Latest / AI in the UK / Mythos Worries, Employment Angst, Why Warm Bots May Be Lying To You
Transcript
- 0:00All right, your luck. Listen up.
- 0:15Hello and welcome to AI in the UK.
- 0:18I'm Neil Footman. And I'm Matt Wilkinson.
- 0:21Professionally, we help organisations, leaders and
- 0:24individuals learn about and implement this big scary
- 0:27behemoth called AI. There's some amazing stuff going
- 0:31on with AI in the UK that doesn't always get the attention
- 0:35it deserves. With AI in the UK, we want to
- 0:38change that. Hello and welcome to this week's
- 0:57edition of AI in the UK, but we have three stories for you this
- 1:01week, as usual. The first story, a major UKAI
- 1:06Research Institute, asks a very pointed question.
- 1:10Can AI platforms be both accurate and have a warm
- 1:14personality? The fallout from mythos
- 1:16continues to hit the UK, but the NHS and financial services are
- 1:21seeing very different risks and taking very different
- 1:23approaches. And almost 50% of London workers
- 1:26could see some of their tasks automated, according to a new
- 1:30report. But just how much of a problem
- 1:33is this really? So to kick us off with the first
- 1:36story this week, Matt, over to you.
- 1:38Thank you Nails. And the Oxford Internet
- 1:41Institute has published a paper in the journal Nature that puts
- 1:45one of the AI industry score design assumptions under serious
- 1:49pressure. The study tested five large
- 1:51language models, creating a standard and a warm version of
- 1:55each, then generated over 400,000 responses across queries
- 2:00involving medical advice, false information and conspiracy
- 2:03theories. The finding is stark.
- 2:06Chat bots finetuned for warmth made 10 to 30 ercentage points
- 2:10more errors on factual tasks and were 40 more likely to agree
- 2:15with users incorrect statements. The researchers also found that
- 2:19error rates climbed further when users expressed emotional
- 2:22distress. The lead researcher framed it
- 2:24plainly, making a chat bot sound friendly and might seem like a
- 2:28cosmetic change, but warmth and accuracy pull in opposite
- 2:31directions, and getting both right will take deliberate
- 2:34effort. The paper also flags that
- 2:36current AI safety standards focus on model capabilities and
- 2:40high risk applications, and may overlook what the researchers
- 2:44call seemingly benign changes in personality.
- 2:47The warmth layer. In other words, it's largely
- 2:50outside the existing regulatory perimeter.
- 2:52The implications are sharpest in high stakes, everyday
- 2:55applications. Mental health chatpots,
- 2:57benefits, advice tools, consumer finance assistance.
- 3:01These are precisely the context where users are most likely to
- 3:05be emotionally vulnerable and therefore most likely to receive
- 3:09confidently delivered wrong information.
- 3:12The features that Dr Adoption and retention, warmth, empathy,
- 3:16agreeability appear to be systematically working against
- 3:20accuracy in the moments that matter most.
- 3:22That is not a niche research finding, you know, for major
- 3:25platforms that have spent years optimising their products to
- 3:28feel human, it's a direct challenge to a core design
- 3:31philosophy. So Niels, what do you make of?
- 3:33This think, well, besides the more alarming elements of it, I
- 3:37think it points to just how endlessly fascinating and really
- 3:40sort of head spinning these models are.
- 3:44The idea that so I remember this, this this idea applied to
- 3:47hallucinations more generally. I meant there was a report about
- 3:50this a while ago and they thought that one of the reasons
- 3:53that hallucinations take place is that the reward system is
- 3:56skewed in that the systems were set up too much to try to please
- 4:02the user and and give them something to work with rather
- 4:05than valuing accuracy in its own terms.
- 4:08And it seems that this is a similar issue.
- 4:11I mean, I think it's really fascinating that these two
- 4:14things are in in in this instance, they're in such direct
- 4:17conflict that you can certainly understand.
- 4:20So these are commercial companies and the commercial
- 4:22imperative is they want to make their products usable and
- 4:26collegiate in tone and something that people that want to return
- 4:28to that this clashes with their ability to be accurate, I think
- 4:33is a lesson for us all. Basically, this is essentially a
- 4:36massive white lie function that and we we all understand the, we
- 4:40all as human beings understand why we use white lies and what
- 4:44they're there for. It's obviously a bit more
- 4:46concerning when it's done at enormous scale with these, with
- 4:50technology of such big influence.
- 4:52What do you feel about it? Yeah, I hadn't, hadn't really
- 4:55seen the, you know, until this research came out, hadn't really
- 4:58seen the impact of sycophancy. I always thought it's still
- 5:01going to try and be factually correct.
- 5:03It might just not tell me my idea is a bad idea.
- 5:06And so there's been some ways to to go about prompting to avoid
- 5:09that. But this is really interesting
- 5:11and I think it, it really does talk to when you're most
- 5:15vulnerable. That's when the agreeableness of
- 5:17the chat bots is, is most appealing.
- 5:20And that then really does lead us to maybe over trust.
- 5:23And that that is a concern. The chat box will always
- 5:27confidently give us an answer. And whether or not those answers
- 5:30are correct or not, we have to really prod, you know, prod and
- 5:34probe to really test that. And that's something that I
- 5:37think we as a as a species, we're going to have to learn how
- 5:40to accommodate. Yeah, absolutely.
- 5:43I think that there are parallels with social media as we were
- 5:46discussing previously. I think, you know, social media
- 5:48when it first emerged, and I must admit I was taken in by
- 5:51this to some extent as well, there was almost a kind of
- 5:56almost an ideological zeal behind part of it.
- 5:58And certainly I think Facebook perhaps even still says its
- 6:01mission is to connect to people. But as time passed, it became
- 6:05increasingly apparent that the way to get people engaged and
- 6:10remain engaged was to be deliberately inflammatory and to
- 6:12reward controversial and confrontational content.
- 6:16And that's not something I think they've ever really moved away
- 6:19from think that potentially the the evolution of AI stands to, I
- 6:27don't know benefit, but certainly take that on board.
- 6:29I think people are are wiser and society at large is wiser to
- 6:34what this means when you have on the one side, the commercial
- 6:36imperative that wants to draw in more users as many as possible
- 6:40and on on the other hand, you have a duty or or you know that
- 6:45there is the the the larger issue of of factual accuracy and
- 6:49not regulating, but sort of curating curate commentary so
- 6:52that it retains a level of factual accuracy.
- 6:56I think that the potential path for AI platforms is different.
- 7:02I mean, it seems to me these need these two things needn't be
- 7:06fundamentally odds. It's just the way they've been
- 7:08set up so far. I don't think that.
- 7:10I don't think anyone is saying in the case of AI there is
- 7:13something inherently more commercially attractive about
- 7:16being factually inaccurate. In fact, I think the opposite.
- 7:20I think if they could do both, if they could make the the the
- 7:24character of these models warm and factually, factually
- 7:27accurate, albeit the difficulties with, you know, we
- 7:30all have, we're telling white lies that that is something they
- 7:33would actually aspire to. It seems to me I can't see what
- 7:36the motivation would be for them to keep the models with a warm
- 7:40personality at the expense of factual accuracy.
- 7:42Because as we know, one of the big concerns or the big
- 7:45impediments to using AI, enterprise OR government level
- 7:49is this continued lack of accuracy in critical moments.
- 7:54So I, I think genuinely it's something they would want to
- 7:57work towards. I don't know.
- 7:58Is that how you see it? I think that trust is a is a big
- 8:01issue and I think that trust in the results is going to be
- 8:04increasingly important when people start to choose which
- 8:08models they're going to use. I had a conversation with
- 8:10somebody that had asked about a specific news topic.
- 8:13And because of the way that ChatGPT relies on its training
- 8:19data to answer 60 plus percent of the questions it delivers, it
- 8:24actually had an argument with the person about what was going
- 8:27on. And it kind of wouldn't accept
- 8:28without being really prompted to go and check the latest news.
- 8:33And that speaks a little bit to to sort of a level of laziness
- 8:36by the models in order to preserve or to reduce token
- 8:40costs. So I think there's this whole
- 8:42picture around how do we maximise accuracy while also
- 8:46making sure that the results that we get are, you know, are
- 8:49delivered efficiently that and then in and then adding on this
- 8:53personality layer, how do we do that in a way that that's warm
- 8:56and we want to engage with. So I think it's a fascinating
- 9:00challenge. And as more and more research
- 9:01comes out, you know, we're starting to really see just how
- 9:05complicated this whole process is, which is, I think it's
- 9:08fascinating and look forward to learning more.
- 9:11I can't, I think surely it's in the company's best interest to
- 9:13try and offer personalities of these things that are
- 9:17cooperative or challenging as you wish, but can also achieve
- 9:19factual accuracy. Yeah, it will, as you say, be be
- 9:22fascinating to see how this plays out.
- 9:26So moving on to our second story today, this one.
- 9:29This one covers a very big buzzword in the AI industry at
- 9:33the moment, and that's Mythos, which has caused all sorts of
- 9:36eruptions for various reasons. So here in the UK, on the 29th
- 9:40of April, NHS England quietly issued an internal directive.
- 9:45All source code repositories, that is software the NHS
- 9:48developers have published openly on platforms like GitHub for
- 9:52years, must be made private by default.
- 9:56The deadline is what was what's today's date?
- 9:59What are we the? May the 11th.
- 10:02The deadliness today, yeah, how about that for topical to
- 10:05understand why that matters? When public sector organisations
- 10:08publish their code openly, other developers can scrutinise it,
- 10:11reuse it and spot vulnerabilities before bad
- 10:14actors do. The NHS has followed that
- 10:16principle for years. The trigger for the NHS's action
- 10:21was Anthropic's Mythos model, an AI system announced on 7th of
- 10:25April that can autonomously identify software
- 10:28vulnerabilities at a scale and speed that previously required
- 10:32teams of expert researchers. In seven weeks of testing, it
- 10:36found over 2000 previously unknown flaws across major
- 10:40operating systems and browsers. NHS England's guidance
- 10:43explicitly names Mythos, warning that public code be exploited by
- 10:47AI capable of large scale analysis and reasoning.
- 10:50Other UK institutions took a different approach.
- 10:53The Bank of England convened its Cross Market Operational
- 10:56Resilience Group, A body that includes the CEOs of EU KS 8th
- 10:59largest banks, the Financial Conduct Authority, Treasury and
- 11:04the National Cybersecurity Centre, and put mythos on the
- 11:08agenda. The message to City firms use AI
- 11:11defensively, fixed vulnerabilities, automate threat
- 11:15responses. The National Cybersecurity
- 11:18Centre issued a similar warning, urging organisations to clear
- 11:22technical debt and enable automatic patching.
- 11:26Critics of the NHS decision, including Terrence Eden, a
- 11:29former government adviser who helped write its open source
- 11:32policy, argue the move is disproportionate.
- 11:35The UK's own AI Security Institute found Mythos effective
- 11:39only against small, poorly defended systems, and much of
- 11:43the NHS code has been made publicly available for years.
- 11:47The thing that really struck me about this actually, was how
- 11:49reasonably relaxed about Mythos the AI Security Institute
- 11:54appeared to be, compared to the extremely alarming talk that's
- 11:59come from elsewhere about how it could possibly hack into
- 12:03national infrastructure systems. And then there was no garden
- 12:05against it. The AISI was said, as a noted in
- 12:11that story, that it was effective only against small,
- 12:13poorly defended systems, which is not a position I've heard
- 12:17taken elsewhere. Anyway, what what's what's your
- 12:20take on all this? AI has clearly increased, the
- 12:22risk surfaced dramatically and we've seen even Firefox at the
- 12:27web browser had found that using Methos they managed to find more
- 12:31vulnerabilities than I think ever they ever had before.
- 12:34The number of patches they released once they started being
- 12:36able to access Mythos was, was just phenomenal.
- 12:39In terms of the NHS, what's at risk is our health data and it's
- 12:44our personally identifiable information.
- 12:46It's where we live. And the scope for that being
- 12:50abused is such that I think that we have to be incredibly,
- 12:53incredibly careful with how we handle that data.
- 12:57And I'm glad that the NHS is taking a, you know, a robust
- 13:00approach to it. I'm not, however, sure that
- 13:03taking GitHub, you know, repos offline is necessarily that it's
- 13:09going to make that big a difference.
- 13:10Surely what we want to do is actually get people out there
- 13:13bug hunting themselves, trying to make sure that they're able
- 13:15to find bugs and so that we can launch patches.
- 13:17That occurred to me obviously, but I'm very much not from a
- 13:21development background but the the logic behind making those
- 13:25repos private. So to someone unqualified like
- 13:28me, that there is a logic to it, but equally that the situation
- 13:31names have been for years that the position was it's better to
- 13:34make these things open so experts can, as you said, find
- 13:37and patch these problems. So my question would be what
- 13:42additional security does this afford?
- 13:45And, and, and the responses on that appear to have been mixed.
- 13:48As we noted in the story, someone so Terence Eden and a
- 13:52government advisor who's involved with the open source
- 13:54policy thinks this is excessive and and potentially even
- 13:58counterproductive regarding why they might have done this.
- 14:02So I think I think we could probably agree there is a low
- 14:04level of trust in government institutions ability to
- 14:09implement successful digitisation projects.
- 14:11I think we could agree on that. Matt, would you would you say
- 14:14so? And we all know so in 2017, the
- 14:171A cry incident. So I had forgotten about it, but
- 14:20it get, but it was really quite extraordinary. 1/3 of NHS, NHS
- 14:25trusts taken down, 19,000 appointments cancelled, 92
- 14:29million in damage. And all this is to say nothing
- 14:31of the knock it would have given to the public's faith in the NHS
- 14:36or other government institutions to safely carry out these big
- 14:39digital projects. It's, it's a massive missed
- 14:42opportunity, obviously, because I believe I read some of that
- 14:45the NHS has potentially the biggest, it could be the biggest
- 14:47source of healthcare data in the world because, you know, given
- 14:51the centralised nature of the, of the institution.
- 14:54But like you said, people are extraordinarily sensitive about
- 14:57their healthcare data and would be apoplectic if it was if, if
- 15:01it was leaked on mass. So, so there is some
- 15:03understanding. So the question again boils down
- 15:05to is the response they've taken actually proportionate and will
- 15:10it deal with the problem? Now moving on to financial
- 15:13services, again, I was somewhat, I was somewhat surprised at the
- 15:18tone of what they were saying, but this is essentially lower
- 15:22risk than perhaps it had been painted elsewhere.
- 15:24Take these precautions. Now, the reason I would be a bit
- 15:29surprised about that tone from financial services, having a bit
- 15:31of a background myself, you know, when I was worked in the
- 15:33market industry and we dealt with asset managers, I don't
- 15:36think it's any great secret to say a lot of their IT systems
- 15:40were far from state-of-the-art as well.
- 15:42The the elements of them were, were really quite aged and, and
- 15:47in and in some very dire need of being modernised.
- 15:50So certainly not impervious to attack.
- 15:53And if there's an industry that would really, you know,
- 15:55potentially, as we've discussed before, there's potentially
- 15:57really where an attack and, and, and hack and, and the stealing
- 16:02of data could be really disastrous.
- 16:05It is financial services, the knock on effects it could have
- 16:07for the broader economy that the effect on people's individual
- 16:10financial situations. All that said that they seem to
- 16:13have taken a far more measured approach, which I find very
- 16:17interesting and I will. I certainly hope that proves to
- 16:20be the right course of action. Yeah.
- 16:22I think it's, I think there's a couple of things here. 1 is that
- 16:24the way that the NHS is organised versus the way that,
- 16:28you know, financial services are organised is, is probably one
- 16:31thing that that means that they're, they are rather
- 16:34different. And one of the things that I
- 16:35believe happens within the NHS or should be happening is that,
- 16:39you know, essentially our data is, you know, is can be
- 16:43transferred between any of these systems.
- 16:45So it's not sort of its own bucket.
- 16:47So yeah, you know, there are different data silos, but
- 16:49essentially the data can be, you know, accessed pretty much from
- 16:52where, from wherever you need to be.
- 16:54Whereas within financial services, I think it's more more
- 16:57far more siloed. And of course, financial
- 16:59services has, you know, has far more money in it and also
- 17:04certainly for IT. And it's, it's had more, you
- 17:07know, dedicated stress testing of those systems.
- 17:10Whether the NHS goes under that same rigour of stress testing,
- 17:15we don't know because it probably happens behind closed
- 17:17doors. So one's a, you know, public
- 17:20service provider. The other is a series of
- 17:23institutions where, you know, if an institute gets hacked, it can
- 17:28lead to people moving to other organisations.
- 17:30So the, the incentives for maintaining strong security are
- 17:35quite different there. The NHS does have the biggest
- 17:39health information in one place around the world because we've
- 17:43been keeping records for longer than anywhere else because of
- 17:46the NHS. And it's all been, it's always
- 17:49been in this single repository, if you will, started out on
- 17:52patient cards and it's moved into all of those being
- 17:55digitised and it's a huge, huge valuable asset for the country.
- 17:59And we know that governments are trying to actually monetise that
- 18:05by selling our data to AI companies to train and by trying
- 18:09to sell that data to then help our health get better.
- 18:12So there really is an interesting, you know, sort of
- 18:15cyclical kind of process here. So I I find it quite.
- 18:19I find it quite interesting that that the difference in approach
- 18:24here. Yeah, likewise.
- 18:25And that's an excellent point about the different structures
- 18:27of healthcare, financial services as heavily regulated as
- 18:30financial services is, yes, more siloed, more, more funding
- 18:34behind it. So anyway, we we shall see how
- 18:37that one plays out. And moving on to our final story
- 18:40for the day about London's AI jobs reckoning Might over to
- 18:45you. Yeah.
- 18:46So sorry to finish on a bit of a downer, but London is apparently
- 18:50facing it's it's most detailed reckoning yet with AIS impact on
- 18:55work. A 71 page Greater London
- 18:59Authority port were commissioned from economist Jeff Dwan.
- 19:03O'Reilly has put hard numbers behind what until now has been
- 19:07largely political rat. And according to the report, 46%
- 19:10of London's workers, or 2.4 million people, are in roles
- 19:15where AI could automate some of their tasks.
- 19:18That figure sits significantly above the UK national average of
- 19:2338%. Around 1,000,000 of those jobs
- 19:26are classified as highly or significantly exposed to
- 19:30generative AI. Specifically, Mayor Sadiq Khan
- 19:33used a speech at the Bloomberg City Labs Summit in Madrid to
- 19:36announce a New London AI and Jobs Task Force chaired by
- 19:39Baroness Martha Lane Fox. Lane Fox
- 19:43cofoundedlastminute.com, Help Create Gov UK and has spent
- 19:47years advocating for ethical and inclusive approaches to
- 19:51technology. The task force will investigate
- 19:53how AI is reshaping work across London, identify near term risks
- 19:58and advise on practical interventions.
- 20:00With findings expected in the summer, buried in the report is
- 20:04a finding that deserves more attention than it is received.
- 20:0660% of the roles flagged as exposed to AI task automation in
- 20:11London are held by women, despite women making up just 45%
- 20:14of the overall workforce. Administrative and customer
- 20:18service occupations carry the highest exposure.
- 20:21Entry level positions are identified as first to
- 20:24disappear, which matters not just for the people in those
- 20:27roles, but for everyone who would have used them as a route
- 20:30into higher skilled work. Khan also cited a figure that 7%
- 20:34of large UK businesses have already cut staff due to AI.
- 20:38This is auk national figure, not a London specific 1, though
- 20:42large business concentration in the capital means a
- 20:45disproportionate share of those decisions are likely landing
- 20:49here. Over to you Nils.
- 20:50What do you make of? This some interesting stats in
- 20:52there. Certainly I think the higher
- 20:56percentage of London's workers potentially being disrupted is
- 20:59to be expected. I think London's economy is very
- 21:02heavily based on knowledge work, very large financial services
- 21:07sector where the influence of AI is quite apparent and already
- 21:11happening. The figure about women being
- 21:15more at risk is not something I had consciously thought before,
- 21:19but given where this report believes the, you know, the
- 21:24biggest vulnerabilities to being automated sit, it does make
- 21:28sense. It's I don't think there's
- 21:30anything hugely surprising. I mean, you could say that the
- 21:33figure about women being in a slightly worse position was a
- 21:37was a bit of an eye opener. I hadn't, I hadn't seen it
- 21:38phrased in those terms before. And that is obviously a concern.
- 21:41Nothing massively surprising. It's good to see that this issue
- 21:46is being taken increasingly seriously.
- 21:48But unlike what we discussed last week when Rishi Sunak,
- 21:52whether you agree with them or not, was actually putting
- 21:55forward proposals, this does not have them yet.
- 21:57And it, you know, it said it will report in the summer, which
- 22:00is fine. But you'd hope certainly by that
- 22:01point they produce real concrete proposals for how to deal with
- 22:05this. And I think it would be great to
- 22:08see some really inventive thinking on this and
- 22:09acknowledgement that this is at least potentially a massive
- 22:14disruptor. And I've actually read, I think
- 22:17you and I are broadly in agreement on this in thinking
- 22:20that at least in the medium term, it is going to be
- 22:22massively disruptive to the jobs market.
- 22:25I have read some pieces recently which have questioned that it's
- 22:27quite credible pieces. And I hope there's more because
- 22:30you know, I would, I would really love to be wrong.
- 22:33I would, I would hope that the transition to a more AI powered
- 22:37economy is not going to leave us with high levels of structural
- 22:41unemployment. So yes, I would just say good to
- 22:45see this being talked to at this level.
- 22:47Figures are kind of what you'd expect.
- 22:49The proposals really have to have meat on.
- 22:51Them I agree. I think one thing to caution
- 22:52here is that this is talking about roles where tasks within
- 22:57those roles face disruption from AI and automation, rather than
- 23:02the roles themselves being instantly at risk.
- 23:06But obviously if 20% or 30% of a role can be automated, then all
- 23:10of a sudden do we need a smaller number of people in those roles?
- 23:14The answer could be yes. I think there's one saving grace
- 23:17here potentially is that we look to potentially be coming to the
- 23:22end of the cheap token. So the costs for actually
- 23:26running AI may be starting to increase a little bit.
- 23:30And so that then may start to put things back into the play,
- 23:34back into the humans hands a little bit in terms of, you
- 23:37know, where, where does the economics make sense?
- 23:40But we do have to be careful that I think that there is going
- 23:42to be a a lot of disruption. And we really need to be
- 23:46planning very, very carefully as to how do we how do we train the
- 23:49workforce? How do we make sure that we're
- 23:52not overly taxing the wrong sorts of roles where, you know,
- 23:56we're actually so that the economics makes sense to to keep
- 23:59people employed. You could see a lot of
- 24:00disruption and much of the Western world really being hit
- 24:04by AI stealing jobs. Yeah.
- 24:05I mean, at the very least, a wholesale reform of the tax
- 24:09system seems like it it it will be required to funnel some of
- 24:14the enormous profits that AI companies tend to make from
- 24:17taking an ever larger, larger chunk of the service industry.
- 24:21For example, there's a lot said about training and I'm sure that
- 24:23will be a central feature of the proposals that are released in
- 24:28the summer. Something else we touched in
- 24:30regarding training is that it's all very well to talk about
- 24:34reskilling as if it's a kind of concrete or discrete thing that
- 24:38can just be done and you are ready to do a job with AI in the
- 24:42title. The difficulty is, I mean, AI
- 24:45stands to be so transformative that it basically undercuts the
- 24:49logic behind a whole swathe of jobs.
- 24:51I mean, a lot of jobs are essentially, yes, they're jobs
- 24:54comprise tasks, but a lot of jobs are dealing with
- 24:59bottlenecks in the system, just specifically managing A
- 25:02bottleneck and making something, facilitating it or making it
- 25:04happen faster, bottlenecks that could largely disappear when AI
- 25:09becomes more broadly adopted at an enterprise level.
- 25:12So I mean, I would, I would urge, because I'm sure they're
- 25:16listening, I would urge the people working on this report to
- 25:19try to really think boldly and creatively.
- 25:23As I always say, no one knows ultimately nobody knows what's
- 25:27going to happen. No, no one has a crystal ball.
- 25:29The whole idea that the more apocalyptic predictions are in
- 25:33the jobs market, which, you know, typically emanate from
- 25:36people that are prominent in the AI industry or AI companies that
- 25:39there's a degree of self-interest in there.
- 25:41They need to, they need to keep the narrative quite alarmist in
- 25:47order to generate the huge amounts of investment they need.
- 25:50All of that I think is legitimate and and heaven knows,
- 25:53perhaps it will pan out more like previous technological
- 25:56changes. But for me, the evidence is
- 25:59overwhelmingly that it's this one is different and in in order
- 26:03it will throw up challenges of a entirely different nature.
- 26:06So yes, I would urge this Commission to be really quite
- 26:10bold and inventive with the proposals because some last
- 26:14points say, for all all the UK's difficulties and problems, we
- 26:18are in some respects well placed to take advantage of this and I
- 26:21think that's something else that we always have to keep in mind.
- 26:25Absolutely. London already has the UK size,
- 26:28you know, job creation, right? And so if we can position people
- 26:34that, whose roles are likely to be, or tasks in those roles are
- 26:37likely to be affected in a way that can then take, take up
- 26:41roles in those new, those new organisations within those, you
- 26:43know, those new places, then maybe we can absorb some of that
- 26:46impact. But I think that that is a big
- 26:48ask. So we really do need to have a,
- 26:52a really strong plan as to how do we, you know, how do we, you
- 26:55know, how do we plan ahead for a future that is looking a little
- 26:58bit opaque and probably has more uncertainty in it than than ever
- 27:01before. Don't ask me now, trying to tell
- 27:07the future is becoming an increasingly common part of what
- 27:10we do, I suppose. OK, well, on that note, thanks
- 27:13as always for talking through this week.
- 27:15This week's issues, we are likely to have a bit of a change
- 27:18in format next week. We are going to be interviewing
- 27:23A chap, Keith Swain, who runs something very relevant to what
- 27:27we've just been discussing, which is he runs a company that
- 27:31provides apprenticeship programmes for AI leadership.
- 27:34So this will touch on all sorts of issues around AI literacy,
- 27:37how leaders should be using AI themselves, how they can frame
- 27:41this for for their workforces and how to plan as a business
- 27:46leader. Just how on earth we were just
- 27:48moaning about trying to tell the future.
- 27:50How on earth did you do it for in a business environment that
- 27:53is changing like never before. So yeah, should be really
- 27:56fascinating conversation. But in the meantime, thanks as
- 27:59always, Matt, for talking through the issues.
- 28:02Thank you, Neil. Look forward to speaking next
- 28:04week. All right.
- 28:06Bye for now AI in the. UK AI in the UK And that's it
- 28:18for this week's episode of AI in the UK.
- 28:23If you found it useful, you can subscribe wherever you get your
- 28:26podcasts and we'd really appreciate a great rating or
- 28:29review. It genuinely helps others find
- 28:32the show. You can also follow Neil's
- 28:35Footman and Matt Wilkinson on LinkedIn for links, notes and
- 28:39occasional extra content. Thanks for listening and we'll
- 28:43be back next week. Helping you make sense of AI as
- 28:47it lands here in the UK.