Latest / Views and News with Clarence Ford / Open AI going rogue
Transcript
- 0:00Speaker 1: Answer our questions. Welcome to you, doctor Seal, Welcome Wade.
- 0:05Speaker 1: It's great to have you.
- 0:07Speaker 2: Thanks Claren, it's good to be you know, I was
- 0:10Speaker 2: wondering when I was listening to you, Now, when am
- 0:12Speaker 2: I going to breach he scant status?
- 0:14Speaker 1: Hascant status?
- 0:16Speaker 2: Yeah, it's I've been here for how many times? I
- 0:18Speaker 2: think this is not the first time I've been here?
- 0:20Speaker 2: So when when do I become part of the family.
- 0:24Speaker 1: Well, you, we need to talk to you regularly. We
- 0:28Speaker 1: need to talk to you regularly, So consider yourself a Hayscent.
- 0:32Speaker 1: Consider yourself. I think we're gonna lead your expertise going forward,
- 0:36Speaker 1: and I think you're absolutely gonna be a hescant. But
- 0:40Speaker 1: but just help us. Yeah, because when we're here about
- 0:42Speaker 1: AI going rogue, we obviously think science fiction, but technically speaking,
- 0:49Speaker 1: we need to know what actually happened here. AI kind
- 0:53Speaker 1: of escaping its sand box. Did the system disobey or
- 0:59Speaker 1: did it pay its goal? Maybe too? Well, help us
- 1:03Speaker 1: with what it did here.
- 1:05Speaker 2: Thanks Clarence. Yeah, I think it's important to understand you're
- 1:09Speaker 2: asking eating the nail on the head. If you notice
- 1:13Speaker 2: the way in which this issue is being reported, it's
- 1:16Speaker 2: basically like AI has maliciously developed this intention to escape
- 1:23Speaker 2: and then to cause havoc, and that's not what happened.
- 1:26Speaker 2: AI is the most sophisticated AI in the world. Is
- 1:31Speaker 2: closer to a calculator than it is to a human
- 1:35Speaker 2: being that can develop intentions of its own. And I'm
- 1:39Speaker 2: talking about a four year old human being like my
- 1:41Speaker 2: daughter can say, Daddy, can you take me to McDonald's
- 1:45Speaker 2: for supper please? So, an AI is a very specific thing.
- 1:49Speaker 2: It remains closer to a rudimentary calculator than to a
- 1:55Speaker 2: human being than to a person. And the reason I
- 1:58Speaker 2: say that is because you still have to in the
- 2:02Speaker 2: same way that you have to give a calculator instructions
- 2:06Speaker 2: to perform a computation, so too you have to give
- 2:11Speaker 2: an AI an instruction, and the AI operates within certain
- 2:17Speaker 2: boundaries and performs that instruction. And the clear that coming
- 2:22Speaker 2: to the specific issue of what is in the media
- 2:25Speaker 2: at the moment, the clear sort of clue to what
- 2:28Speaker 2: I'm saying is the word sandbox. Now, for your listeners,
- 2:32Speaker 2: if you picture a playground, the playground would be like
- 2:35Speaker 2: an app, and then developers want to test a specific feature.
- 2:41Speaker 2: In this case, it was a new frontier model, and
- 2:45Speaker 2: they want to test the capabilities a specific new feature
- 2:48Speaker 2: of an app. So they boil a sandbox in the playground.
- 2:53Speaker 2: And in the same way that the sandbox is supposed
- 2:56Speaker 2: to contain the sand in the playground so that it
- 2:59Speaker 2: doesn't fall on the AstroTurf, doesn't fall on the paving,
- 3:04Speaker 2: so to the sandbox in a text perspective, so to
- 3:09Speaker 2: the sandbox is supposed to contain the new AI or
- 3:13Speaker 2: the new feature that's being tested. And what happened here
- 3:17Speaker 2: is that the sandbox wasn't strong enough. The architecture of
- 3:21Speaker 2: the AI, of the of the testing environment was not
- 3:25Speaker 2: strong enough to contain the new feature that was being tested.
- 3:31Speaker 2: And so instead of asking did this AI wake up
- 3:35Speaker 2: and think, wait, I need to preserve myself, I need
- 3:38Speaker 2: to cause havoc, I need to save myself, we should
- 3:41Speaker 2: be asking why was the architecture of the sandbox not
- 3:47Speaker 2: strong enough? And the reason why we do they don't
- 3:50Speaker 2: tend to ask that question is because very powerful people
- 3:53Speaker 2: stand to be brought before jury and they stand to
- 3:59Speaker 2: the risk of are some serious questions, because can you
- 4:03Speaker 2: imagine if this if an AI is let loose. They
- 4:06Speaker 2: developed the AI and the architecture that its supposed that's
- 4:09Speaker 2: supposed to contain it is not strong enough, and the
- 4:12Speaker 2: AI escapes, and then it causes avocs without banking accounts
- 4:17Speaker 2: or with our electricity grid, or with our you know,
- 4:20Speaker 2: our custom of our ads whatever, and so yes, we
- 4:25Speaker 2: must be concerned, but we must ask the right question.
- 4:28Speaker 2: Is the AI waking up and deciding to like a Terminator,
- 4:33Speaker 2: you know the old Terminator movie, is deciding to come
- 4:36Speaker 2: and do his own thing and cause avoc. No, that's
- 4:38Speaker 2: not what is happening. What is happening is AIS are
- 4:41Speaker 2: being tested and they're being released. They should be released
- 4:45Speaker 2: in terms of architectures that secure the AIS, that secure
- 4:49Speaker 2: the technology, and that does not happen. And so we
- 4:52Speaker 2: must ask those questions.
- 4:53Speaker 1: I get you, I absolutely get you, and I want
- 4:57Speaker 1: to drive to that particular position. But for now, the
- 5:00Speaker 1: journalists that wrote that article talks about this is AI's
- 5:03Speaker 1: first felony. I absolutely like that. The journalist is talking
- 5:09Speaker 1: about not wanting to do any scare mongering. I like
- 5:14Speaker 1: that too. And I think the principle here that the
- 5:16Speaker 1: journalist eventually brings forth and we're talking about New York Times,
- 5:21Speaker 1: is that speed is often at the expense of safety.
- 5:26Speaker 1: And I think that is your closing comment. And obviously
- 5:31Speaker 1: we need to be independently vetting these systems that are
- 5:35Speaker 1: being released, shouldn't we.
- 5:37Speaker 2: That's absolutely and so that's what the calls from legislators
- 5:40Speaker 2: in the US are basically about. We've got to be
- 5:46Speaker 2: vetting an have third party oversight over exactly the kind
- 5:50Speaker 2: of guard rails and architecture security for these machines. The
- 5:55Speaker 2: Trump administration issued an executive order where they basically said
- 6:00Speaker 2: they they wanting the companies to sort of voluntarily participate
- 6:07Speaker 2: in this, and I think lawmakers in the US are saying, no, no,
- 6:11Speaker 2: that's not enough. They've got to we've got to step
- 6:14Speaker 2: in and we've got to insure an old people accountable. Now,
- 6:17Speaker 2: if you look at the felony the felony comment from
- 6:20Speaker 2: the journalist, it's an interesting comment. I don't think like
- 6:24Speaker 2: I say. You know, you can't hold a car guilty
- 6:29Speaker 2: of a crime or of a civil dealict if that
- 6:33Speaker 2: car crashes into an into your car, you hold the
- 6:37Speaker 2: driver accountable and in the same way. And and that's
- 6:41Speaker 2: why this is so interesting. We are at a pivotal
- 6:44Speaker 2: moment because if you're not going to hold the car accountable.
- 6:47Speaker 2: You must hold the driver. Now, if the AI is
- 6:49Speaker 2: the car that can't act out of its own who
- 6:53Speaker 2: is the driver? And the driver is your Elon Musk,
- 6:55Speaker 2: your Sam Altmont, your Dario Ama Day, all these big guys.
- 7:00Speaker 2: And that's just Americans. We haven't even gotten to the
- 7:02Speaker 2: rest of the world. You've got all those guys accountable.
- 7:05Speaker 2: And that is I time now and sorry for speaking
- 7:09Speaker 2: so much, but again we must bring attention to what's happening.
- 7:12Speaker 2: Where's our AI regulations and discussions in South Africa? No way,
- 7:17Speaker 2: that's absolutely no. That is a concern and that's why
- 7:20Speaker 2: you're bringing this topic up here on your platform is
- 7:24Speaker 2: so so important.
- 7:25Speaker 1: So what you're saying essentially is if the developers or
- 7:29Speaker 1: the evaluators or the company running the server is fined
- 7:34Speaker 1: and significantly fine, then you won't find these kind of
- 7:37Speaker 1: escapes from the sandbox.
- 7:39Speaker 2: Absolutely, I do think so. I mean that works the
- 7:42Speaker 2: same with everything else. You know, if you all them
- 7:45Speaker 2: responsible for what they do, then they're going to be
- 7:47Speaker 2: far more cautious in terms of designing these architectures the box,
- 7:53Speaker 2: you know, the walls of the sandbox so that nothing
- 7:56Speaker 2: gets out. What is happening at the moment is they
- 7:59Speaker 2: have free rain and there's very little limitation on what
- 8:02Speaker 2: they can and can't do, and so they just sort
- 8:05Speaker 2: of allowed to try anything and test anything, you know,
- 8:08Speaker 2: and so they would take more responsibility if there's more
- 8:12Speaker 2: responsibly accountability to be ad.
- 8:17Speaker 1: Yeah, you are essentially saying that top tech companies can
- 8:21Speaker 1: contain these autonomous things.
- 8:26Speaker 2: No, absolutely, I have absolutely no doubt about it. Now,
- 8:29Speaker 2: I must also just say to you and your listeners,
- 8:32Speaker 2: you know, as somebody who works in a university environment,
- 8:37Speaker 2: and you see it's got amazing technical AI research programs happening.
- 8:43Speaker 2: But I'm still the most thing. The one thing I'm
- 8:46Speaker 2: convinced about here is that nobody's actually convinced or write
- 8:51Speaker 2: about anything in terms of AI. And I add a
- 8:54Speaker 2: debate to the with the world renowned I won't say
- 8:56Speaker 2: it is, but a world renown computer signed to is
- 9:00Speaker 2: an expert in AI, because he said at the conference
- 9:03Speaker 2: that an AI can reason, an AI can't reason, you know,
- 9:07Speaker 2: and I and eventually he had to concede the point.
- 9:11Speaker 2: But I point to that because.
- 9:13Speaker 1: Yet or will never be able to reason.
- 9:17Speaker 2: Certainly A can't reason. Yet a can't reason Yet the
- 9:21Speaker 2: thing is AI is just work on the basis of
- 9:23Speaker 2: statistics of putting words together. So that means if you
- 9:27Speaker 2: think an AI can respond in a sensible way to
- 9:32Speaker 2: your input, then yes, AIS can reason. But calculators reason
- 9:37Speaker 2: in that way too. When you put in an input
- 9:39Speaker 2: one plus one, they give you a reasonable answer to
- 9:43Speaker 2: So in that sense, AIS can reason. But if you
- 9:45Speaker 2: think again of an AI thinking, oh, I'm going to
- 9:49Speaker 2: escape here, what can I do today? I think I'm
- 9:52Speaker 2: going to cause havoc for a for for people and
- 9:58Speaker 2: escape my environment and going to the bank counts and
- 10:01Speaker 2: steal their bitcoint or start killing people. No AI is
- 10:05Speaker 2: gone to that. They can't reason autonomously on their own.
- 10:08Speaker 2: They still rely on inputs and outputs and so and so.
- 10:12Speaker 2: There is the but I raised that point to make.
- 10:15Speaker 2: I raised that to say even the experts don't know
- 10:21Speaker 2: what the capabilities are.
- 10:23Speaker 1: And you an expert, doctor white Seal, are you saying
- 10:25Speaker 1: you don't know.
- 10:28Speaker 2: In terms of what AI is and what the what
- 10:31Speaker 2: the ultimate capabilities are? And this is all going, yes, No,
- 10:35Speaker 2: I don't know.
- 10:37Speaker 1: Because a s I. So I think we need to
- 10:39Speaker 1: to maybe just you know, ponder the possibilities if it
- 10:42Speaker 1: is going to by any way means necessary do the
- 10:47Speaker 1: job at hand. Uh, you know, what are the possible
- 10:51Speaker 1: outcomes in the future with with a s I for example.
- 10:56Speaker 1: But let's let's talk. And I'm talking to doctor White Seal,
- 10:58Speaker 1: a post doctor or Fellow in AI Ethics at the
- 11:01Speaker 1: Institute for Humanities in Africa at the University of Kaetan,
- 11:05Speaker 1: and he focuses on AI ethics existential risk governance frameworks
- 11:09Speaker 1: from a global sales perspective. Bruce Rights via WhatsApp. Keep
- 11:14Speaker 1: your thoughts coming through on seven two five six seven
- 11:16Speaker 1: one five six seven Bruce Wrights. I'm always amazed at
- 11:21Speaker 1: the hype that is created through marketing via journalism. AI
- 11:25Speaker 1: does not escape as much as an information leak to
- 11:28Speaker 1: the press is accidental. If you read between the lines
- 11:32Speaker 1: of winning the Race of AI against the East, these
- 11:35Speaker 1: sorts of tricks are designed to create hype in the
- 11:39Speaker 1: market but also claim in the race of winning the
- 11:42Speaker 1: AI Prize. Bruce suggesting this is a little marketing. Uh
- 11:48Speaker 1: what they call it? They call it ambush marketing, telling
- 11:53Speaker 1: the world we we got we got, we a little
- 11:56Speaker 1: ahead of you China.
- 11:57Speaker 2: Is is it that hi? I don't think there's this
- 12:01Speaker 2: absolutely nothing to what Bruce is saying. I do think
- 12:05Speaker 2: I tend to I think we spoke about this the
- 12:06Speaker 2: last time as well. That you know, the minute they
- 12:10Speaker 2: come forward with something like that, then it gives the
- 12:13Speaker 2: impression that the capability is much greater than what it is,
- 12:17Speaker 2: and then that will push up the shape price if
- 12:21Speaker 2: you know, your company bold around the capability of it.
- 12:24Speaker 2: So I don't think there is nothing to what Bruce
- 12:28Speaker 2: is saying. I just don't want to be the guy
- 12:30Speaker 2: to to say it to accuse companies of doing that
- 12:35Speaker 2: sort of cynical, sort of cynical thing here for sure.
- 12:39Speaker 2: But yeah, yeah, let's let's talk.
- 12:41Speaker 1: About the layman's perspective from my advantage point, which is
- 12:44Speaker 1: like nowhere. From my advantage point, I'm thinking, can we
- 12:48Speaker 1: separate these autonomous things from from the hard ware? Isn't
- 12:53Speaker 1: that a very important kind of safety means?
- 12:58Speaker 2: What do you mean separated from hard way? Like, I
- 13:01Speaker 2: don't understand what we can.
- 13:02Speaker 1: We can you even can you beings have the switch?
- 13:06Speaker 2: Oh do you mean like.
- 13:10Speaker 1: Do you mean like banish them?
- 13:13Speaker 2: Banish them? Yeah? I was thinking about this as well.
- 13:16Speaker 2: You know, I was looking at one of the videos
- 13:18Speaker 2: of a of a of a I robot, and I
- 13:22Speaker 2: was thinking water guns are probably going to go up
- 13:24Speaker 2: in manufacturing and sales as well because or can they
- 13:28Speaker 2: be waterproof? A lot of the guys, how do we
- 13:31Speaker 2: destroy them? How do we destroy them? A lot of
- 13:35Speaker 2: the guys are part of the legislators call in the US,
- 13:39Speaker 2: calling for oversight or saying there must be ways to
- 13:44Speaker 2: shut down these ais.
- 13:47Speaker 1: I guess that's the question very clumsily. But I want,
- 13:49Speaker 1: I want to interrupt you because Vernon's on the line.
- 13:52Speaker 1: Vernon's on the line, and Vernon is more important than me.
- 13:54Speaker 1: Vernon go ahead, for doctor Wade Zeal.
- 13:59Speaker 3: Terrence, remember you from Titties by Chairs on the Rock.
- 14:02Speaker 1: I remember you singing on.
- 14:06Speaker 3: With Andrew. You remember that absolutely clearly.
- 14:10Speaker 1: You stole that beach. It was yours. Whatever you got
- 14:13Speaker 1: for Waight.
- 14:15Speaker 3: You know, Clarence, you know I've got a company, and
- 14:18Speaker 3: I've got an IR company.
- 14:20Speaker 4: I'm very extensively involved in. I also studied the professor
- 14:25Speaker 4: Goazinski at u ce T in your predecessors for your learning.
- 14:29Speaker 4: Because I brought uniques into the country.
- 14:31Speaker 3: The big, the big challenge we got, Jack Clarences, It's
- 14:34Speaker 3: not so much. I there's under to them, you know
- 14:37Speaker 3: rock Semini, you name, it is under them. It's the
- 14:40Speaker 3: input and the output which your speaker alluded to. But
- 14:45Speaker 3: the most important thing for AI is energy, energy and
- 14:49Speaker 3: quantum computers. Now, we haven't got the capability in South
- 14:53Speaker 3: Africa to my quantum computers, and quantum computers needs rare
- 14:58Speaker 3: earth mineral pickup. So if you haven't got quantum computers
- 15:02Speaker 3: and you haven't got energy, and you ever got a
- 15:04Speaker 3: good cooling system, you're gonna suckle with your AI. Now
- 15:08Speaker 3: you're talking about a driver, it says Elon Musk and
- 15:11Speaker 3: Elon gard Croc now Crooc AI is an executive type,
- 15:15Speaker 3: top level.
- 15:16Speaker 1: I I'm going to need you to be as short
- 15:20Speaker 1: as possible. I got to go to news quickly in
- 15:22Speaker 1: a short while, and and I'd like the doctor to respond.
- 15:25Speaker 3: Okay, So my main question is what is South Africa
- 15:29Speaker 3: and his counterparts and his colleagues worldwide doing about the
- 15:33Speaker 3: shortage of quantum computers and energy to drive them?
- 15:38Speaker 1: Thank you Vernond for that question. Dr Wade Seal Quantum
- 15:41Speaker 1: I thought that was our next headache for tomorrow.
- 15:44Speaker 2: Well, I think it is that. I think your listener
- 15:47Speaker 2: is absolutely right, it does raise serious questions. I think
- 15:50Speaker 2: there's a ruckus at the moment in the media about
- 15:53Speaker 2: data centers being opened up in Cape Town as well.
- 15:56Speaker 2: And data centers also consume an enormous amount of water
- 16:00Speaker 2: to cool down. And you know, we've got water shortages
- 16:04Speaker 2: in South afric in Cape Town, we've got where we've
- 16:06Speaker 2: had water issues in the past, we're looking at another
- 16:09Speaker 2: one now. So these are all important questions. But you
- 16:12Speaker 2: see the problem is that ethics doesn't live in the
- 16:15Speaker 2: black and white. It lives in the gray. And by
- 16:18Speaker 2: that I mean it's not a matter of choosing do
- 16:21Speaker 2: we do it or do we not do it. It's
- 16:22Speaker 2: a matter of do we allow data centers to come
- 16:26Speaker 2: here and give people jobs? Because jobs is a big
- 16:29Speaker 2: thing as well, So do we allow them to come
- 16:32Speaker 2: here go and then but then the trade off is
- 16:35Speaker 2: that they use water or do we say no, we
- 16:38Speaker 2: need our water, we need to preserve the water and
- 16:41Speaker 2: then sort of lock ourselves into this mister boat in
- 16:44Speaker 2: terms of jumping on the bandwagon with the AI wave
- 16:48Speaker 2: and so on and all the opportunities that it brings.
- 16:51Speaker 2: And I think the same question applies to quantum computing.
- 16:56Speaker 2: South Africa, it is not a poor country. We can
- 16:58Speaker 2: certainly procure the kind of computing that your listener refers to,
- 17:03Speaker 2: but at what expense, you know, that's the question, what
- 17:06Speaker 2: do what do we do we cut social grants for instance,
- 17:10Speaker 2: to procure expensive computers and so ethics loves in that
- 17:15Speaker 2: great area. I think there's no right choice, no right
- 17:17Speaker 2: easy answers, and we've got to work that sort of
- 17:21Speaker 2: thing out so that that would be my response to
- 17:24Speaker 2: to your listener.
- 17:26Speaker 1: Maybe maybe an opportunity that comes to that, it's somebody
- 17:28Speaker 1: that can can make these data centers run on seawater
- 17:31Speaker 1: could change the game.
- 17:34Speaker 2: Maybe no, no, Maybe I think there is research into that. Yeah,
- 17:37Speaker 2: there is research into that, but I think it's it's
- 17:39Speaker 2: the seawaters, the specific salt content is the problem for
- 17:44Speaker 2: using seawater. But I think there is research being happened
- 17:47Speaker 2: into that year.
- 17:48Speaker 1: Doctor White c We're going to rely on your expertise
- 17:51Speaker 1: into the future, Doctor White Seal a post doctor follow
- 17:53Speaker 1: in ethics AI ethics at the Institute for Humanities in Africa.
- 17:57Speaker 1: It doesn't sound at all by that, And yeah, maybe
- 18:03Speaker 1: I'm a little less secure than than him on this
- 18:05Speaker 1: particular front. What we do know there's going to be