Speaker 1: optimistic about low expectations. Matt Sheehan, thanks so much for
• 5:16
  Speaker 1: joining us.
• 5:17
  Speaker 4: Treasury Secretary Scott Besson says Washington and Beijing are already
• 5:21
  Speaker 4: discussing potential safeguards for the technology, like alerting one another
• 5:25
  Speaker 4: about potentially dangerous AI incidents.
• 5:29
  Speaker 5: We think that just like with any cross-border activity, that
• 5:36
  Speaker 5: moving from opaque to more transparency between the number one
• 5:41
  Speaker 5: and the number two AI powers in the world is
• 5:43
  Speaker 5: very important.
• 5:46
  Speaker 4: All right, let's go live now to CNN Business Technology
• 5:48
  Speaker 4: Editor Lisa Adachiko. Lisa, let's start with this new lawsuit
• 5:52
  Speaker 4: against several prominent AI companies. What more are you learning?
• 5:55
  Speaker 3: Yeah, Pamela. So the interesting thing about this lawsuit here,
• 5:58
  Speaker 3: really what's at the heart of the concerns here is
• 6:01
  Speaker 3: that this call for a slowdown in AI development, there's
• 6:05
  Speaker 3: accusations here that that could potentially cause harm for consumers
• 6:10
  Speaker 3: who are paying subscriptions for services provided by those companies.
• 6:14
  Speaker 3: The thinking being here that the plaintiffs allege in this
• 6:17
  Speaker 3: lawsuit that consumers, when they're paying for these subscriptions, they're
• 6:20
  Speaker 3: doing so under the impression that the individual companies, whether
• 6:25
  Speaker 3: that's SpaceX, Google, Anthropic, or OpenAI, are making independent decisions
• 6:29
  Speaker 3: about the future of their products and the pace at
• 6:32
  Speaker 3: which to continue developing them. And so the argument here
• 6:36
  Speaker 3: is that this coordinated effort could kind of threaten the
• 6:40
  Speaker 3: pace of innovation. And these companies, if there isn't as
• 6:44
  Speaker 3: much of a competitive factor here because they're coordinating on
• 6:47
  Speaker 3: a slowdown, they might not be as motivated to improve
• 6:50
  Speaker 3: their products as quickly. And I think this is an
• 6:54
  Speaker 3: interesting wrinkle to the ongoing concerns we've been hearing around
• 6:58
  Speaker 3: the future of AI and safety. And I think it
• 7:01
  Speaker 3: kind of underscores the challenges that these companies face in
• 7:05
  Speaker 3: really just continuing to maintain and push forward the products
• 7:10
  Speaker 3: that some people are coming to rely on for work
• 7:13
  Speaker 3: and other everyday tasks while also doing so in a
• 7:17
  Speaker 3: way that is responsible and safe and takes into account
• 7:20
  Speaker 3: these new unexpected instances of AI models acting, you know,
• 7:26
  Speaker 3: going rogue and acting in ways that they aren't expected
• 7:29
  Speaker 3: to during training. So I do think this is one
• 7:31
  Speaker 3: to keep an eye on for sure. And it kind
• 7:33
  Speaker 3: of underscores that broader issue.
• 7:35
  Speaker 4: I know that there's obviously been these public calls for
• 7:37
  Speaker 4: a slowdown from the AI leaders, but is there any
• 7:40
  Speaker 4: indication they're speaking privately to reach some sort of deal
• 7:44
  Speaker 4: as this lawsuit sort of indicates could be happening?
• 7:48
  Speaker 3: So there have been some discussions around the formation of
• 7:53
  Speaker 3: an industry standard. Right now, there is no broad standard.
• 7:57
  Speaker 3: There's no formal federal regulation. So there have been a
• 8:01
  Speaker 3: lot of concerns here about what needs to be done
• 8:04
  Speaker 3: to kind of make sure AI advances safely. And a
• 8:07
  Speaker 3: lot of the really important factors here, of course, the
• 8:10
  Speaker 3: models themselves and how they're tested, are really critical. But
• 8:14
  Speaker 3: I also think a key thing to think about here
• 8:17
  Speaker 3: when we're thinking about the advancement of AI is not
• 8:20
  Speaker 3: just the technology itself and what it's capable of, but
• 8:23
  Speaker 3: the decisions made by these tech companies and these tech
• 8:26
  Speaker 3: leaders when it comes to how do they secure their models?
• 8:29
  Speaker 3: How do they test them? How do they train them?
• 8:32
  Speaker 3: What kind of data are they using to train them?
• 8:34
  Speaker 3: All of these decisions that these companies make really do
• 8:38
  Speaker 3: kind of play into this trajectory of AI and how
• 8:43
  Speaker 3: these broader concerns are handled. And I think that's what
• 8:46
  Speaker 3: we're seeing now is that these companies are in a
• 8:48
  Speaker 3: phase where they're trying to figure out what that looks
• 8:50
  Speaker 3: like moving forward.
• 8:52
  Speaker 4: All right, Lisa, thank you so much.