Goldilocks and the Big Shiny Project: Why Government IT Programmes Fail

Episode links & transcript

Links

DrDoctor LinkedIn

Tom Whicher LinkedIn

Professor Joe McDonald LinkedIn

Professor Joe McDonald Digital Health

FHIR and Loathing in Las Vegas

The Access Group

Transcript

Tom - So, welcome to DrDoctor Will See You Now, the next in our podcast series. And I’m really, really delighted to have Professor Joe MacDonald here with us today - advisory board member of the Access Group, 30 years as a clinical psychiatrist, clinical lead for mental health on the National Programme for IT. And most importantly, you’ve just written a book, Digital Royalty. Joe, welcome. Tell us a bit about yourself.

Joe - Thanks a lot. Well, obviously, I’ll start with the book, FHIR and Loathing in Las Vegas, available on Amazon, £1.77, if you want to download it - or if you want the coffee table, full colour version, quite a lot more.

I set off as a medical director of a mental health trust, but I didn’t start there - that’s what I was about 20 years ago. My trust merged with the trust next door, as NHS organisations tend to do, and I suddenly became deputy medical director.

And this was the first time that my career had had to take a sort of step down. And I was looking for somewhere else to go and be really important, not number two. And I picked up the PMJ, and there was an advert in the paper for National Clinical Lead for IT - brackets mental health.

And I thought, well, you know, I’ve climbed the greasy pole as, you know, the trust management climbing frame. Why don’t I try and be something national?

So I went to London and had an interview to join the National Programme for IT - Tony Blair’s £12.7 billion National Programme for IT - with responsibility for the mental health arena.

And I had recently built a website. Now, not a lot of people had built a website 20 years ago. I built a one pager to rent out my house in Turkey, which I couldn’t afford, but I bought anyway. So I bought Microsoft FrontPage, which was a first WYSIWYG website building tool. And the CD-ROM, when I put it in my Amstrad, it cost £460.

But because I built a website, when I got to Westminster for the job interview, I was like, you know, wow, this guy actually built a website. You’ve got the job.

And I remember ringing my wife, standing in Whitehall, and said, I think I’ve got the job. She said, oh, shit, what are you going to do now? I said, well, I don’t know. I’ll just have to make it up as I go along. So I did.

So I was National Clinical Lead for IT within the National Programme for four years. And it was an education. It was an absolute education. I mean, none of us really knew what digital health was at that point, right?

Tom - You were making it up as you went along in the most positive way possible, right?

Joe - Yeah. Yeah. I mean, the funny thing was, I learned a lot. I mean, it is all in the book, to be honest.

I didn’t really write the book - the book wrote itself over 15 years in 1000 word instalments written for Digital Health online magazine. And I’ve just stapled it all together, really. But it actually reads like a diary of the last 15 years of digital health. So it’s all in there.

But one of the key lessons that we learned was there’s a Goldilocks size to an IT project - a health IT project. And if it’s too big, you’d think, oh, they’ve given me £3 billion. How can I possibly fail? But in fact, actually, you can have too much money at the start of a health IT project. In fact, if you have a really huge amount of money, you will almost certainly fail.

Tom - Yeah, because the huge amount of money will warp people.

Joe - As a psychiatrist, I thought I knew something about human nature. But when you put a billion pounds on the table, it warps space time, it warps human behaviour, it makes fat Geordie middle-aged psychiatrists sexually irresistible to software sales staff - all manner of madness would go on. And that was news to me. And I learned a great deal.

Tom - I’m super interested in these massive infrastructure projects, because I don’t think globally there’s any examples of that sort of scale succeeding. And I’m really interested in - we obviously had the National Programme for IT, it did lay some great foundations. And I think what we forget, when we look back, is how far we’ve come, right?

We really didn’t have any digital infrastructure 15 years ago. And actually, we have got now in the NHS a really good ecosystem of products and solutions. It’s a bit of a patchwork, which some people don’t like. I personally think is a good thing, because I think a patchwork is more resilient.

But it does seem like the national conversation is flipping back to big central projects. And I’m really interested in your take on that. Like, the pendulum swings one way, and then it swings the other. And is it a good thing? Is it dangerous? What can we learn?

Joe - It’s really dangerous to have really big projects. I don’t know if you’re familiar with the work of Waldo Jaquis in the United States. He’s a guy who works for a government organisation, and he works purely in the field of how do they get best value for money out of their IT.

And his golden rule is, if you’re spending more than $10 million, you will fail. So he says, as soon as the money gets over $10 million, you will fail. And he’s got half a dozen examples.

We have plenty of examples, not just in healthcare. I mean, the Post Office kicked off at the same time. And I’ve just listened to a podcast on the train here, actually, about the justice system, which has spent a billion pounds on a system which isn’t working.

And these great big government IT projects, I believe, are predestined to fail by their very size.

And when I look at things that have succeeded - look at the things that succeeded from the national programme. Let’s go back 20 years. What’s still here? Spine. Relatively small project, actually. Small number of people. Open source. And that’s still great infrastructure.

My wife works on the electronic prescription service. So every time you get a prescription now, it’s at Boots before you are.

Tom - Yeah, which is magic.

Joe - You know. But again, it was a team of less than a dozen people. And they didn’t have a massive separate budget. They kind of went under the political radar and got it done.

Tom - That’s so interesting, isn’t it? And the thing that they both have in common is that they’re sort of part of the fabric. They’re infrastructure. They’re enabling services.

And when I think about what central should be building, it should be building enabling services. How do we plumb stuff together? Spine is such a great example of that. How do I get the demographics of any patient out of any GP system in real time? So enabling.

But I think the tendency is, unfortunately, if you’re a leader at DHSC or a politician, the allure and the draw of the shiny thing that you can put your name against is so great. There’s not much that’s sexy about an enabling piece of infrastructure, sadly.

Joe - There isn’t. And it’s politically unattractive.

I mean, let’s take the single patient record that has been mooted at the moment. That is such a sexy idea that we’re about to do it for the fourth time. And we know that it will fail.

You know, we did HealthSpace. We did Lorenzo, Summary Care Record, and the other one, which I can never remember. But we’ve done it to death.

It is such a sexy idea. Let’s just have one single source of the truth, they always say. So alluring. And the minister gets it.

We’re still on the national programme - it was the prime minister’s idea. And there’s only one thing worse than a minister with an idea, and that’s a prime minister with an idea.

And, you know, it is such a sexy idea. It’s so appealing. Unless you’ve had a go at it. And then, you know, actually - oh, we have got an installed base that we’ve got to deal with.

We can’t throw out 25 years worth of data - 40 years in the GP record, to be honest. We’ve got to build on our installed base.

And there isn’t one ring to rule them all. And even if there was, it would be a terrible idea, because they would just turn the handle on the price of it every year. And there’d be nothing you could do.

So it’s a terrible idea, but a great idea at the same time. And unfortunately, the newcomers, new governments and ministers - I don’t know - it’s the arrogance of youth, really. They think, oh, well, those silly old buggers. They didn’t get it right. I’m sure we will, though.

And actually, the complexity is in the task. It can’t be taken out by a little more money or a little more testosterone or a few thousand more developers.

Tom - Yeah. I mean, the institutional memory isn’t there, is it? And everybody thinks the people that went before them were fools and they can do it better.

And the reality is, as you say, the devil is always in the detail. And it would be a shame.

You know, we have shared care records. We have the Summary Care Record. Not only are they products that are out there and working and being used every day, but there’s also a whole ecosystem of businesses that are delivering those things.

And I think it’s really poisonous for innovation in the NHS every time we talk about one of these things. Because, you know, if you’re Graphnet or Orion or, you know, one of many other businesses that supplies a shared care record, I suspect business dries up because everyone goes, well, there’s a new central thing coming.

And if we as a country don’t support our innovative homegrown businesses like Graphnet, then we are going to fail and we’re going to end up with no innovators.

And I think that’s the other sadness here. You know, we don’t learn from the past and we don’t respect the innovation that’s already happened and we don’t lean into it and we don’t go, do you know what? There’s some great stuff happening there. How do we do more of that rather than starting in here with our kind of new white elephant that we’re going to ride in on and ultimately sort of-

Joe - Yeah, it is very sad. I mean, you know, 30 years ago there were 30 GP system suppliers. Right. Now there are two.

Tom - Yeah. And we’re desperate for a third or fourth because we’re worried about the duopoly.

Joe - Yeah, we are. We are. We’ve got a managed divergence policy for general practice records. Yeah. But I think we’ve still got a managed convergence policy for hospital records.

Tom - We do. Which is interesting. It’s interesting in the current time.

Joe - But I do think, um, I mean, the reason I joined Access Group is because they’re not trying to take over the world.

Tom - Yeah, they are trying to join it up.

Joe - And at a British price the NHS can afford rather than, you know, the mighty corporations who aren’t that interested in us and the way we do things.

And, you know, who are, you know - well, I always say to people when they’re buying an EPR, has your supplier got a Formula One team? Does the chief executive have their own jumbo jet? You might be overpaying.

Tom - Yes, totally. You know, it’s worth saying for listeners that DrDoctor and the Access Group have a partnership, which we’re really proud of.

I believe really strongly in an ecosystem - an ecosystem of innovative British businesses that are going and doing good stuff at a fair price. I think the work we’re delivering independently and together are really good examples of that.

There isn’t one business that can do everything. I find it incredibly curious that we think, you know, we have to put everything in one EPR.

And, you know, obviously, if you’re a clinician, you don’t want to be having to do more clicking than you need to do. But I often will compare it to what we do.

You know, we’re a tech business. We use a very modern stack, but I still use six or seven different tools every day. And they work together really well, because I actually - I don’t want to just use the Microsoft toolset. I want to be able to use Slack and I want to be able to use Miro and Figma and lots of other tools.

And I think our job as innovators and suppliers in the NHS is to prove that by working together, we can move quickly, we can deliver value and ultimately something great for patients and our clinical users.

Joe - I’ve probably done more national UX surveys on EPRs than anybody in the world. And one of the things that I learned from them was - I mean, I did the first one 15 years ago when I surveyed all the EPRs in mental health.

And that was a two man band. It was me and Darren McKenna, who was my CIO at my trust.

Anyway, we did a national user survey. And we found some really interesting things. One was, it didn’t matter that much which system you had. It was about how well it had been implemented.

So you would see RiO would score 32 in one trust, 84 in another out of 100.

And I’ve done a number of national surveys together with CLASS and NHS England over the last few years. And what emerges is that little niche systems that do one thing really well will come top of the league.

The broader the system is, and the more it’s trying to do, the lower the usability score.

So try and prize out of the midwife’s hand her copy of BadgerNet, and you will not - they will not have it prized out of their hand because it’s been designed exactly to do what midwives need to do. And those niche systems are the best beloved.

And it tends to be the larger suites, which people aren’t really very keen on.

So there’s something about, if it’s been designed to do a little module of what the NHS does, it can do it really well. But if you try and make it do everything, it becomes cumbersome and people don’t like it and it causes burnout.

So I’m a big fan of Professor Margaret Aanestad, who was at the University of Oslo. And my wife went to the Digital Academy in cohort one. And I was watching her watch a video of Margaret Aanestad over her shoulder on a laptop one time.

And she gave this 15 minute lecture about the nature of IT. And she explained that architecture is the wrong language. It’s gardening and cultivation.

And that what you have to do is you plant modules, and those which work and connect to each other, you water them with cash, and they develop into ecosystems which are usable and functional. And they’re very close to the end user. And the end user can see that it’s good. And then you can make progress.

If you try and parachute it in from outside, with a great big budget attached to it, it will almost inevitably fail.

I would say there are some exceptions, but I can’t think of any. I can’t think of a gigantic successful billion dollar imposed-from-the-centre programme. Because you can’t do a billion dollar piece of work and, as you say, listen to the individuals and water their individual needs. That’s not possible. And I also think you can’t innovate in that way.

You can’t innovate if you have a monolith - you know, a big monolithic system - because it’s trying to do so much. You can’t go and do the bit of magic that’s required.

And then that really, I think, is a risk that we have right now in the NHS. We’ve kind of over indexed on massive monolithic systems. We haven’t spent very much money over the last two or three years. And so we’ve killed lots of our innovation, sadly.

Tom - No, I agree. And there’s a lot of narrative around we need to solve it with big systems.

And I think it would be really sad to see all of that learning - you know, your 15 years of experience, which is in the book - everything that we’ve done as innovators. You know, DrDoctor is nearly 14 years old, which blows my mind.

Joe - God, is it?

Tom - Yeah, it is. 2012.

Joe - I remember when it was the height of a grasshopper.

Tom - Yeah, exactly, right. And we were doing stuff back in the day where we still had N3. The idea of opening an N3 port to the public internet was very scary to people.

Joe - Yes, oh God, I remember that.

Tom - You know, and we were having to really persuade people at NHS Digital that that was okay.

These days, you can spin up a bit of Azure infrastructure and just do it in 30 seconds, which is a fantastic thing. But man, those early years of breaking ground.

But the sad thing is, you know, I’ve watched the digital health space explode with all this interest. And at the moment, apart from a couple of areas - like some of the ambient scribe technologies in AI, which we should talk about - it’s a bit moribund. I think people are leaving and small businesses - they’re hanging on, but there isn’t a lot of positivity.

Joe - I mean, young medics who are coders, who are IT literate in a way that we - my generation just weren’t and aren’t - come to me sometimes and say, how will we sell our lovely new application into the NHS?

And I say, don’t, it’s too hard. A terrible customer, very fickle, they may well go and invest everything in, you know, the new shiny thing, and you’ll be left, you know, for dead.

And I hate myself when I give that advice.

I do generally say, go to Saudi - you know, get yourself on the plane to Dubai, sell it there. They’ve got money and they’re a better customer than we are. And that’s a real shame, cards on table.

I’m a card carrying member of the Labour Party. I thought we were going for growth. I’m not sure how we’re growing this market sector by handing it lock, stock and barrel to a group of oligarchs in America.

Tom - Yes, yeah.

Joe - It’s a bit political, I shouldn’t have said that.

Tom - No, I agree to be honest with you. I mean, I think we were all excited about the Labour government coming in. I continue to feel positive about having a Labour government. But the digital health industry in the UK feels a bit neglected and forgotten.

And it is full of people who are smart, that care - the passion that they have is so real.

I mean, I actually sat next to someone at dinner last night - he was a PhD student who’d invented a way of making prostheses fit on a first fitting, using a scan of the person and some AI to save the sort of five or six repeat visits you normally have. And he’s gone to the US with it.

Joe - Yeah.

Tom - He’s gone to the US with it. And I sat next to him and I said, it’s really sad, but it sounds like you’ve made the right decision.

Joe - It is sad.

That has to change. I mean, I do talk to my MP and she’s great. She happens to be chair of the Science and Innovation Select Committee, Chi Onwurah. Sometimes I sit next to her at St James’s Park and she gets it, but she’s not a minister. She doesn’t have her hands on the levers.

I think there is so much innovation that’s possible in a place like the NHS and the rewards that we can get in the NHS are unique. I don’t think anywhere else in the world has the data that we have.

Go right back to the beginning - you know, 20 years ago when I started off as national clinical lead for IT, I was inspired by a friend who rang me knowing that I was the only adolescent psychiatrist that he’d been at university with and said, my daughter’s got first episode psychosis. What’s the best treatment?

And I’d already been a consultant for 10 years. And I said, I don’t know. I don’t know what the best treatment for your daughter is.

I’m pretty sure on a genome, there’s something that encodes which antipsychotic would be best for her. And if I get it right first time, she’d be well in six weeks. She’ll do her A-levels and go to medical school. If I don’t get it right until we’re on drug number 12, she’ll have been in an institution for three years and she will never regain her trajectory.

So in my naivety, when I applied for the job of national clinical lead for IT - mental health - I thought I’d gather up all that data and I’d be able to analyse it and find out what’s the best treatment for first episode psychosis. And that’s still the dream.

The only place you can do it is here. Everywhere else in the world, the data is too distributed. We have a national health service with, you know, 30, 40 years worth of data in a get-at-able format.

We now have tools that can make sense of what’s data soup, to be honest, but we now have tools that could maybe answer some of those questions if we can free it up.

Tom - Are we doing that? Is that happening anywhere?

Joe - I think very soon it will happen. If you look at some of the great work that’s going on - Ben Goldacre’s OpenSAFELY project - where they found a way to run queries on data without it leaving where it is.

And we’ve got plans in the Access Group, actually, because, you know, my dream is that at some point I can say, Alexa, show me everybody in Newcastle tonight who’s thinking about killing themselves, you know, and there’ll be clues in that data if we can get it into an analysable format.

And I don’t think we’re a million miles away from that. We are beginning to look at ways in which we can start to answer those questions. What’s the best treatment for schizophrenia? Who’s in trouble tonight?

Because as a former medical director of a mental health trust, Friday evening, five to five, you’re thinking, what I wouldn’t give to know where my hot 100 are this evening. Who’s going on the bridge? Who’s going off a cliff? And are there clues in the data? And I think there are.

You know, the indicators are that they’re all there.

And could we actually make that difference? And I think the last 20 years of work in digital mean that we are getting close to that - being able to really answer some very difficult questions.

But it is a unique opportunity. The Americans aren’t sufficiently connected to do what we can do.

I’ve always said this. I’ve said that the NHS is the best place to build a digital health business. And if you can prove it and you can scale it, then you should be able to export it.

And the NHS should be an innovation factory, creating exports that go and deliver value. And it’s a shame that we’re not that.

Tom - I want to dive a bit into what you just said, because, you know, the state of mental health services at the moment is really tricky, right?

Lots of people, lots of clinical burnout, lots of patients waiting a long time. I suspect the number of serious mental health incidents that we’re having is increasing because of the lack of capacity in the system.

And it does feel like there’s an opportunity, particularly in those services, to use data to change the way we practice, almost more than anywhere else.

And to use the combination of that big data set, perhaps remote monitoring, utilising some of the tools that already exist better.

What’s your take on that as someone who is still practising?

Joe - I think there are tremendous opportunities to use technology to make things better in the mental health spaces. Some of it’s really obvious.

So, the ambient voice scribing - I know there’s been a lot of worry about it. And does it hallucinate, etc? And is it safe to use?

I believe all of that is doable. You know, we have integrated ambient voice in the new version of RiO. And it does save a tremendous amount of time.

So, I did stop practising for a while - gave up my licence to practise. I’m back now, actually, because the only bit I ever enjoyed was the bit where you were talking to the patient, listening to the patient. The bit that I hated was all the admin. I hated writing it up afterwards.

Tom - Me too, Joe.

Joe - GP letters and all the rest of it. And now the AVT has taken that all away.

So, there’s a genuine productivity gain there. And if everybody had AVT, if every mental health worker in the country had AVT, they’d all buy, I reckon, 10 minutes in the hour.

10 minutes in the hour multiplied by the thousands and thousands of people who work in mental health - I’m pretty sure we can solve most of our problems in a couple of years by saving 10 minutes in the hour.

Tom - That’s our productivity target done.

Joe - Exactly. And once you get into the virtuous circle of that list coming down, and I have great hope, actually, that waiting lists can be disappeared because I’m old enough to remember the last time we disappeared the waiting lists. And we did.

I mean, we had massive psychiatric - especially child psychiatric - waiting lists back in early 90s. And within a couple of years - I know Tony Blair’s not everybody’s cup of tea - but within a couple of years of Tony Blair coming to power, they’d gone.

I had no waiting list. I was a child psychiatrist without a waiting list. So, I know it’s doable.

Tom - Yeah. That’s the positivity we need. Was it one thing or was it - like normal - it’s a thousand small interventions?

Joe - It was many things. But mostly, they went low enough in the organisation and said, you’ve got a big waiting list problem. How are you going to fix this? And what do you need?

And we literally, we came in on Saturday mornings. And we worked on Saturday mornings.

Tom - You got reasonably well rewarded for it.

Joe - I got reasonably well rewarded for it.

Tom - Yeah. Which is fair, right?

Joe - Absolutely, it is. But once you’ve got rid of the waiting lists, actually, other things become possible.

You begin to make other efficiencies. You begin to get other people to do what the psychiatrist should - only the psychiatrist should do.

Other bits of the job can be formed out to people who aren’t quite so expensive as a consultant psychiatrist. And that is possible in all sorts of ways.

But once we got out from under the tyranny of the crushing waiting list, then you can start to think, well, how can we do this even more efficiently?

Because you’ve got the time, you’ve got your nose up from the grindstone long enough to take a look.

Tom - A little bit of space.

Joe - We could do this better. Maybe we could automate some stuff.

Tell you what, instead of spending the first hour of your first meeting with the patient, explaining to mum how to fill in a Conners rating scale for ADHD, maybe you could put that on video.

Tom - Right, right.

Joe - It could be explained really well. Every time.

Rather than by, you know, in a slightly rushed manner by somebody worrying about they’ve got to pick the kids up from school - and they might actually get a better chunk of knowledge out of the service in a video clip than they will in a real life encounter.

And then there are things - lots of things that we’ve done. I mean, it’s a really old joke. Why did the patient go to the clinic? Well, to remind the consultant to look in the notes. But it’s kind of true.

That is often what’s going on in outpatient clinics.

So there are many ways in which we can improve the efficiency of the NHS, which technology can definitely help with. AVT being the lowest hanging fruit.

Tom - Yeah. And I think there will hopefully be a big AVT boom next year. And in a year’s time, most people will be using it as default. I think that is quite likely.

You know what? I’ve spent probably 15 years pushing technology onto my clinical colleagues who didn’t really want it. And now I’ve got people in a queue for AVT. Literally.

We want to go next. Trusts complaining that we’ve let their neighbours go before them.

And do you think those tools will end up living in the EPR? Will people buy separate things that live outside of it? Like, where are we going to end up there if you had a crystal ball?

Joe - So I’ve travelled the world and looked at what’s on offer. There are very few examples. RiO is one of them. InterSystems is the other really good one, where they’ve absolutely integrated the AVT into the EPR.

So there’s no cutting and pasting or taking out of one application and putting it into another. And I think that’s the way.

I’ve been very keen that that’s how we do things in RiO. Because I think that integration is absolutely key.

The other thing for me is, if it’s baked into the EPR, after how many ADHD assessments does the machine become able to do most of the assessment?

I’ve done hundreds. And after a while, they get quite similar. And your process for reaching a decision about prescribe or not, psychological methods or not - I’m pretty sure modern AI could sift out the cases that need to see the human.

Then you begin potentially to drop the cost of that treatment pathway. And I think that’s a really interesting area.

But within a year or two, we’ll have enough of those consultations, which have been described by AI, to begin to go, well, which of these really need the human?

Tom - I think what’s interesting about this is, so I have quite a strong hypothesis that the digital health market is going to see a convergence in value proposition.

So instead of I’m an EPR, I’m an AVT tool, I’m a patient engagement tool, I’m a skin analytics piece of AI, we’re all going to find ourselves actually in the same sort of place, which is how much of this stuff can be automated?

And how much can we allow you as a clinician to focus on the high value patients that are complex and need the human touch?

And the point that we achieve that, I think we haven’t solved the problem, but we’ve moved forward so far.

The interesting thing for us as suppliers in the space is that convergence, I think might happen quite quickly. And what we traditionally thought of as an EHR could look really different in 12 or 18 months time, I think.

Joe - I agree. But the really difficult bit - there’s a chapter in the book about, I go out in the middle of the night to see a patient who’s psychotic in a police cell in Gateshead.

Now, AI is never going to know how to get to the police station in Gateshead. Well, it probably can do that, actually. But it won’t know how to pull a 16-year-old psychotic patient out of the nightmare that is psychosis after they’ve been banged up in a police station. That’s a way off yet.

But that’s where I felt I was adding the most value. Carrying out routine assessments of ADHD in my clinic - much of which I felt might be automated - not so much.

But in those really difficult cases where actually your 20 years of experience absolutely counts, they’re where people who have the knowledge and consultants should be applied. They should only be doing what only they can do.

Tom - 100%. There are some things that only humans can and should do. I believe that very strongly.

And the key to the technology is removing the burden so that you can go and do that.

I was really interested - I saw on LinkedIn this morning, actually, that a mental health AI business called Yara.ai had decided to shut down out of choice.

And it was a business that was using LLM-based technology to create therapeutic chatbots. And the founder said, two years into our journey, we’ve decided to close this down for ethical reasons because we’ve realised that the chatbots aren’t capable of giving patients who have serious mental health needs what they need.

And I thought that was a really brave and impressive thing for that founder to do.

I’d add the word “yet” to that statement. I think he’s done the right thing. And I don’t think they’re up to it at the moment.

And there’s a very real risk when there’s a two year wait for an NHS mental health appointment, that people are going to turn to chatbots - ChatGPT - and try and therap themselves. And I don’t think there’s any doubt that that is dangerous right now. And it’s happening.

But it is happening. And I know that steps are being taken to try and stop it happening. I’m quite sure they’ll work.

But the answer, I think, is let’s concentrate and get rid of waiting times. Let’s make sure that the humans are doing what only the humans need to do. And then people can get access to therapy, you know, when they need it.

That’s where we need to go at the moment.

But after AVT has sat in enough - yeah, a year, two years, 10 years, 20 - pretty sure it will get pretty good.

Now, I think there may be an element in therapy which is so human, that there are things that AVT and AI will never, ever be able to do.

I suspect there’s an element that somebody in this conversation cares about me. And that is like an enabler.

And I think there is some magic in the human contact. Yeah. And if you take it all out, yeah, I suspect it probably won’t work.

Tom - I wonder if there’s a risk almost with some of these, you know, computer chat interfaces that it does the opposite. It sucks people inside themselves, because it reinforces perhaps some of the negativity or anxiety.

I’m sure there’s huge potential for good, but potentially huge potential for harm at the same time.

Joe - Yeah, I think that’s true. I mean, you know, if you look at social media - let’s say you’ve got anorexia nervosa, and you join a group online, and you’re talking about your anorexia nervosa - you might be in with somebody who’s really positive, help you out there. Or it might turn it into a dieting competition. And it might actually make you worse.

I worry that AI is insufficiently able to tell the difference between what’s the good stuff and what’s the bad stuff here, because it’s just making associations.

I saw Geoffrey Hinton at the Royal Institution a couple of months ago. And he was explaining - and I thought I understood how AI was working - he explained that language models, all they do is they see that words are multifaceted.

And he described them as having hands and different sized hands, different coloured hands. And they would just grab onto all the words that they fit with.

This means that words actually don’t mean anything on their own. They’re entirely dependent on the context in which they exist.

And that’s dumb. You know, that’s not what people do, or at least I don’t think it’s what people do.

Consequently, these models would need an awful lot of training to produce anything meaningful from a therapeutic point of view.

Tom - Yeah, totally. It is just kind of statistically linking words together rather than producing anything real.

I’ve actually found my use of LLMs reduced recently, which is interesting. Having started using them a lot in my professional life, I started using them less and less because you start hitting the limits quite quickly - all sorts of use cases.

I also think it’s quite sad that the conversation around AI has become so LLM centric. I think because of the big investments from the big tech firms into that particular technology.

But, you know, like my - I was sort of laughing last night with some of my university dissertation. I wrote a genetic algorithm, which is an AI, but we didn’t call it an AI then. You know, that was.

And there’s all these different forms of machine learning and artificial intelligence that, to solve some of the problems we’ve just touched on - like finding your hundred patients who are at risk - we should be investing in.

And that side of the conversation has got a bit lost, I think, amongst the rush towards AVT and the rush towards LLMs.

So, yeah, there’s definitely an opportunity there, I think, to think about how we change practice.