Latest / AI in the UK / UK's IP Battles Heat Up; OpenAI Opens Key Research Lab in London; UK AI Biology Cluster Takes Shape
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 on this week's
- 0:56edition of AI in the UK, the UK legislative battle over using IP
- 1:02to train AI models is heating up.
- 1:05But with frontier model research depending so heavily on data
- 1:08mining, how much can the UK really defend copyright?
- 1:13Open AI has chosen London to be it's biggest research centre
- 1:17outside the US. Is this a big leap for British
- 1:21AI or is it just window dressing?
- 1:24And three announcements suggest Britain is building its own AI
- 1:27powered biology cluster. Can the UK turn its academic
- 1:31brilliance into major commercial success?
- 1:35So to kick us off with the first story, Matt, over to you.
- 1:39Thank you, Niels. And this is a story about the
- 1:42battle for Britain's creative intellectual property.
- 1:46With the UK creative industries mounting A coordinated copyright
- 1:50defence against AI training, three separate British creative
- 1:56sectors are pressing for formal frameworks governing AI training
- 2:00data. The Publishers Association has
- 2:03called on ministers to reject the government's proposed text
- 2:06and data mining exemption, which would let AI companies train on
- 2:10content unless rights holders actively opt out, arguing the
- 2:16book trade should shape AI development through negotiated
- 2:19commercial deals, not blanket permissions.
- 2:24The BBC, the Financial Times, Guardian, Telegraph and Sky News
- 2:28have now jointly launched what they've called SPUR Standards
- 2:33for Publisher Usage Rights, a coalition modelled explicitly on
- 2:37a NATO for News concept seeking industry wide technical
- 2:41standards and licencing frameworks.
- 2:43It won't set fixed prices or act as a collective licencing body,
- 2:48with individual publishers able to create their own deals with
- 2:53AI companies, but it will enable them to have some control and
- 2:59some technical standards in place.
- 3:01And then the third set up is HEPI released a framework urging
- 3:08universities to develop distinctly human capabilities
- 3:11alongside AI. The universities sit in a
- 3:15different position entirely as both content producers and
- 3:20increasingly as AI research partners.
- 3:23The three initiatives proposed genuinely different models,
- 3:27bilateral commercial deals, shared industry standards and
- 3:31capability development. What unites them is a challenge
- 3:35to the assumption that AI developers can freely chain on
- 3:39British cultural and intellectual assets.
- 3:42The government's proposed TDM exemption sits directly in
- 3:46conflict with the publishers position, creating a legislative
- 3:50choice that EUUS and Japan have each resolved differently.
- 3:55So, Nils, there's a lot to unpack here, but where do you
- 3:57sit on this opt in versus opt out question?
- 4:03You know, I thought I had a relatively clearview of this on
- 4:08on what I thought of it. I thought opt out, meaning by
- 4:12default all the data is available for the AI companies
- 4:17to use unless you specifically opt out was a necessary evil.
- 4:21Essentially the the the line the companies trot out.
- 4:26And I, I, I thought and still think to some degree there is,
- 4:29there is some validity to it is that in order to stay
- 4:32competitive, particularly with the, the fast improving Chinese
- 4:37models, they need access to the, to those huge reams of data.
- 4:41And having an opt in system would simply be too bureaucratic
- 4:45and costly to achieve. Now, I say this as someone who's
- 4:50from, from a somewhat creative background myself.
- 4:52I used to be a writer still, well, I still do, you know, I
- 4:55still still do write, but I used to do it professionally.
- 4:58And I think it's, it's clearly massively unfair that these huge
- 5:02companies essentially stole all that data and without any
- 5:06recompense. My feeling was they should be
- 5:08able to wait, work out some way to to compensate them.
- 5:12But looking into this story, the, the, the whole thing about
- 5:15the opt in is, is far more complicated and, and
- 5:20multifaceted than I thought. So for example, if if the UK was
- 5:27to adopt an opt out position, it basically means that it's
- 5:31throwing in it's lot with the the pre training stage of
- 5:35frontier models. This these massive training
- 5:39programmes where they suck in as much data as possible to create
- 5:42these huge models. Now it's not unreasonable to
- 5:46ask, is this where the UK should be focusing?
- 5:50We don't have any native UK frontier models.
- 5:55The closest thing we have is DeepMind, which is owned by
- 5:59Google, but a lot of EU KS real value is on its creative
- 6:04industries and creative content which is not used or or is or
- 6:09can be used more productively in later training stages in RAG,
- 6:15for example, that you know the the retrieval augmented
- 6:17generation where it can offer real value.
- 6:20So perhaps that's where the UK should be focusing.
- 6:23So I, I'm, I'm a bit torn. I think probably I still fall
- 6:27somewhere on opt out. But how about you?
- 6:29What what? What's your view?
- 6:31I'm torn as well on on the exact mechanism, but who's going to go
- 6:36to the effort of opting in? A lot of people aren't going to
- 6:40do that. And I think the real risks
- 6:42involved here are ones that we're not going to see straight
- 6:45away. But there are risks that down
- 6:49the line, what we'll see is as the models progress, the UK's
- 6:52cultural heritage will be will become less and less represented
- 6:59within the training data. And all of a sudden our
- 7:02influence within the tools that are going to be influencing how
- 7:07the world is shaped over the next decade or more is going to
- 7:11disappear. So our influence by default will
- 7:14decrease. And I think that's a real risk
- 7:16that these conversations aren't really paying that much
- 7:20attention to now. So can I just start the first
- 7:23thing, just just to clarify something you said Matt.
- 7:25So you're saying if we adopt an opt in, if we adopted an opt in
- 7:30model the risk would be the big companies would find it
- 7:33uneconomical to do any pre training?
- 7:36With this the vast streams of UK data and as a result the UK
- 7:41component would be lost from the pre training element of very big
- 7:45models which would be disadvantageous to the UK in the
- 7:49long run. Is this?
- 7:50Is this your argument? I believe so.
- 7:52And it's not just necessarily disadvantageous to the UK, but
- 7:54it's the, it's the UK belief system, which, you know, we can
- 7:59see geopolitics being incredibly fragmented these days.
- 8:05And I think that one of the, the challenges is that we really
- 8:08need our voice to be, to be heard.
- 8:10We want our values to at least be represented.
- 8:14And I don't know that at the moment they are represented as,
- 8:18as well as they could be in the on the world stage.
- 8:22You know, successive geopolitical shifts have have
- 8:26maybe reduced that. And I'm, I'm concerned that what
- 8:32I believe is kind of still a belief in the, the rules based
- 8:36order, the, you know, international law and, and a
- 8:39whole sort of history over the last, you know, certainly since
- 8:45the, since the end of the Second World War history of, of, of
- 8:48international collaboration and, you know, treating people
- 8:55fairly, it, it could disappear and could be under represented.
- 8:58And that's something I'm concerned about, that is.
- 9:01A really interesting point. I, I just wonder, have you seen
- 9:06evidence of this? Have you, have you read anything
- 9:07about this? The, the degree to which, and I
- 9:09know that there is a lot because we're in English, because we
- 9:13have a kind of structural advantage of the amount of
- 9:16British content that will have been used to train some of the
- 9:19big models. I mean, have you ever seen any
- 9:22evidence of the the degree to which a single country's
- 9:24contribution influences the kind of cultural tone of of the
- 9:30output of these models? So I think that the, well, yeah,
- 9:34there's, there's some very good examples of this in imagery.
- 9:40There were, I think it was TechCrunch ran a, ran a series
- 9:44where essentially they, they went through and they asked,
- 9:48they asked one of the diffusion models, I believe it was mid,
- 9:52mid journey to, to create a series of Barbie images of
- 9:55Barbie dolls across, across, you know, from different countries.
- 10:00And, you know, as they went around the world, there were
- 10:04predominant pictures of, you know, the stereotypes from each
- 10:07of those countries. So there were, you know, when
- 10:10they went to Germany, it was a blonde haired lady who was
- 10:13wearing a black jacket with sort of some silver signals on the
- 10:15things that sort of went back to maybe some Second World War kind
- 10:19of invocations, if you will. There were pictures from, you
- 10:24know, all around the world. Very, very archetypal imageries
- 10:28of those countries. The sort of the concern is, is
- 10:31that you go then go to somewhere like Sudan where most of the
- 10:35images that have been taken have been taken in recent years and
- 10:40during times of incredible strife and upset and civil war.
- 10:46And so those pictures showed, you know, a Sudanese lady
- 10:50wearing Sudanese traditional garb holding an AK47.
- 10:57That's how the training data impacts the output.
- 11:01Now, you know, real life human feedback, RLHF can influence
- 11:06the, you know, the quality of those outputs.
- 11:08But we don't know if, if the data isn't there in the
- 11:12beginning, it's not even going to come up to be able to then
- 11:15get to that post sort of training piece.
- 11:18I I, I read something interesting the other day that a
- 11:21lot of the real life human feedback came from Nigeria.
- 11:26And that is the reason why the word delve shows up so much in
- 11:31AI answers. Because it is a word that is
- 11:34used far more commonly in Nigeria than it is in the UK or
- 11:39in the US. And so the the location of where
- 11:42the work is done and the location of and and where the
- 11:46inputs come from directly influence the outputs.
- 11:50OK, good point. A couple of couple of things.
- 11:53I'd come back to you on that, though I wonder the degree to
- 11:55which with future training models, it's difficult to think
- 12:00that using future training models they would explicitly
- 12:03extract existing elements of British input, British content
- 12:08and culture, if only because it's technically very difficult.
- 12:12So even if even if an OPT, an opt in model in future made it
- 12:19uneconomical for the big AI training companies to use
- 12:22British data in future training models, it would not be excluded
- 12:27you would assume from what's already been there.
- 12:29The second point I'd make is, like I said, you make a really
- 12:34interesting point, but I think that the way British AI
- 12:38companies have thrived, the ones that are doing particularly well
- 12:41are not frontier models. They, they are built on frontier
- 12:44models and they're, they're using the, they're doing the
- 12:47later stages of training. They're building their products
- 12:49on top of models, which is not something for which this is a
- 12:53relevant argument because the models already exist.
- 12:56So the, so the likes of 11 Labs say, and what's the video one
- 13:02that we talked about Synthesia. So, so companies like that and,
- 13:07and others and, and the one we talked about last week, David
- 13:09Silva, he wants to build a frontier model, but not based on
- 13:14scaling AI, scaling big data sets.
- 13:18He wants to do it based more on reinforcement learning.
- 13:21So you wonder whether kind of a the the degree of relevance to
- 13:27British AI companies now and how relevant it will be in future
- 13:30given that the the growing move around the growing move away
- 13:34from scaling as the solution to reaching AGI I mean.
- 13:39You raised some interesting points, but with the vast
- 13:46majority of content having already got into these models
- 13:50anyway, if we were to, to now go to an opt in only model, so I
- 13:57have to say yes, I would, I would allow you to use my model
- 14:00and you know what, use my data and my, you know, my, my
- 14:02intellectual property. If that only happens from
- 14:07anything from now onwards, then 95% of all British content
- 14:14doesn't, doesn't get covered anyway.
- 14:17So it has to go back historically.
- 14:19Otherwise it's a pointless law, which I think is why it's on to
- 14:23a really it's, you know, it's the losing vote really because
- 14:27of just the complexity and the fact that how do you, I don't
- 14:33even know how you how you go into that.
- 14:36I mean, is it that if you've got a, you know, you're a British
- 14:38person with British IP, but you've, but you're the company
- 14:42that you, that you wrote your, you know, published your book
- 14:44through was a British company that again then gets bought by
- 14:47an American company. Where does that IP sit?
- 14:49That the, the it's such a complex scenario that I just
- 14:54can't see it working, which is sad because I completely believe
- 14:59in the right of, you know, of the creators to, you know, to be
- 15:04rewarded here. But it is, I can't see that opt
- 15:09in only model working. I think we're going to have to,
- 15:12we're going to have to wait and see what happens.
- 15:14But I'm really interested to see how this falls out.
- 15:17I think this is a really big gnarly issue that that could
- 15:21take weeks of us discussing to to and and probably never get to
- 15:26a probably never get to a decision.
- 15:28We haven't reached a conclusion it it is considerably more
- 15:30complex than I than I had initially considered there.
- 15:34It's a fascinating issue and it's one that's going to run and
- 15:36run and one I'm sure we will return to.
- 15:40But we have other things to discuss today.
- 15:44Moving on to our second story is about London and solidifying its
- 15:49position as an AI capital within Europe through some news from
- 15:54Open AI and some major funding as well.
- 15:58So 2 significant developments this week reinforced the UK's
- 16:01growing position as a centre for serious AI research.
- 16:05Open AI has designated London it's largest international
- 16:08research hub outside San Francisco, a signal that the
- 16:12city is consolidating it's place as Europe's leading hub for
- 16:15frontier model work. Ironic given what we were just
- 16:18talking about. And at the national level, UK
- 16:21Research and Innovation has committed 1.6 billion pounds to
- 16:24AI over the next four years, it's single largest investment
- 16:28area. Open AI has designated London as
- 16:31its largest international research hub outside San
- 16:33Francisco, significantly expanding on an operation it
- 16:37first established in the capital in 2023.
- 16:40Researchers there will take ownership of key components of
- 16:43Open AI frontier model development, including work on
- 16:46safety, alignment and reliability, placing London at
- 16:49the centre of the company's most consequential research.
- 16:53Separately, UK Research and Innovation, EU KS main research
- 16:57funding body, has unveiled its first dedicated AI strategy
- 17:01backed by 1.6 billion targeted directly at the AI sector
- 17:05between 2026 and 2030. The strategy covers 6 priority
- 17:10areas including advancing AI technology, transforming
- 17:14scientific research through AI and building world class
- 17:16infrastructure. London's position in global
- 17:20rankings reflects genuine strength but also genuine
- 17:23competition. The 2025 Global Innovation Hub
- 17:27Index places London 5th overall in research and innovation
- 17:30capacity behind San Francisco, New York, Beijing and the
- 17:34Guangdong, Hong Kong, Macau region.
- 17:38So a big question here, Matt, around all of this.
- 17:42Just how significant is it? So I will kick off, I'll say the
- 17:48news about Open AI is encouraging.
- 17:52So it's not just opening an office.
- 17:55London researchers are already contributing to specific
- 17:58products, including codecs and GPT 5.2, and will own core
- 18:04components of frontier model development, including data
- 18:07evaluation, alignment and reliability work.
- 18:11That is a meaningful transfer of research responsibility, I would
- 18:15say. And it's not not not merely a
- 18:18satellite operation or PR exercise.
- 18:22Yeah, it's an interesting 1, isn't it?
- 18:24Because while that sounds great, Open AI hasn't actually
- 18:28disclosed any headcount targets or real investment figures.
- 18:32You know, roughly 30 researchers.
- 18:34Currently, the largest international hub label is being
- 18:38applied to an operation smaller than your typical university AI
- 18:42lab. So I think until we see the
- 18:44numbers, this could be just more of a designation rather than
- 18:48demonstrating commitment. I hope I'm wrong with that.
- 18:52I, I really do. But my fear here is that this is
- 18:57one of those nice PR spins that Sam Altman is so good at, at
- 19:01doing. And, you know, really they're to
- 19:04sort of stroke UK government ministers kind of give them
- 19:08those strokes to sort of make them feel good about themselves,
- 19:10make them feel good about the, the UK's position in the AI
- 19:13race. When really it's, you know,
- 19:17we're going to keep nearly everything in the US and, and
- 19:20we're going to have a little satellite office in the UK and
- 19:23we'll give it some of those. We'll give it some of those
- 19:26tasks where actually we don't need lots and lots of
- 19:30infrastructure nearby to do it. Because, you know, UKUK
- 19:35headcount is probably cheaper. Oh, it's definitely cheaper than
- 19:40Silicon Valley headcount. Well, yes, it is always easy to
- 19:44be cynical, Matt, particularly given you're saying he's got a,
- 19:50you know, a, a solid touch with the PR.
- 19:52But his recent, the the opening eyes, recent travails with the
- 19:56Department of War might suggest otherwise.
- 19:58That was not not their finest PR moment.
- 20:00However, there is there are there are other reasons to
- 20:04suppose that this is more than just more than just APR exercise
- 20:09or or some kind of outpost. As we keep on saying, deep deep
- 20:14mind, yes, it is no longer British owned, but being based
- 20:18here has been enormously important for building an AIAI
- 20:23ecosystem and that that is a sighted reason for why Open AI
- 20:29has chosen to expand its research facilities here.
- 20:33DeepMind has in the region of 2000 staff or 2000 just
- 20:38researchers apparently primarily based in London, which gives
- 20:43London a pool of frontier AI talent that most other
- 20:47competitor cities. Obviously San Francisco very
- 20:50much aside, we're talking in in the kind of second tier.
- 20:54Most cities can't match that, including rivals like Paris,
- 20:59which have their own strengths. But this, this in particular is
- 21:01a seems to me to a particular strength of London's position
- 21:05and exactly why open AI might might choose to, to expand here
- 21:11and focus here along with the universities, of course.
- 21:13So I think there could be more to it than than you know a a a
- 21:18less flattering reading might suggest.
- 21:20I think you're right and those are all good reasons, although I
- 21:22mean it, it is worth just noting that Paris is closing the gap
- 21:26and in some research metrics such as patents, university spin
- 21:29out founders already leads London.
- 21:32You know, the assumption that London's European primacy is
- 21:35secure, you know, may not hold until 2030.
- 21:38So now while I hope, you know, while I'm hope that all these
- 21:42things are wrong and that we really are building something,
- 21:46you know, something strong and foundational, if you'll excuse
- 21:49the pun. I'm I'm a little bit sceptical
- 21:53still. OK, no, fair enough.
- 21:56I mean, what I would certainly hope would happen is, is
- 21:59something along the lines of some of the influence DeepMind
- 22:03has had. I mean, the ideal scenario is
- 22:07open AI does significantly and meaningfully expand its research
- 22:10operations here and attracts other people in, brings people
- 22:13in from universities and, and other people from overseas.
- 22:17And this in turn further cements a kind of AI ecosystem in the
- 22:22UK. Some of these people leave, they
- 22:24start their own companies and it grows from there.
- 22:26So yeah, commonly hopelessly naive, but I, I, I'm hopeful for
- 22:31this. I think this could be a really
- 22:32positive move for the AI industry in the UK.
- 22:37I'm, I'm hopeful too, but I think that we might have better
- 22:43bets in another sector. And that sort of moves us nicely
- 22:45into our sort of third story, which is Britain's bio AI boom.
- 22:51There've been 3 funding deals in the last week and that sort of
- 22:54signals a shift from policy to deployment.
- 22:58We've had recent weeks from, you know, a number of different
- 23:02companies. Cardiff based anti verse closed
- 23:06it at 9.3 million Series A to scale.
- 23:08It's AI driven antibody design platform targeting so-called
- 23:13undruggable proteins, GPCRS and ion channels where conventional
- 23:18discovery has repeatedly failed. The four month design timeline
- 23:23widely quoted in coverage is an investor's line, not a company
- 23:27claim. But it is still incredibly
- 23:30exciting, but it doesn't cover the full development cycle from,
- 23:34you know, thinking about a target through to getting to,
- 23:37you know, getting to, to the clinic.
- 23:39It's, it's more about sort of actually getting the the design
- 23:41done. Following that, Cambridge based
- 23:46Bind Bridge raised 3.8 million in seed funding for its Bridge
- 23:49platform, applying AI to design molecular glues for targeted
- 23:54protein degradation in agriculture, a drug discovery
- 23:58mechanism being adapted for herbicides and crop protection.
- 24:02The company was founded in March 2025 and you know, it's, it's
- 24:06still scaling up. Earlier this month, Sable Bio
- 24:11raised 3.8 million to scale it's AI drug target safety platform
- 24:17and was that was a company founded by former benevolent AI
- 24:21scientists and it already holds commercial agreements for three
- 24:26top ten global pharma firms. And another one in the past week
- 24:32was that Google DeepMind and the Welcome Sanger Institute have
- 24:35announced A postdoctoral fellowship in AI and genomics
- 24:40applications open late spring 2026.
- 24:43And so, so, so obviously no fellow has yet been appointed,
- 24:47but I think that's going to be a really important, you know,
- 24:49important, important piece of research.
- 24:53The common thread is, you know, the alpha fold breakthroughs in
- 24:57protein structure prediction, which you know, have made
- 25:00computational foundations to molecular design viable at small
- 25:03team scale. Now that really does look to
- 25:05sort of look at sort of, you know, unblocking commercial
- 25:08bottlenecks. You know, one of the the big
- 25:11challenges is of course is that regulatory approval timelines in
- 25:15pharma and aggregate, you know, chemicals, those currently
- 25:18remain unchanged. And so we might be looking at
- 25:22companies being able to to get things into the clinic quicker
- 25:26and maybe more reliably, but then it's still not necessarily
- 25:30going to shift the years of inhuman research that are needed
- 25:34to really prove that a drug is safe and effective in, you know,
- 25:38in real life. So what do you make of this
- 25:41Neil's? Just just as an aside,
- 25:45interesting to know alpha folds in been mentioned there.
- 25:50All roads really do lead back to deep mind don't they?
- 25:53AA fold is is also a product of deep minds research and it was a
- 25:59massive breakthrough in protein structure reading which is which
- 26:03is thought to be a next impossible task.
- 26:05But anyway, anyway, as usual, I will not I will not suppose or
- 26:12will not presume to understand all the life the scientific
- 26:16issues in great depth. That's more your department one.
- 26:19So allow me to adopt the cynical, the somewhat sceptical
- 26:24side on this one, because one thing I did notice was it is it
- 26:30is a habit of journalism in general to try and identify
- 26:35anything as a trend. I remember I used to work as a
- 26:37journalist and there was, it was a bit of a, it was a bit of a,
- 26:40an ongoing joke about for people desperate to find an angle for a
- 26:44story, they would try and just identify anything that happened
- 26:48twice or three times and call it a trend.
- 26:51So is this really a cluster? So these are four companies in
- 26:54different cities, different regulatory regimes, different
- 26:57investor bases. We've got Cardiff, Cambridge,
- 27:00London. Is that really a cluster or is
- 27:05it, as we were discussing earlier, more of APR spin?
- 27:10So I think that you're right that sort of Cardiff might be an
- 27:13outlier in where it sits. But we have long had a, what's
- 27:19been known as the Golden Triangle, which sort of looks at
- 27:22London, Cambridge and Oxford as being a real, you know, biotech
- 27:27cluster, sort of nascent in its form, Maybe nowhere, you know,
- 27:31not as powerful as those in Silicon Valley and sort of in
- 27:34the Boston area, but still a, a real global powerhouse with some
- 27:38of the world's best researchers and, and, and, and a lot of
- 27:43pharmaceutical companies, many, you know, many of which have
- 27:46moved their headquarters to their cities.
- 27:48In fact, AstraZeneca moved their headquarters to Cambridge not
- 27:52too, no, not too long ago. So I think it's, I do think it's
- 27:56important. I think what's really exciting
- 27:59about this is that this, as you know, as you mentioned, alpha
- 28:02fog coming out of deep mind. I think it's the convergence of
- 28:05those two centres. So you on one hand, you've got,
- 28:11you know, an area that does have an awful lot of deep science
- 28:15knowledge and you've got an area that has a lot of deep AI
- 28:20knowledge. And those two don't overlap that
- 28:22frequently. And I think that's why this is
- 28:26this is so exciting and it it could well be a cluster forming.
- 28:31OK. But the the news of the
- 28:33individual companies themselves is, I mean, it's noteworthy and
- 28:38we always have to maintain a, a, you know, a sense of
- 28:41perspective. And it seems from from what's
- 28:44being mentioned, Binebridge hasn't run a lab experiment yet.
- 28:48Sables revenues are fairly minimal and the fact the
- 28:53fellowship hasn't yet started. So what's actually new here?
- 29:00Well, that's the exciting thing, isn't it?
- 29:02If you were to ask me what was happening in AI 3-4 years ago,
- 29:05we'd have probably not, we wouldn't have been having this
- 29:07podcast. So what's new is everything.
- 29:09And I think it's the convergence of, you know, the, the talent
- 29:15and the need. So you've got both the, the AI
- 29:18talent in the area, you've got the, the, the, the deep
- 29:22expertise in life sciences and then, you know, and the
- 29:26companies that have the need for this as well.
- 29:28And so I think that really the, that what's new is that of
- 29:31course investors are coming in and seeing the opportunities to
- 29:35invest in these companies. And many of them are, you know,
- 29:38start-ups, spin outs. And so of course they're at the
- 29:42earliest stages of, you know, of their development.
- 29:45You know, we didn't question whether David Silver's startup
- 29:49had had, you know, really had many revenues yet.
- 29:52And so I think it does look to, yeah, it looks promising from my
- 29:57perspective. Whether it will, whether it will
- 29:59go anywhere, we don't know. OK.
- 30:03I mean, fair enough. I think this is a field you've
- 30:05followed closely for many years. So I think your perspective
- 30:08means something. I would just say that David
- 30:11silver starts no, no revenues, but a billion in seed funding.
- 30:14So I think that that that does put in a different category
- 30:17somewhat. So would you sum out then as
- 30:22saying something like it's, it's what it's really about?
- 30:25Is London's AI growth landing on top of an already exceptional
- 30:29life science base elsewhere in the country, something like
- 30:31that? Well, I wouldn't just say
- 30:33elsewhere, I'd say in, in within London and the, the surrounds
- 30:38because you know, you can get a train to Oxford or Cambridge
- 30:40within an hour from, from London.
- 30:43So I think it's it that's the, that's the, that's the core
- 30:45piece and that's the core bit of the cluster.
- 30:48And the $64 billion question, as it probably would be nowadays,
- 30:54how about commercialisation? Where, where do we sit with
- 30:57that? Are we, are these companies any,
- 31:00have have they learned the lessons of the past?
- 31:01Are they are they better positioned to do that?
- 31:06Well, that's, that is a, you know, that's the, that's the
- 31:08million, billion dollar question, right.
- 31:11We'll we'll have to wait and see, but I think we can start
- 31:14with with probably what's actually true.
- 31:18We know that investments coming in.
- 31:21We know that the, the technique, you know, the technologies are
- 31:26there and the industry is ripe for disruption and evolution.
- 31:32And I think that that's a really exciting thing that that we we
- 31:36have people in the country able to start companies that are
- 31:41receiving funding. OK, well, Amen to that.
- 31:47I'm I'm all for that. Just as a last point around
- 31:51this, So this, the, the AI is going to transform drug
- 31:59discovery and, and, and treatment and so on is one of
- 32:02the most persistent narratives around AI and why it's
- 32:06potentially such an exciting technology.
- 32:09As a, as a non specialist in that field, I'm not sure I, I
- 32:14desperately want that to be the case.
- 32:16And I can understand why it having used using AI so much
- 32:19myself and its ability to synthesise data and discover
- 32:23patterns and spot patterns, which is one of the biggest
- 32:25challenges in scientific research.
- 32:28But it's hard not to be sceptical.
- 32:30It seems like very little has happened yet.
- 32:34Where do you stand on that? Yeah, I, I mean, it's a, it's a
- 32:39great question, but let's start with what's actually true.
- 32:42We know that 90% of all drugs that enter the clinic and enter
- 32:48sort of clinical trials fail. Every biotech pitch sites that
- 32:53as a, as real, you know, 9 out of 10 drug candidates fail.
- 32:58That's not a statistical quirk. It's the cost of an industry
- 33:01that has historically been unable to distinguish good
- 33:03candidates from bad ones early enough to stop spending on them.
- 33:08That's partly because the models, the biological models we
- 33:11use are not human models. We don't put things straight
- 33:15into human beings. Otherwise we'd have a lot of
- 33:16people dying because the drugs aren't safe or the, you know,
- 33:21and there are a lot of trials where, you know, people just,
- 33:25you know, the drugs aren't efficacious enough.
- 33:26They look good in lab results. They hit the target, but
- 33:29something happens in the body that means maybe the cascade
- 33:32doesn't actually that we're targeting in the body doesn't
- 33:36actually lead to the desired effect.
- 33:38Now, AIS 1st and most defensible contribution into that picture
- 33:43is making that failure faster and cheaper.
- 33:46So Sable bio is the clearest example of this.
- 33:49Moving toxicity assessment earlier doesn't shorten the
- 33:53regulatory clock, but you know, failing on the right grounds
- 33:57before hundreds of millions go into, you know, into a project
- 34:01is genuinely valuable. You know, three top 10 pharma
- 34:05firms are already paying for those sorts of services, and I
- 34:07can only see that growing now. Antivirus is a more ambitious
- 34:12claim. They're not saying that they
- 34:14make drug development faster. They're saying they can attempt
- 34:17targets that were previously closed.
- 34:19GPCRS and ion channels, protein families where a third of all
- 34:23FDA approved drugs work, but where almost no approved
- 34:27antibodies exist because conventional methods couldn't
- 34:30crack them. That's not a faster treadmill,
- 34:33that's a different room. And that's a chat.
- 34:34That's a real challenge. So that's a bit more
- 34:36speculative, you know, So there are two versions of what AI does
- 34:40to drug discovery. The 1st is make the same process
- 34:42more efficient, cheaper, cheaper failures, faster iteration.
- 34:46The second is unlocking a new map.
- 34:49Targets that were biologically inaccessible become druggable.
- 34:52Disease areas with no viable chemistry become viable.
- 34:55We're probably in the early innings of both simultaneously
- 34:59and the market market hasn't really decided which story it's
- 35:02pricing. Now, if the second one is true,
- 35:05you know, if alpha fold era tools genuinely expand the
- 35:08druggable target space, the long run implications for diseases
- 35:12are hard to everstate. The question is whether Britain
- 35:15converts the academic infrastructure it already has
- 35:18into commercial output that stays here.
- 35:20You know, that's the real test. And we've seen time and time
- 35:23again that these companies, they get bored up by, you know, big,
- 35:28big multinationals and, you know, they end up leaving.
- 35:33Yes, as I said, it is something we've seen time and again and
- 35:37this is a story that is going to run and run as well.
- 35:40There's just so much to so much to talk about here, but
- 35:43unfortunately not in this episode.
- 35:46We've we've definitely reached the end of our time for this
- 35:51week. So with that I will say thank
- 35:53you as everyone for your time and for chatting over these
- 35:56issues. Thank you, Nils.
- 35:58It's been there's been a lot of good news this week.
- 36:00Indeed. All right, see you next week.
- 36:13And that's it for this week's episode of AI in the UK.
- 36:18If you found it useful, you can subscribe wherever you get your
- 36:21podcasts and we'd really appreciate a great rating or
- 36:25review. It genuinely helps others find
- 36:27the show. You can also follow Neils
- 36:31Footman and Matt Wilkinson on LinkedIn for links, notes and a
- 36:35case.