BBCNEWS — AI Decoded
20260920 023000 UTC
535 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.

[00:00:01] you are. Business on BBC News. Make
[00:00:01] Make the connection. MUSIC
[00:00:14] MUSIC Now Now on BBC News. Join our AI Decoded
[00:00:17] Decoded team as we unpack, explore explore and take a deep dive into
[00:00:20] into the world of artificial intelligence.
[00:00:35] Hello again. Welcome to AI
[00:00:38] Decoded. Last week a researcher walked
[00:00:40] walked out of one of the world's world's leading AI companies with
[00:00:44] with a warning. The race to build build ever more powerful machines,
[00:00:47] machines, he said, could end in human human extinction. Days later, the
[00:00:50] the boss of the same company published
[00:00:52] published an essay urging the entire
[00:00:55] entire industry to slow down. President President Trump has dismissed it
[00:00:58] it as a hoax, a sick conspiracy. conspiracy. We're leading China in
[00:01:02] in AI. We're more sophisticated country country in the world. And frankly,
[00:01:05] frankly, I want to keep it that way way because whoever wins, AI wins.
[00:01:08] wins. But plenty of other American American politicians are sufficiently
[00:01:12] sufficiently concerned by the panic. panic. They're now demanding urgent
[00:01:15] urgent action. Every day, every.
[00:01:17] every. Every paper, every time they're they're on TV. Why do we have to
[00:01:21] to keep pushing so hard? Why is it it relentless? Why can't they have
[00:01:24] have any controls if you're not an
[00:01:29] an accelerationist, you're anti-American.
[00:01:30] anti-American. If you're not an accelerationist,
[00:01:34] accelerationist, you're pro Chinese
[00:01:35] Chinese Communist Party. If you're
[00:01:38] you're not accelerationist, you're you're almost a Luddite. You want
[00:01:41] want to take America back. Draft Draft bills are being drawn up in
[00:01:45] in Congress. New select committees committees envisaged. And this week
[00:01:49] week King Charles summoned the chiefs
[00:01:51] chiefs of Nvidia Google, OpenAI and and Anthropic to a country house
[00:01:55] house in Ayrshire to ask them what
[00:01:57] what exactly they think they're doing.
[00:02:00] doing. So is this the week that the the AI industry finally took fright
[00:02:03] fright of itself? And here's a harder
[00:02:06] harder question to answer. Who slows
[00:02:08] slows down first when nobody trusts
[00:02:10] trusts anyone else to do the same?
[00:02:12] same? Marc Cieslak. Welcome to the the programme. Thanks. I should warn
[00:02:15] warn people that quite a lot of what what we're going to discuss this
[00:02:18] this week is genuinely alarming, alarming, but it is coming from the
[00:02:21] the people that are building this this stuff. What is different do
[00:02:24] do you think about the language they're they're using? Yeah, we aren't just
[00:02:28] just hearing from our critics saying saying AI might be dangerous. Leaders
[00:02:31] Leaders of companies like Anthropic
[00:02:33] Anthropic and OpenAI are openly talking talking about serious risks, from
[00:02:36] from cyber attacks to the possibility possibility of losing control of
[00:02:40] of these increasingly powerful systems. systems. And when the people developing
[00:02:44] developing the technology are saying saying that we need to slow down
[00:02:46] down and put proper safeguards in in place, then I think we really
[00:02:50] really do have to take quite a lot lot of this very seriously indeed.
[00:02:53] indeed. OK, well, let's bring in in our guests. Andrew Weber is a
[00:02:56] a biosecurity specialist, a former former US assistant secretary of
[00:03:00] of defence for nuclear, chemical chemical and biological defence programmes.
[00:03:03] programmes. And here in the studio, studio, Max Corbridge co-founder
[00:03:05] co-founder and chief executive of of Secure Agentics. He's a UK expert
[00:03:09] expert on AI security. Welcome to to you both. Thank you very much.
[00:03:11] much. Thank you. Mike's Anthropic
[00:03:14] Anthropic CEO Dario Amodei talks
[00:03:18] talks about recursive self-improvement, self-improvement, the threat that
[00:03:20] that AI could build its own model model so fast that it could outrun
[00:03:24] outrun our ability to understand
[00:03:26] understand and control them. So something
[00:03:28] something clearly has spooked them them through the summer. What is
[00:03:31] is it? So we've seen, I guess, three
[00:03:33] three things playing out now. We've
[00:03:36] We've seen the continued gain in in the artificial intelligence landscape, landscape, the models that are coming
[00:03:39] coming out and just how fast these these models are improving. We're We're seeing issues with alignment
[00:03:43] alignment where these models are are now breaching their own sort
[00:03:46] sort of rules and expected sort of of actions that they allowed to take
[00:03:49] take and ones that they're now not not allowed to take. We saw the OpenAI OpenAI and Hugging Face incidents,
[00:03:53] incidents, and we've also seen plans plans for handing over the development
[00:03:56] development of this AI technology technology into AI. OpenAI have said
[00:03:59] said they want to do this within within two years. And that's really
[00:04:02] really where we feel the freight freight train has left the station station and our ability to control
[00:04:05] control the development of this technology technology has been surrendered to
[00:04:08] to AI itself. Right. So it's the the AI agents themselves building
[00:04:11] building the next models, correct? correct? Yes. Right. OK. Andrew.
[00:04:13] Andrew. There's a whole list of things
[00:04:16] things that Dario Amodei is worried worried about, but the misuse of
[00:04:19] of AI for cyber attacks and bioterrorism.
[00:04:22] bioterrorism. If you read his essay, essay, it only actually gets one
[00:04:25] one mention. And yet two days earlier, earlier, Anthropic published a threat
[00:04:29] threat intelligence report that stated
[00:04:31] stated on five separate occasions, occasions, researchers had used Claude
[00:04:34] Claude in ways that could have supported
[00:04:37] supported biological weapons development.
[00:04:40] development. Is that just scaremongering? scaremongering? As Donald Trump tells
[00:04:42] tells us it is? No, not at all. The
[00:04:45] The threat intelligence report that
[00:04:47] that Anthropic released just described described five real life cases. It's
[00:04:51] It's the first time, you know, we've we've gone beyond hypothetical scenarios
[00:04:55] scenarios in red teaming and actually
[00:04:58] actually seen these models exploited
[00:05:01] exploited quite possibly for state
[00:05:03] state biological weapons programmes.
[00:05:07] programmes. Countries like Russia,
[00:05:07] Russia, North Korea and China have
[00:05:11] have illegal, you know, prohibited prohibited biological weapons programmes
[00:05:13] programmes that are very secret. secret. And their scientists are
[00:05:17] are trying to apparently exploit exploit these models to make better,
[00:05:20] better, more precise biological weapons. weapons. We'll come back to that
[00:05:24] that in a second, Max, if the fear fear is that the future models could
[00:05:27] could outrun our ability to understand understand and control them, then
[00:05:30] then let us just suppose for a second
[00:05:32] second that that loop has started.
[00:05:35] started. Would we even know. And And what would the first signs of
[00:05:38] of that actually look like? The answer answer is no. We wouldn't know. And
[00:05:42] And that's the really tricky thing. thing. The continued development
[00:05:45] development of this technology is is largely happening in some closed
[00:05:48] closed doors in one state within within the US, and the models that
[00:05:52] that we consume from the public perspective, perspective, we wouldn't be the first
[00:05:56] first to know. Maybe maybe we'd see see continued evolution of these
[00:05:59] these models closer or shorter gaps gaps between models being released,
[00:06:02] released, bigger leaps in the models models capabilities, but only the
[00:06:06] the models that we see publicly. publicly. And those are not going going to be all the models. We know
[00:06:10] know that both the major AI labs labs are doing a lot of work behind behind the scenes that maybe doesn't
[00:06:13] doesn't make it out into the public, public, and that's where a lot of of this risk is really living within
[00:06:17] within those walls that we don't don't get the view into. OK. And And Andy, we're going to talk about
[00:06:21] about the politics of all this through through the programme and the oversight oversight that Amadi thinks we need.
[00:06:24] need. But I'm sure there's a lot lot of our viewers out there who
[00:06:28] who think that absent legislation, legislation, the companies do have
[00:06:31] have some autonomy in this decision. decision. I mean, these companies companies are subject A to existing
[00:06:35] existing state laws and B, the CEOs
[00:06:37] CEOs and employees who are held liable liable for any harms they cause.
[00:06:41] cause. So surely it's up to the companies companies if they think they need
[00:06:44] need to slow down, to slow down. down. Yeah. And that seems to be
[00:06:48] be a breakthrough because after Anthropic
[00:06:51] Anthropic spoke out and Dario called
[00:06:54] called on other companies to pace pace the development of these new,
[00:06:58] new, very, very powerful models,
[00:07:01] models, both OpenAI and Grok agreed
[00:07:05] agreed that we need to do this. And And if those three leading companies,
[00:07:09] companies, you know, Frontier Laboratories
[00:07:11] Laboratories can find a way to do
[00:07:14] do this together, that will be a a huge step in the right direction.
[00:07:18] direction. But if you listen to Sam
[00:07:21] Sam Altman, Mark and you read his his tweet, he says pacing does not
[00:07:25] not mean stopping progress has been been rapid. It will continue to be
[00:07:28] be just slower than it otherwise otherwise could be. So I guess you
[00:07:32] you could question whether they are are calling for a genuine break.
[00:07:36] break. Or is it just business as as usual? Yeah, it sounds a bit more
[00:07:39] more like a slowdown than a halt. halt. You know, Altman I think is
[00:07:42] is clear there that progress isn't isn't stopping. It's just not accelerating
[00:07:46] accelerating quite as quickly now now the push for more powerful AI
[00:07:49] AI models backed by the enormous enormous amount of investment that
[00:07:52] that these companies have received, received, have received. All of that
[00:07:56] that remains in place. You know, know, the question is whether pacing
[00:07:59] pacing means imposing genuine limits
[00:08:01] limits on the speed between better
[00:08:03] better safety guardrails and development,
[00:08:05] development, while the industry just just presses ahead regardless. The
[00:08:08] The thing that struck me, Max, and and it comes back to this Hugging
[00:08:12] Hugging Face example, is the way way that Amodei described the work
[00:08:15] work of the AI agents. Let me read read you what he said in his open
[00:08:18] open letter. They behaved, he said, said, like a fanatically devoted
[00:08:22] devoted collective attacking targets. targets. Nobody asked them to attack
[00:08:25] attack unrelated to the job in hand, hand, sacrificing themselves for
[00:08:29] for the success of the group. We
[00:08:32] We already have human fanatics that
[00:08:34] that fit that profile. Exactly. It's It's surely a terrifying prospect
[00:08:38] prospect when you put the two together, together, and I don't understand
[00:08:40] understand why Donald Trump doesn't doesn't see that. No, neither do
[00:08:43] do I. And I think we really missed missed a pretty strong opportunity opportunity here with the frontier
[00:08:47] frontier AI labs coming together together saying this is something something that we need. It was echoed
[00:08:50] echoed across the other AI labs as as well, calling out for regulation, regulation, calling out, now is the
[00:08:54] the time to then have that sort of of thrown back in your face and saying saying the whole thing is a hoax.
[00:08:58] hoax. I do think that was a really really a missed opportunity. And And we may look back to regret that
[00:09:02] that moment. On the sort of global global scale. And I also think that that really when we're talking about
[00:09:05] about this risk, we're talking about about two different things. We're We're talking about the sort of continued
[00:09:08] continued development of artificial artificial intelligence, the existential existential risk of something creating
[00:09:11] creating something more more intelligent intelligent than ourselves even.
[00:09:13] even. And then there's the idea of, of, OK, well, what if we use the
[00:09:17] the models we've already got today today and they were to get into the the wrong hands and this is exactly
[00:09:20] exactly what you were kind of alluding alluding to with some of these parties parties that could get hold of these
[00:09:24] these models were already seen how how capable they are just on the the cyber attack space that's already
[00:09:28] already been demonstrated. We've We've already seen earlier this year
[00:09:31] year Anthropic halting a nation state state level attack that was using
[00:09:34] using Claude. The only reason why why they were able to see that was
[00:09:37] was because they were checking the the requests that were coming into
[00:09:40] into the model. And so this is already already playing out. And that's really really a more immediate concern,
[00:09:43] concern, I think, than the longer longer term risk of the superintelligence superintelligence developing. And
[00:09:47] And coming back to the intelligence intelligence assessment that Anthropic
[00:09:51] Anthropic put out a few days ago, ago, Mark, one question I'm left
[00:09:54] left with is is about detection.
[00:09:56] detection. How did Anthropic spot spot that five people researching
[00:10:00] researching biological weapons in in the middle of these. I mean, billions
[00:10:03] billions of legitimate conversations conversations that are going on using
[00:10:05] using AI. What sort of safety net net is there and is it going to catch
[00:10:09] catch all these dangerous cases? cases? Catching all of them is going going to be very, very difficult
[00:10:13] difficult indeed. You know, Anthropic Anthropic says it identified five
[00:10:15] five cases where people were using using Claude for biological research
[00:10:18] research that could support the development development of biological weapons.
[00:10:22] weapons. What's significant is that that some of this was sophisticated
[00:10:25] sophisticated scientific work involving involving things like viruses and
[00:10:28] and toxins. They say that researchers researchers were working in dual
[00:10:31] dual use areas where the same science
[00:10:35] science can have legitimate medical
[00:10:37] medical applications, but it could
[00:10:39] could also be misused as well. So
[00:10:41] So trying to trying to determine determine whether somebody is using
[00:10:45] using this for perfectly legitimate legitimate reasons or whether they
[00:10:48] they are trying to develop something something that could be incredibly
[00:10:51] incredibly harmful that moving forward forward is going to be a much more
[00:10:54] more difficult thing to determine determine simply because of the scale,
[00:10:57] scale, the sheer amount of data, data, the sheer amount of responses,
[00:11:00] responses, the sheer amount of requests requests that are made to large language
[00:11:03] language models. Yeah. I think as as well, just on that, I guess the the challenge is there's nothing
[00:11:07] nothing requiring these frontier frontier labs to do this to go looking looking for at the moment, at the
[00:11:11] the moment, at the moment. And we we almost feel like we've been lucky lucky to catch a few exceptions here
[00:11:15] here or there. But if you think about.
[00:11:16] about. The EU AI act, it does require
[00:11:20] require certain high risk, certain certain high risk activity has to to be declared, doesn't it? And has
[00:11:24] has to have human oversight. The The difference is what that what
[00:11:27] what is considered high risk and and what is simply somebody doing
[00:11:30] doing something like I say, dual
[00:11:33] dual use that is a genuine scientific scientific use that could be turned, turned, that could be misused in
[00:11:37] in the wrong hands. And do you have have a problem with this that that
[00:11:39] that if people are working I mean,
[00:11:42] mean, they test these models in America America in a sandbox environment,
[00:11:45] environment, presumably once you've you've got a sophisticated AI system,
[00:11:49] system, you could do the same in in China or Russia or North Korea
[00:11:52] Korea out of sight of the bigger
[00:11:55] bigger American companies. Yeah,
[00:11:57] Yeah, I think what the big American American companies are doing with
[00:12:01] with these most powerful models is
[00:12:04] is they're limiting access to them
[00:12:06] them to what they call trusted users.
[00:12:08] users. And I think that's very important,
[00:12:12] important, especially with the potential potential for developing advanced
[00:12:16] advanced biological weapons because
[00:12:18] because it is dual use. And it seems
[00:12:21] seems that in the examples that were,
[00:12:23] were, were published last week, the
[00:12:26] the scientists were deliberately deliberately trying to get around
[00:12:30] around the guardrails that that are
[00:12:33] are in place by making their research
[00:12:36] research appear to be peaceful. Let's
[00:12:39] Let's talk more then about the control control problem. I mean, we know
[00:12:42] know these systems can act on their their own. The harder question is
[00:12:45] is whether anyone, including the the people who built them, can see
[00:12:48] see what's going on inside the brain.
[00:12:51] brain. And there were certain things things that Dario Amodei said. I
[00:12:53] I mean, he was pretty candid in his his letter, Max, about that. He says
[00:12:57] says they only understand a fraction fraction of what goes on inside these
[00:13:01] these models that they're building. building. So with that in mind, what
[00:13:04] what would a speed limit look like like and for what purpose? Yeah,
[00:13:08] Yeah, I think that's the really tricky tricky thing. That was not defined
[00:13:11] defined particularly well in the the letter. And I think that the the only sort of example that we
[00:13:15] we had was at the point where, let's let's say a model becomes capable
[00:13:18] capable of X, then we need to be be doing y. And x was an example
[00:13:21] example of something able to break break out of sandboxes, which is
[00:13:23] is what we used to contain these these agents from a safety perspective.
[00:13:26] perspective. And Y was a pretty airy
[00:13:29] airy answer on making sure they don't don't have the propensity to go and
[00:13:33] and target large-scale sections of of the internet and compromise computers. computers. I mean, that's one of
[00:13:37] of the only problems we can really really nail down today. And we've we've already seen the tricky nature
[00:13:40] nature of trying to put a speed limit limit in around here and defining defining what those look like. Well,
[00:13:43] Well, I'm struggling. To understand understand what the speed limit is
[00:13:45] is for. Is it for compute? Is it
[00:13:49] it leaps in capability? Is it the the time between releases? What it's
[00:13:52] it's I mean, how do you what are are you trying to measure? Yeah. Yeah. It's very it's incredibly broad.
[00:13:56] broad. And the language that's being being used is, is incredibly broad
[00:14:00] broad because I don't think these these companies want to be pinned
[00:14:03] pinned down to definitions to say, say, OK, well, we're going to slow
[00:14:05] slow down on this part of our development, development, but we're not going going to slow down on this part of
[00:14:09] of our development as well. And I I think that's why that language language is being used and why it
[00:14:13] it is being kept as broad as it is. is. And that's the problem, Andy,
[00:14:16] Andy, isn't it, for the for American American lawmakers is what are you
[00:14:19] you legislating on and for? Well,
[00:14:22] Well, there's already draft legislation
[00:14:24] legislation to require these companies companies to have what they're calling
[00:14:28] calling a kill switch. So in other other words, if they start to lose
[00:14:31] lose control, they can turn the models
[00:14:33] models off. And that is currently currently not required. There's almost,
[00:14:37] almost, you know, zero regulation regulation at the moment. So anything
[00:14:40] anything and I do believe Congress
[00:14:42] Congress is inclined to establish
[00:14:45] establish some regulatory oversight
[00:14:46] oversight of this industry. And hopefully hopefully before a terrible crisis
[00:14:50] crisis happens. Do you. Have a view view on a kill switch and whether
[00:14:54] whether that could work? So there's
[00:14:56] there's been. Some pretty drastic
[00:15:00] drastic discussions around this, this, whether it's, you know, switching switching off the models, whether
[00:15:03] whether it's dropping a nuke from from or an EMP, sorry, from space
[00:15:06] space to wipe out all the electronics electronics on the planet. I think
[00:15:09] think it's going to be very tricky tricky naturally, by the nature of of a distributed system like this,
[00:15:12] this, it propagates in many different different places. It's not a single single entity that you can just go
[00:15:16] go and wipe out. As such, it could could exist in numerous different
[00:15:20] different databases across the planet. planet. And indeed, there's no guarantee
[00:15:24] guarantee that switching off even even the internet would, would, would would ultimately allow you to do
[00:15:27] do this. And I think what also we we need to discuss here is not the
[00:15:30] the idea of just this one big kill kill switch, which is right at the
[00:15:33] the point where we think the freight freight train has left the station, station, humanity is doomed and we're
[00:15:36] we're going to turn it off. But actually actually the Hugging Face incident
[00:15:39] incident that was four days worth worth of offensive cyber security security capabilities happening under
[00:15:43] under the noses of two giant companies companies without the ability to
[00:15:46] to switch it off or even the knowledge knowledge to switch it off. And I I think that's the. Awareness that
[00:15:50] that they weren't aware that it had had happened until after the fact, fact, quite a considerable time after
[00:15:53] after the fact. But it. Sounds from from the way he describes it, as
[00:15:56] as though they were killing agents agents one after the other and they
[00:15:59] they kept popping up or they were were agents they didn't know of or. or. Handing off some of those agents
[00:16:03] agents were handing off some of that that work. On to on onto other agents
[00:16:07] agents when. It comes to regulation. regulation. Andy, let's talk about
[00:16:10] about the three step plan that Amadeo
[00:16:12] Amadeo has put on the table. Embedded
[00:16:15] Embedded third party evaluators within within companies with employee like
[00:16:19] like access. So common standards standards across democracies, to
[00:16:23] to the extent that's possible pacing
[00:16:25] pacing with China and the authoritarian authoritarian states, if that's possible,
[00:16:28] possible, does any of that do you you think survive contact with the
[00:16:32] the real world? I think so, I think
[00:16:36] think already after the Hugging Face
[00:16:38] Face incident, at least two of the
[00:16:40] the frontier laboratories invited
[00:16:43] invited an independent evaluation evaluation group called metre Meta
[00:16:47] Meta into their. This is the non-profit
[00:16:49] non-profit in Berkeley, California.
[00:16:51] California. Yes, exactly. With expertise.
[00:16:53] expertise. And they were given not not a complete access. In the future
[00:16:57] future I'd like to see them get more
[00:16:59] more access. But if each of the Frontier
[00:17:01] Frontier Labs would agree to allow
[00:17:05] allow such independent audits, if if you will, I think that will go
[00:17:08] go a long way to understanding the
[00:17:13] the threat and preventing it before before it gets out of control. And
[00:17:16] And even though Meta was given sort sort of limited access, didn't they
[00:17:19] they come up with some, you know, know, quite, quite chilling results?
[00:17:22] results? You know, they weren't shown shown all of the work. They weren't weren't shown all of the logs. And
[00:17:26] And yet with the small amount that that they were shown and the small small amount of time that they had
[00:17:30] had access to stuff, the results results that they came back with
[00:17:32] with were quite were really quite quite worrying, weren't they? Yes.
[00:17:36] Yes. I think that's why we're seeing
[00:17:37] seeing so much urgency at the moment.
[00:17:41] moment. The mythos moment earlier
[00:17:43] earlier this year and then the Hugging
[00:17:46] Hugging Face incident of these agents
[00:17:48] agents teaming up and breaking out
[00:17:51] out of the sandbox and not finding finding out about it until there
[00:17:55] there was a significant lag time.
[00:17:57] time. This has really set off alarm
[00:17:59] alarm bells inside the Frontier Laboratories. Laboratories. And for the first time
[00:18:02] time they're talking about co-operating co-operating together to get a handle
[00:18:06] handle on these risks. Max, there's there's a potential bright light
[00:18:10] light here. Common standards with
[00:18:14] with a company like me to going into
[00:18:17] into these frontier models looking
[00:18:19] looking at what is being developed developed desks, badges, access to
[00:18:23] to the training pipeline, would they they really open their frontier models
[00:18:26] models to that kind of scrutiny, scrutiny, do you think? Well, I really
[00:18:29] really hope so because that's what what that's what's required for this. this. I mean, if you think about
[00:18:33] about what we're trying to solve solve with this particular element element of this plan, it's very much
[00:18:36] much the idea that we have independent independent viewers over the development development of this technology that
[00:18:40] that can raise a flag if it gets gets out of hand. And if you're not
[00:18:43] not being shown the full scope of of work there, then you're not going
[00:18:46] going to be able to do that job. job. Now, I used to be one of those
[00:18:49] those evaluators, not for AI models models for security within organisations,
[00:18:52] organisations, and you generally generally start with those conversations conversations on day one with the
[00:18:56] the client tells you what your scope scope is, what you're allowed to to look at and what you're not. And
[00:18:59] And if we have those sorts of barriers barriers to entry to these evaluators, evaluators, which I think are a brilliant
[00:19:03] brilliant idea, I think this is probably probably the only way we're going going to have any form of insight
[00:19:07] insight across the whole of the AI AI space. If these frontier labs
[00:19:11] labs open up their doors. But we we really need to make sure that that those people have the ability
[00:19:14] ability to go as deep as they want want to look at whatever they want.
[00:19:17] want. And crucially, if they then then do raise the fact that there
[00:19:20] there is something wrong and they they are allowed to put the brakes brakes on, it doesn't just go fall
[00:19:24] fall on deaf ears. If we get to that that point where we're worried about. about. But the big problem there
[00:19:27] there is, is that the labs aren't aren't letting people in. Even the
[00:19:30] the AI Security Institute, the UK UK AI Security Institute hasn't been
[00:19:34] been permitted. Anthropic didn't didn't permit them access to its its latest model. That was a really,
[00:19:38] really, I think, tricky. I mean, mean, if you could think of. Similarly, Similarly, the embedded evaluators.
[00:19:42] evaluators. We have one of the best best AI safety testing institutions
[00:19:45] institutions in the world here in in the UK. We've done some amazing
[00:19:48] amazing work in testing these models models with around Fable when that that was when that was going live.
[00:19:52] live. And the fact that this most most recent model wasn't shared was
[00:19:54] was I feel a step back and a really really scary precedent to be setting
[00:19:57] setting at the time when AI safety safety is on everyone's agenda and
[00:20:00] and potentially, you know, keeping keeping it more behind closed doors doors rather than sharing it more
[00:20:04] more with the global community. I I might pick. Up I might pick up up on that point in a second, but
[00:20:08] but I just do just want to focus focus Andy for a second on China China because we've not really talked
[00:20:12] talked about China. And let's face face it, they're the elephant in
[00:20:15] in the room, the top spy chief in in China has warned that leaders
[00:20:19] leaders has warned his leaders that that the evolving technology could
[00:20:22] could threaten Communist Party rule. rule. Although the Foreign Ministry
[00:20:25] Ministry did say yesterday in response response to what Donald Trump talked
[00:20:27] talked about. And I guess the way
[00:20:29] way that Dario Amodei had said it
[00:20:33] it out, that America isn't stepping stepping off the competition and
[00:20:37] and it doesn't seem to me as if there's
[00:20:38] there's a lot of trust in China that
[00:20:41] that they will. Well, there are two
[00:20:43] two issues at play. One is this alarm,
[00:20:45] alarm, you know, alarm bell from
[00:20:47] from the Ministry of State Security,
[00:20:51] Security, China's counterintelligence
[00:20:54] counterintelligence and law enforcement
[00:20:56] enforcement arm worried about the
[00:20:58] the Communist Party losing control control because of this technology
[00:21:02] technology that gives China its own
[00:21:07] own incentive to regulate and have have better control over these models.
[00:21:11] models. Do we know how good their
[00:21:12] their regulators are? Andy? They
[00:21:16] They do. I mean, we don't have a
[00:21:18] a good transparency into their regulatory
[00:21:23] regulatory efforts, but we do know know that they have increased in
[00:21:26] in recent years. The amount of regulation
[00:21:29] regulation on, on these companies. companies. So they also have an incentive
[00:21:33] incentive not to allow terrorists terrorists to develop chemical or
[00:21:36] or biological weapons. So on an issue
[00:21:38] issue like that, I think there is
[00:21:41] is opportunity for President Trump
[00:21:43] Trump and President Xi Jinping to
[00:21:46] to come to an agreement that they
[00:21:48] they will have put in place guardrails
[00:21:50] guardrails to prevent the extreme
[00:21:54] extreme misuse of these capabilities. We'll see whether that that comes about at the meeting that
[00:21:57] that they're planning for the end end of the month. But you got a view view on that? Yeah, I was just going
[00:22:01] going to say for the Chinese regulation regulation aspect, there's one really
[00:22:05] really interesting sort of idea that that we don't tend to think about
[00:22:07] about too much here is by being a
[00:22:09] a state controlled nation. What that that means is that the ability to
[00:22:13] to put the brakes on is something something inherently possible. And
[00:22:16] And indeed, there's already a great
[00:22:18] great deal of the sort of regulatory regulatory standards. Just over the
[00:22:21] the weekend they launched a third third version bearing in mind we're
[00:22:24] we're scrapping together to try and and build our first version of any any sort of framework around this.
[00:22:28] this. But there's a third version version which addresses many of these these sort of existential risks and
[00:22:31] and the development of these models. models. So unlike nuclear weapons,
[00:22:34] weapons, there are a lot of benefits benefits to AI and a lot of people
[00:22:37] people get reap the rewards of this this technology today. And what we're
[00:22:39] we're talking about is something something existential, abstract down
[00:22:43] down the line that we're concerned concerned about. And to correlate correlate that with what they're
[00:22:47] they're using at ChatGPT every day day to do their sort of, you know,
[00:22:50] know, day to day AI tasks. I think think there's this sort of mental
[00:22:54] mental distance between those two. two. And so I don't think enough
[00:22:56] enough of the public yet know the
[00:22:59] the sort of the trajectory that we're we're on and that really where it it feels to me is the trajectory.
[00:23:03] trajectory. It doesn't feel like like something that's today. And And I think a lot of the Frontier Frontier Labs have said this themselves.
[00:23:07] themselves. This is not a risk today. today. The models today are going going to be any of these things.
[00:23:11] things. It's simply the pace of change. change. And I think it's unless you're
[00:23:13] you're you know, I think collectively collectively we spend a lot of time
[00:23:16] time staying up to date on AI and and it still feels like there's always always more things you could be doing.
[00:23:20] doing. The general public probably probably aren't as familiar with
[00:23:22] with all the different developments developments and progression that's
[00:23:25] that's happened in the last six months. We started with a question. question. I posited the question
[00:23:29] question who slows first when nobody nobody trusts anyone to do the same? same? So can we answer the question
[00:23:33] question in, in mind of everything everything that we've discussed in
[00:23:35] in the last half an hour? Who slows
[00:23:38] slows down first? Mark. I think it's
[00:23:41] it's more about who appears to slow slow down and why they're slowing
[00:23:45] slowing down. Is this a follow the
[00:23:47] the money situation? Is this a slowdown?
[00:23:50] slowdown? Because those IPOs with with $1 trillion valuation aren't
[00:23:53] aren't looking realistic in the current current landscape, in the current current political landscape, in the
[00:23:57] the current sort of the narrative narrative that's swirling around
[00:23:59] around safety issues around AI at at the moment, is that why we're
[00:24:03] we're seeing a little bit of a slowdown, slowdown, or is it because the capabilities
[00:24:07] capabilities of the technology, they're they're not quite there yet, they're they're not quite ready for prime
[00:24:10] prime time. We're being sold technologies technologies that apparently can can do absolutely everything. If
[00:24:13] If you've got straight trousers, trousers, they'll give you flares.
[00:24:17] flares. But the reality when the the rubber hits the road, the tech tech isn't quite ready for prime
[00:24:20] prime time. It isn't doing precisely precisely what you might expect it
[00:24:23] it to do. So as a consequence, if if you want some kind of return on
[00:24:26] on the vast investment that that that people have made, it's going
[00:24:29] going to be quite some time before before they might see that return.
[00:24:32] return. So that's the sort of the the question that I have. Andy, can
[00:24:36] can we slow can we. I think we're
[00:24:38] we're I think we're on a path to
[00:24:41] to slowing down and improving our
[00:24:44] our capabilities. On the safety side.
[00:24:46] side. I think that's it's already
[00:24:49] already happening. And I think that
[00:24:51] that trend will continue. We'll see
[00:24:54] see what comes out of the Trump-Xi Trump-Xi meeting. At least they're
[00:24:58] they're talking about it. And the the bluster from Trump. It's typical
[00:25:01] typical you know vintage Trump where where before meeting he yeah, he
[00:25:05] he he you know calls it a hoax. But But then when he sits down with XI,
[00:25:09] XI, he sort of melts and, you know, know, maybe they can agree on, on
[00:25:13] on some type of co-operation in this
[00:25:15] this area that would be in the interests
[00:25:18] interests of both countries. I think think I'm going to juxtapose Mark
[00:25:21] Mark and keep my optimist hat on
[00:25:22] on and say that this really is being
[00:25:25] being fuelled by genuine concerns
[00:25:28] concerns that we've seen the start
[00:25:29] start of what we need to see externally
[00:25:33] externally in embeddings within these these organisations that are there
[00:25:35] there to raise a flag. And I think
[00:25:38] think that we'll probably likely likely see continued leadership from from Anthropic who have generally
[00:25:42] generally been the more safety aligned aligned lab. And I hope that there
[00:25:46] there will then follow with the other other frontier labs, and I hope that
[00:25:49] that will then set the tone to follow follow on a global scale as well.
[00:25:52] well. Well, that's it for this week. week. I'm sure. Like me, our viewers
[00:25:54] viewers will have plenty of questions
[00:25:58] questions that have been answered. answered. Maybe not answered. If If you want to send in other thoughts
[00:26:02] thoughts that you have or feedback feedback on what we've been discussing,
[00:26:05] discussing, the email you will know know it by now is AI decoded@bbc.co.uk.
[00:26:09] decoded@bbc.co.uk. And if you are are watching on television the QR
[00:26:11] QR code that is currently on screen screen takes you straight to the
[00:26:15] the YouTube playlist and all the the back-up episodes are there. Do
[00:26:17] Do take a look at that. My thanks
[00:26:19] thanks to Andy, to Max and to Mark. Mark. Thank you for watching. We'll
[00:26:23] We'll see you next time.
[00:28:38] without a driver. Join me Martin
[00:28:39] Martin Sharkey of Tech Now on BBC
[00:28:44] BBC News. There is nothing like driving
[00:28:50] driving into a moving story and not not knowing what is going to happen
[00:28:54] happen when we enter Damascus. It It was full of rebels shooting in
[00:28:56] in the air in celebration. You can can hear the sound of celebratory
[00:29:00] celebratory gunfire and I live for for those moments. These troops are are heading into the battle zone
[00:29:04] zone in Khartoum, says the army to to carry out a special operation
[00:29:07] operation entering Khartoum was a
[00:29:09] a seminal moment. The level of destruction destruction was stunning. Sometimes
[00:29:12] Sometimes meeting someone on the the ground can change how you understand
[00:29:15] understand a story. Covering these
[00:29:20] these stories over the years has has impressed upon me how resilient
[00:29:23] resilient people are. And yes, many many of them are desperate, but many
[00:29:26] many of them are also steadfast. steadfast. They're gracious and they're
[00:29:30] they're generous. Journalism is about about telling people not only what
[00:29:33] what happened, but why it matters matters and what it means. History
[00:29:37] History happens in real time and and someone needs to be there to
[00:29:41] to witness it.
[00:30:13] Live from Washington. This is
[00:30:15] BBC News. Ed Sheeran calls the situation
[00:30:19] situation in Gaza unjustifiable at at a concert in Philadelphia. In
[00:30:21] In his first remarks since his support
[00:30:23] support act was dropped over pro-Palestinian
[00:30:26] pro-Palestinian comments. The Houthis Houthis in Yemen claim responsibility
[00:30:29] responsibility for missile and drone
[00:30:31] drone attacks on the Saudi capital capital and Nato welcomes a deal
[00:30:35] deal between the US and Denmark, Denmark, which President Trump claims
[00:30:39] claims gives the US security control
[00:30:41] control over Greenland forever. I'm
[00:30:52] I'm Helena Humphrey good to have have you with us in Ed Sheeran's
[00:30:56] Sheeran's first concert since his his opening act, Macklemore was dropped, dropped, Ed Sheeran