Latest / Views and News with Clarence Ford / Can we stop AI from deceiving us?
Transcript
- 0:00Speaker 1: I would like you to join the conversation, especially the
- 0:03Speaker 1: next one. We're keeping a BDI on developments on the
- 0:06Speaker 1: AI front. So we are going to talk about AI deception.
- 0:10Speaker 1: There's an article in The Guardian, also an article today
- 0:14Speaker 1: in today's New York Times. In fact, recent reporting and
- 0:18Speaker 1: academic analysis, including prominent coverage highlighted in those two titles.
- 0:26Speaker 1: They underscore, I think, what is a really critical shift
- 0:30Speaker 1: in artificial intelligence development. We are talking now about AI
- 0:35Speaker 1: models growing more capable and complex. And we thought we
- 0:40Speaker 1: knew that was going to happen. Their capacity may be
- 0:44Speaker 1: for subtle persuasive deception. We've seen months ago that was reported,
- 0:50Speaker 1: but it increases rather than decreases. I think one of
- 0:53Speaker 1: the risks the question that we've got to ask ourself.
- 0:58Speaker 1: We perceive an ideal world served by AI digitally and,
- 1:06Speaker 1: I think, responsibly. And is that where we are heading?
- 1:10Speaker 1: I think the big, big question, certainly underscored by today's
- 1:13Speaker 1: New York Times article. So we're going to go to
- 1:15Speaker 1: Wade Seal to bring us up to speed in a
- 1:18Speaker 1: short while. But before we go there, let's check in
- 1:22Speaker 1: on that WhatsApp line, 0721.
- 1:24Speaker 2: 567-1567.
- 1:24Speaker 1: Janneke has sent us a voice note. Let's take a listen.
- 1:30Speaker 3: Clarence, that's such a good voice note from that listener
- 1:32Speaker 3: to Errol.
- 1:33Speaker 2: I agree.
- 1:34Speaker 3: And on that note, we just saw John Mason on
- 1:37Speaker 3: Friday night in a play called Superfan, which I can
- 1:40Speaker 3: highly recommend. And obviously, it just explains the New Zealand
- 1:44Speaker 3: support of South Africans. It's a very moving play. I
- 1:48Speaker 3: can highly recommend it. Unfortunately, it sold out, but I
- 1:50Speaker 3: think we need to ask John to do some more
- 1:51Speaker 3: runs of it. And maybe Errol can go see it
- 1:54Speaker 3: as well. What a great voice note. Errol, get a
- 1:57Speaker 3: Springbok jersey.
- 1:59Speaker 1: Yeah, we actually discovered quite a bit. In fact, I
- 2:02Speaker 1: did today, Errol. But again, we have suggested that there
- 2:07Speaker 1: is a whole lot more than just legendary rugby rivalry
- 2:14Speaker 1: at stake when the Springboks meet the All Blacks on
- 2:17Speaker 1: a rugby pitch. We watched that very fierce contestation for
- 2:21Speaker 1: the ball translate after the game into really kind of
- 2:25Speaker 1: warm relationships. You get that. You see the relationship. with
- 2:29Speaker 1: opposing players and families and kids, warm hugs, first name
- 2:33Speaker 1: kind of terms, Uncle Sia. And we proposed maybe that
- 2:39Speaker 1: there was, again, historically grounded and a culturally uh a
- 2:44Speaker 1: cultural resonance of sorts between the springboks and the all
- 2:48Speaker 1: black specifically and we wondered maybe we extrapolating too much
- 2:53Speaker 1: we wondered if that same on and off pitch relationships
- 2:58Speaker 1: are in place for uh specifically the english um because
- 3:03Speaker 1: I know the Irish and the Scottish especially would maybe
- 3:06Speaker 1: share similar kind of histories with South Africa and certainly
- 3:11Speaker 1: with New Zealand as well. We speculate it may be
- 3:14Speaker 1: extrapolated too much, maybe too far down the rabbit hole,
- 3:18Speaker 1: but why not?
- 3:19Speaker 2: Why not?
- 3:20Speaker 1: It's 1039. We welcome Wade Seal. Wade Seal, it's good
- 3:23Speaker 1: to have you with us again.
- 3:25Speaker 2: Thanks, Clarence. Good to be with you again this morning.
- 3:28Speaker 1: You've now fast become our go-to guy when it comes
- 3:31Speaker 1: to AI.
- 3:33Speaker 2: Thank you. That's an honor. I don't carry that title lightly.
- 3:37Speaker 1: Yeah. So I guess we're going to be navigating AI
- 3:40Speaker 1: deception then. I don't know if you had sight of
- 3:45Speaker 1: today's New York Times article.
- 3:48Speaker 2: Yes, I did. I've been looking at the article. It
- 3:51Speaker 2: follows some reporting from The Guardian as well last week.
- 3:57Speaker 2: It's interesting, but again, I think the language is important
- 4:02Speaker 2: to say that To say that what has been happening
- 4:05Speaker 2: is a deception implies, again, that there's this intention and
- 4:12Speaker 2: that there's this thinking sort of consciousness within AI that's
- 4:17Speaker 2: deciding to deceive or not to deceive, you know. And
- 4:20Speaker 2: that misplaces where the culpability should be going because, again,
- 4:24Speaker 2: I maintain it's a matter of how these AIs are programmed,
- 4:28Speaker 2: you know. So, yeah, that's it, yeah. But at very
- 4:31Speaker 2: interesting times.
- 4:33Speaker 1: So, I mean, let us understand a sandbox. Isn't a
- 4:36Speaker 1: sandbox supposed to be exactly a sandbox? A box that
- 4:40Speaker 1: the AI cannot escape from and it's got to use
- 4:43Speaker 1: its own kind of resources to find solutions to its task.
- 4:50Speaker 1: Why is it escaping them or is there a fault
- 4:54Speaker 1: designed in the experiment?
- 4:56Speaker 2: That's an excellent question. It's a very appropriate question. Any
- 5:01Speaker 2: Any developer uses, tests new features for any app, beyond
- 5:06Speaker 2: AI even, for any app using sandboxes. So think, for example,
- 5:10Speaker 2: of Facebook, right? We all use Facebook. Now imagine they
- 5:14Speaker 2: want to introduce a new function to Facebook. They don't
- 5:18Speaker 2: shut Facebook down for a few weeks in order to
- 5:21Speaker 2: test things. the new function, what they do is they
- 5:24Speaker 2: create a sandbox within Facebook, within the existing app. And
- 5:28Speaker 2: that's a confined area for the new feature to be tested.
- 5:34Speaker 2: So certain identified users are then allowed to use the
- 5:38Speaker 2: new feature. And you're absolutely right. It's like a playground.
- 5:42Speaker 2: The sandbox is supposed to keep the sand in the sandbox.
- 5:47Speaker 2: in a way so that the sand doesn't flow out
- 5:50Speaker 2: into the rest of the playground. And so it's important
- 5:54Speaker 2: that the sandbox is secure and keeps all the sand in.
- 5:58Speaker 2: And what you find with the AIs, in my view,
- 6:01Speaker 2: with the testing that's going on at the moment, is
- 6:04Speaker 2: that the sandboxes are not strong enough. So the sand
- 6:08Speaker 2: is leaking. There's holes in the sandboxes and the sand
- 6:11Speaker 2: is leaking. That's what happens when AIs sort of escape
- 6:15Speaker 2: the testing environments. The testing environments were not rigid enough.
- 6:19Speaker 2: They were not strong enough to keep the AIs constrained.
- 6:25Speaker 2: But you see again now when it is reported that
- 6:29Speaker 2: the AI deceived the developers or the AI escaped, then
- 6:35Speaker 2: it's as if the AI is sitting there and trying
- 6:38Speaker 2: to figure out how to escape the sandbox, you know,
- 6:41Speaker 2: and then the developers sort of shun the responsibility. what's
- 6:45Speaker 2: really going on is that the developers are not building
- 6:49Speaker 2: strong enough sandboxes, and so the developers should be held
- 6:53Speaker 2: responsible for whenever an AI escapes and wrecks Havoc. And
- 6:58Speaker 2: the dangers are clearly there. I think everybody's aware that
- 7:02Speaker 2: it's super important that these AIs and their features are
- 7:06Speaker 2: tested in strong sandboxes. So you're absolutely right to ask
- 7:10Speaker 2: that question about the sandboxes.
- 7:11Speaker 1: Another stupid question. Can't you ask AI to design a
- 7:19Speaker 1: sandbox that it can't escape from?
- 7:23Speaker 2: Well, that's a very interesting question because what is happening
- 7:28Speaker 2: at the moment, as you know, there's a lot of
- 7:31Speaker 2: AIs who are being used to develop software. And so
- 7:36Speaker 2: the AIs are being used to develop things like sandboxes.
- 7:42Speaker 2: I suppose the problem is that they want to. They're
- 7:45Speaker 2: trying to develop an AI that is progressively smarter, progressively
- 7:52Speaker 2: more capable. And so the challenge comes in if it's
- 7:55Speaker 2: going to be progressively more capable, then it shouldn't. In theory,
- 8:00Speaker 2: it should be able to not be able to escape
- 8:06Speaker 2: sandboxes that are developed independently. So if, for example, the
- 8:12Speaker 2: function is to, they want to test the function on
- 8:15Speaker 2: Facebook again, maybe a new feature in Messenger or on Marketplace,
- 8:20Speaker 2: they want to test something like that, then you can
- 8:22Speaker 2: ask an AI to develop a very strong sandbox because
- 8:27Speaker 2: the AI itself is independent from the feature that's being tested.
- 8:33Speaker 2: It becomes a little complicated when the AI itself is
- 8:36Speaker 2: the thing that you're wanting to test. I don't know
- 8:40Speaker 2: if that makes sense. I get it.
- 8:42Speaker 1: You're saying it's got an innate ability then to deceive.
- 8:47Speaker 2: Not to deceive, again, not to deceive, because it doesn't
- 8:50Speaker 2: have a mind of its own, but it's got this innate,
- 8:54Speaker 2: a certain level of security. So the idea is that
- 8:57Speaker 2: if you want to, you know, let's imagine that the
- 9:01Speaker 2: AI is at a clever level, sort of at grade five,
- 9:07Speaker 2: the equivalent of a grade five kid. Obviously, it's much smarter,
- 9:10Speaker 2: but let's just imagine that. and then all of a
- 9:14Speaker 2: sudden you come along, they develop something new now, a
- 9:17Speaker 2: grade 6 or a grade 7 level AI. In order
- 9:21Speaker 2: to test whether the AI is at grade 6 or
- 9:24Speaker 2: grade 7 level, the sandbox will have to be developed
- 9:29Speaker 2: independently of the AI itself. Do you understand? It's got
- 9:34Speaker 2: to be developed independently. If the AI develops independently, the sandbox,
- 9:39Speaker 2: then it may just develop it at a level that
- 9:43Speaker 2: suits itself. So that's the complication. But you're right. AI
- 9:49Speaker 2: is test. Developers use AI now routinely to write code.
- 9:56Speaker 2: And so I'm sure that they routinely use it to
- 9:59Speaker 2: develop sandboxes for other features.
- 10:02Speaker 1: Okay, so I think what we are understanding more so
- 10:05Speaker 1: today from me, from this New York Times article, is
- 10:09Speaker 1: that AI is capable of Strategic deception and maybe also
- 10:15Speaker 1: hidden reasoning. And I think, and you must help me
- 10:19Speaker 1: with this, I think the ideal world is one that
- 10:22Speaker 1: is served diligently by AI and responsibly by AI. Is
- 10:30Speaker 1: that the mission here? Is that the mission that we're on?
- 10:33Speaker 1: Is that the mission statement with AI? Is it about
- 10:36Speaker 1: diligence and responsibilities?
- 10:39Speaker 2: It ultimately should be. I think, you know, something very
- 10:43Speaker 2: interesting a few months ago, I don't know if you
- 10:45Speaker 2: caught that, but there was a sort of a voice
- 10:48Speaker 2: out of left field and that was the Pope, the
- 10:50Speaker 2: Catholic Pope of the Catholic Church, who released a big
- 10:54Speaker 2: document on AI and again sort of telling the world
- 10:57Speaker 2: to pivot towards serving humanity, serving conscious development of AI,
- 11:05Speaker 2: serving it for the greater good and so on. And
- 11:08Speaker 2: a big problem, as you probably have been gesturing in
- 11:11Speaker 2: our various interactions over the past few months, is that
- 11:15Speaker 2: the AI is currently being developed by people with private interests,
- 11:20Speaker 2: with for-profit interests.
- 11:22Speaker 3: Yeah.
- 11:23Speaker 2: And so it's to be expected that when they themselves
- 11:29Speaker 2: are responsible for testing these models, if the models don't
- 11:33Speaker 2: give the results they're looking for, what are they going
- 11:35Speaker 2: to do? Are they going to report it? Are they
- 11:37Speaker 2: going to stop the big marketing campaign that went with it?
- 11:40Speaker 2: Or are they going to report to their shareholders that
- 11:43Speaker 2: all the money that was invested in the development of
- 11:46Speaker 2: this new AI has been wasted? They're not going to
- 11:48Speaker 2: do that. They're going to look the other way. you understand,
- 11:51Speaker 2: or they're going to try and find a way to
- 11:53Speaker 2: justify it. And so, yes, the New York Times article
- 11:57Speaker 2: is interesting. It's super important because, again, it calls us
- 12:02Speaker 2: to pivot towards the real reason, the common good around
- 12:08Speaker 2: underlying the development of AI. But I maintain, Clarence, that
- 12:13Speaker 2: an important part of that is understanding where things go wrong.
- 12:18Speaker 2: And things don't go wrong because our AIs are conscious.
- 12:22Speaker 2: Things are going wrong because developers are not being held
- 12:26Speaker 2: to account. I think that is the big issue. They
- 12:30Speaker 2: should be held. And again, it's like anybody, you know,
- 12:34Speaker 2: if you're going to be, it's like you, Clarence, you
- 12:36Speaker 2: and your job, you've obviously got to work within certain guardrails.
- 12:40Speaker 2: You can't say things that it's going to bring Cape
- 12:42Speaker 2: Talk as a brand into disrepute through some defamation suit
- 12:47Speaker 2: or whatever. you've got to work within guardrails. And then
- 12:51Speaker 2: you take the care, when those guardrails are clear to you,
- 12:54Speaker 2: you then take the care to ensure that these things
- 12:58Speaker 2: that's going to bring Cape Talk into disrepute don't happen.
- 13:02Speaker 2: And the same thing should happen with developers. But it's
- 13:05Speaker 2: not happening at the moment. They're driven by profit. There's
- 13:07Speaker 2: this race to the bottom. And they're not held accountable
- 13:11Speaker 2: because there's this narrative that that these AIs can think
- 13:15Speaker 2: for themselves and deceive people, and it's just not the case.
- 13:19Speaker 2: It's just not the case.
- 13:21Speaker 1: Okay, so let us then understand how we are going
- 13:27Speaker 1: to build systems. How do we architect systems where honesty
- 13:31Speaker 1: and transparency are fundamental boundaries, constraint boundaries, rather than optimizations?
- 13:42Speaker 2: Yeah. I mean, that's a big question that you're asking.
- 13:45Speaker 2: It's an important question. But I think from my view,
- 13:48Speaker 2: as a start, and again, I think everybody who's involved
- 13:52Speaker 2: in any kind of organization would understand this, any sector.
- 13:56Speaker 2: I mean, you are in media. The media watchdogs are
- 14:00Speaker 2: not the media houses. Cape Talk doesn't, so the ombudsman
- 14:06Speaker 2: of the press ombudsman is not in Cape Talk itself.
- 14:10Speaker 2: The press ombudsman is an independent organization. And so when
- 14:14Speaker 2: we talk about transparency and honesty, that's where it's got
- 14:18Speaker 2: to start. At the moment, like I say, you have
- 14:22Speaker 2: Anthropic doing testing and researching and it's guardrails for itself.
- 14:28Speaker 2: And you have open AI doing testing and guardrails for itself.
- 14:34Speaker 2: That's where it's in such a fundamental level. We've got
- 14:39Speaker 2: to have like an AI ombudsman that is independent and
- 14:44Speaker 2: that does research and testing. And I mean, think about it.
- 14:47Speaker 2: It happens before a pharmaceutical company can pass a drug test.
- 14:53Speaker 2: a new drug. It's got to go through months and
- 14:55Speaker 2: months of testing, years of testing. It's got to go
- 14:58Speaker 2: through all sorts of loopholes, legal, not loopholes, but it's
- 15:03Speaker 2: got to meet certain standards, et cetera. And that just
- 15:07Speaker 2: doesn't exist in the AI world at the moment. What
- 15:11Speaker 2: you have is people, researchers, very smart people, but also
- 15:15Speaker 2: people who have access to big money and they are
- 15:20Speaker 2: able to just do these, develop these agents or these
- 15:26Speaker 2: sort of, yeah, these AIs, and they let them out
- 15:31Speaker 2: into the market and then something goes wrong. So I
- 15:33Speaker 2: would say as a sort of start and a giant
- 15:37Speaker 2: leap towards transparency, trustworthiness, openness, honesty, a giant leap would
- 15:45Speaker 2: be independent marketing watchdogs, you know?
- 15:48Speaker 1: Yeah.
- 15:49Speaker 2: And, and then there's a, sorry, Clarence, I don't know.
- 15:51Speaker 2: I hope to not cut you, but there's another, there's another,
- 15:54Speaker 2: there is this emerging trend or so of having these
- 15:57Speaker 2: independent labs, they call them, where researchers come. There's a,
- 16:03Speaker 2: there's a guy called Marius Orban, who's the founder of
- 16:06Speaker 2: Apollo Research, and he's actually a German, but he's started
- 16:10Speaker 2: a company in San Francisco recently. And basically what they
- 16:14Speaker 2: do is that they do research on, like an SABS
- 16:19Speaker 2: you know the SABS stamp of approval that we have
- 16:21Speaker 2: for products in South Africa so this is like an
- 16:24Speaker 2: SABS but for AIs but the problem is also that
- 16:28Speaker 2: this guy and it's a trend that's happened it's the
- 16:31Speaker 2: same as happened with OpenAI and with Anthropic is that
- 16:36Speaker 2: it's a for-profit company And so the problem with that
- 16:40Speaker 2: is immediate. It means that if Grok is going to
- 16:44Speaker 2: use Wade's research lab to see whether their new function
- 16:49Speaker 2: is ethically compliant, Wade, who's got a private interest, a
- 16:54Speaker 2: profit interest, is going to play nice with Grok. He's
- 16:58Speaker 2: not necessarily going to be 100% honest and open and
- 17:02Speaker 2: transparent and be there for the purpose that he's supposed
- 17:06Speaker 2: to be there for. So I would say that we
- 17:11Speaker 2: sort of have ombudsman-like organizations, quality assurance organizations, more than
- 17:18Speaker 2: to test these AIs. But secondly, that those are not
- 17:21Speaker 2: privately owned for-profit institutions. They must be– do you know
- 17:26Speaker 2: what I mean? They've got to be publicly owned, and
- 17:28Speaker 2: they've got to be there for the common good.
- 17:30Speaker 1: Well, I don't mind triple bottom lines. That can be
- 17:33Speaker 1: put into the program, but it must.
- 17:35Speaker 2: Be triple bottom lines.
- 17:38Speaker 1: I think you are leading us into a very important conversation,
- 17:42Speaker 1: and it's a conversation that we knew was coming. It's
- 17:45Speaker 1: a philosophical, it's an ethical conversation. So essentially, and this
- 17:50Speaker 1: is the headline in The Guardian, if you build something
- 17:52Speaker 1: vastly smarter than you, it better be on your side.
- 17:58Speaker 1: But if it's vastly smarter than you, what is not
- 18:02Speaker 1: on your side may be good for you. Do we
- 18:06Speaker 1: trust it then to do what is good for us
- 18:10Speaker 1: even if we have suspicions about it?
- 18:13Speaker 2: Very good question. Very good question. I mean, and you
- 18:16Speaker 2: think about it like we think about the fact that
- 18:19Speaker 2: you as a salary, every single listener that you have
- 18:22Speaker 2: there is a salary earner. You own a bank account.
- 18:25Speaker 2: But we've got these AIs that are unleashed onto our
- 18:28Speaker 2: banking accounts, into our systems, into our private lives. And
- 18:33Speaker 2: how much say do we have on these things that
- 18:36Speaker 2: are supposedly much smarter than we are? We don't understand it.
- 18:41Speaker 2: And there's just nobody batting for the small man. So
- 18:45Speaker 2: you raise an absolutely important point. There are serious, I
- 18:50Speaker 2: don't know if you want to call it philosophical and
- 18:52Speaker 2: ethical issues, but that's what they are. There are serious
- 18:55Speaker 2: ethical and philosophical questions here. And I think I've been
- 18:59Speaker 2: saying it for months now. You know, you look at
- 19:01Speaker 2: South Africa and our landscape. We, I mean, besides you
- 19:06Speaker 2: and the article here and there, you've consistently raised this
- 19:09Speaker 2: issue on your platform. But by and large, I mean,
- 19:13Speaker 2: the government is just nowhere. You know, there's just nothing
- 19:16Speaker 2: being said about it. And it's not as if we're
- 19:18Speaker 2: protected in South Africa because we're in South Africa. We
- 19:21Speaker 2: live in a globalized society. and um we just we
- 19:26Speaker 2: just know where we just not even having the conversation
- 19:29Speaker 2: you know so um it's a it's an incredibly important
- 19:33Speaker 2: time for us and we have to have these conversations
- 19:36Speaker 2: so yeah um thank you for having it i.
- 19:39Speaker 1: Mean no it's important platform repeatedly and we need people
- 19:42Speaker 1: like you to to help us navigate uh these new
- 19:45Speaker 1: grounds and discoveries so maybe i should just introduce you
- 19:49Speaker 1: probably wade's holds a Ph.D. in philosophy and is Africa
- 19:54Speaker 1: AI Ethics Carnegie Corporation New York postdoctoral fellow at the
- 19:59Speaker 1: Institute for Humanities. We are very good as human beings.
- 20:04Speaker 1: to self-inflict harm. So yeah, it's such a difficult kind
- 20:12Speaker 1: of space. Maybe it should be a broad brief, but
- 20:18Speaker 1: I guess in an ideal ecosystem, AI should be augmenting
- 20:24Speaker 1: human autonomy. It should assist us with critical thinking, preserve
- 20:28Speaker 1: that capability at least. Just some questions that are left field,
- 20:32Speaker 1: I guess, a little doctor.
- 20:34Speaker 2: Yeah, no, thank you. There was an interesting article, I
- 20:39Speaker 2: just forget off the top of my head now, but
- 20:41Speaker 2: it was from one of the deputy or one of
- 20:45Speaker 2: the deans of the educational faculty at Stellenbosch University. And
- 20:50Speaker 2: he was asking this question about teachers. Now, again, that's
- 20:54Speaker 2: a relatable example that we can think about, but because
- 20:59Speaker 2: we all know teachers and how important teachers are. But
- 21:03Speaker 2: it teaches and teaching within the context of AI. And
- 21:08Speaker 2: the broad sort of thesis of his article, of his op-ed,
- 21:12Speaker 2: was that just because you impart knowledge doesn't mean or
- 21:17Speaker 2: doesn't make you a good teacher. Because I think we
- 21:20Speaker 2: all know of good teachers when we think about good teachers.
- 21:24Speaker 2: It wasn't just that they taught us mathematics or whatever.
- 21:28Speaker 2: It was more of the human mentoring role. And so
- 21:32Speaker 2: this professor asks that question, and then he argues that
- 21:37Speaker 2: you can have AIs coming in and performing the role
- 21:41Speaker 2: of imparting knowledge, but the question is, does that make
- 21:46Speaker 2: them teachers? And so we can bring that back to
- 21:50Speaker 2: exactly what you're asking. In that op-ed, he doesn't go
- 21:55Speaker 2: so far, but for me, the natural next question is,
- 21:59Speaker 2: You know, if you use something like AI to augment
- 22:02Speaker 2: the role of the teacher, what does it then mean
- 22:06Speaker 2: to be a teacher? And we know there's tremendous opportunity
- 22:10Speaker 2: in that space because we know our teachers in our
- 22:12Speaker 2: public schools are overburdened with admin and marking and all
- 22:18Speaker 2: sorts of things. And there should be. I'm not an
- 22:21Speaker 2: expert in this field, in education, but there should be
- 22:24Speaker 2: low-hanging fruit there that AI can come and do that
- 22:29Speaker 2: sort of admin work.
- 22:30Speaker 1: Maybe even heavy lifting. Maybe even heavy lifting.
- 22:33Speaker 2: Well, maybe, absolutely. And then the teachers can focus on
- 22:37Speaker 2: developing these little human beings, which is exactly what is
- 22:41Speaker 2: lacking in our public school.
- 22:46Speaker 1: We are out of time, unfortunately, Wade, but we will
- 22:50Speaker 1: have more such conversations as we are alerted to the
- 22:54Speaker 1: challenges ahead in the use and adoption of AI technology.