Welcome to the latest episode of the Pain Free Living podcast with hosts Bob Allen (osteopath) and Clare Elsby (therapy coach).
Have you ever heard of Artificial Intelligence (AI)?
If you said no, then this episode is for you, and even if you said yes, this episode is still worth your time, as it is such a huge subject that you might learn something new.
AI is such a broad subject that we have covered it in a three-part mini-series where we take you on a jargon-free tour of what AI is, where it came from, and why it matters in therapy coaching, and healthcare.
You’ll find out how early AI tools began in the 1950s with Alan Turing’s big question — “Can machines think?” — and how the first therapy chatbot appeared as early as the 1960s.
Bob explains why early medical AI struggled with real humans, why “if–then–else” decision trees failed, and how today’s systems rely on large language models rather than true reasoning.
Clare explores cyberchondria, covering the rise of health anxiety driven by people looking for their symptoms online and why context is essential when using AI for anything health-related.
You’ll also learn why using AI is not the same as asking Dr Google questions as it can “hallucinate,” when it doesn't know the answer. In addition, providing AI with vague prompts can make things worse, and we show you how to ask questions that produce safer, more reliable answers.
Don’t worry — getting to grips with AI isn’t about becoming technical; it’s about understanding how it works, what it can do, and why guardrails are important.
This episode sets the foundations. In part two, you’ll hear how AI compares to real therapy, why dependency can be a concern, and where AI can genuinely support wellbeing.
5 Key Takeaways
AI is a prediction machine (think sophisticated autocorrect), it doesn’t know anything!
Humans are very complex, and simple algorithms miss the bigger picture.
Cyberchondria is increasing as people search for symptoms without context and accept AI answers.
Good AI use depends on good prompting and detailed questions.
AI can support therapy, but it cannot replace human care.
Note: This podcast provides general information for educational purposes only. It is not medical advice and should not replace professional assessment, diagnosis, or treatment. Always seek qualified healthcare advice if you have pain or other symptoms that seem to be getting worse, or any concerns about your general health, before starting exercise or self-care routines.
A warm welcome to everyone to the Pain Free Living Podcast with myself, Claire Elsby and Bob Allen.
Speaker B:
Hello.
Speaker A:
We thought we would do a series of episodes. There's going to be three of them. We think there might be more, but there's at least three.
All about AI, but specifically about higher effects, coaching, therapy or therapy and healthcare. So. So AI itself is not new, but it is much more accessible and it's much more visible these days to us as the general public.
And the one key takeaway is it's not Google.
So we wanted to explain in simple language as possible, jargon free, what AI is in terms of therapy and healthcare, how you can use it, what tools are available and what to watch out for.
Speaker B:
I think that sums it up pretty well.
Speaker A:
Okay, well, I'll start with the history of AI.
te surprised to learn that in:
Speaker B:
Turing test.
Speaker A:
Yeah.
And so the Turing Test which you've just mentioned is about, if a human has a conversation with a machine, will the human know that it's having a conversation with the machine? And that was his test.
Speaker B:
Yep.
Speaker A:
So that was:
But the difference was it was a conversational tool and though the output wasn't actually therapy, it didn't deliver therapy per se. But what the individuals, the people who used it felt that they'd actually been heard whilst using the tool. So I find that really interesting.
Roll forward.
s,:
So it was very much data in, data out, but it, it was able to summarize all sorts of symptoms and diagnostic tools that doctors could use. So it was a sort of forerunner to the assistant that we try and think of AI as today.
We now roll forward to today and there's probably about nine popular platforms that is available. The most popular one in the UK is by far is ChatGPT. But there are other ones.
There's Microsoft, Copilot, Google, Gemini, even Elon Musk has got one called Grok. And the AI that we have today is very much further advanced than we've had. And what we're getting is it's all based on large language models. So.
So where we are today with AI is there's obviously lots of ongoing development. We've now got about 12 platforms of which ChatGPT, particularly in the UK, is by far the most popular.
There is also Microsoft Copilot, there's Google Gemini and Elon Musk has brought out his own version called Grok.
Speaker B:
Excuse me.
Speaker A:
Oh, sorry.
Speaker B:
And I forget Perplexity and Claude and.
Speaker A:
Various others, there's lots of different ones that people can use and access. But the thing to note is it's not Google. It's a very different way of questioning and prompting and we'll come to that later.
As to the best way to actually use ChatGPT or use AI, I should say, did you want to talk a little bit about the history affecting physical therapy and how that's been developing slightly.
Speaker B:
ke off until around the early:
So that's the way that it's looked at in the nhs, for example. It'S very decision based. It's what they call an if, then else type scenario.
Speaker A:
So.
Speaker B:
So the medical model is very reductionist in that. So rather than taking into account the whole.
So we have 11 systems, or 12 systems, depending on what you read, rather than taking into account that whole. Modern medicine currently tries to keep things as simple as possible. So it looks at a particular aspect.
So if you had knee pain, you then go through kind of what they call a decision tree. So you look at the condition and then you have two branches off from that.
So if it's this, then you do this, if it's this, then you do this and you go down that route to hopefully come to a simple conclusion which is easy to treat. As they found out in the early models that they use for physical therapy, human beings are much more complex than that.
So it's not a case of if this, then do this. The medical model is quite simple and it tries to keep it simple because then it makes the treatment relatively simple.
Lots of modern medicine is moving towards the field where they where what we call the biopsychosocial model.
So it's not just looking at one particular aspect of a problem, it tries to take into account all the different systems because all the different systems can potentially be affected. So the biopsychosocial model looks at the biology. So, you know, so they have knee pain.
Okay, what the biopsychosocial model will look at is, well, okay, you know, when did it start? How did it start? Did you have a trip or a fall? Did you do this? Did you do that?
So that's the kind of ties in a little bit with the medical model, but it also looks at the psychology of the person. Are they stressed? Because if you're stressed, and we've covered this in other podcasts, if you're stressed, that can make you more sensitive to pain.
Are they happy? What is their social environment like? You know, are they, are they a group type person? Do they like, do they have lots of friends?
Because one of the things that we know is that the more sociable the person is, the better the social life they have. The. It helps with long longevity. I'm not sure if they've exactly narrowed down exactly what it is because it's not a simple one thing.
But the model that they tried to use for the physical therapy side of things tried to keep it simple and basically it didn't work. Humans are complicated. So fast forward to today. What they're trying to find is different roles for AI. AI isn't clever.
AI is a prediction machine like Autocorrect, but it's just got a much bigger data bank, the large language model. But it doesn't think so.
The thing about physical therapy is that we don't just look at where the pain is coming from, we look at the person as a whole. Because there are so many different factors.
I said earlier from the biopsychosocial model, there are so many different factors that can be involved in that person's pain. So AI isn't intuitive.
Whereas if, like, for me, for example, I'm hitting my 17th year as a physical therapist, so I have a massive data bank of information and I can pull information from different cases that I've seen to come up with a conclusion. AI can't do that.
I would like to say it can't do it yet, but I would be very, very surprised if you could do that five, ten years down the line either. But that's a whole nother story which we'll cover another time.
So, yeah, so from a physical therapy aspect, early, the early tools that they had were not great. They were great. They were okay for simple problems, but because we're not simple, they.
They were limited in their use and that kind of fell away forward to the modern day. They're now looking at using AI for slightly different aspects of physical therapy.
Speaker A:
So in terms of a particular interest that I have is about hypochondria We've talked about this. We have talked about this and there is actually a new. So let me explain what hypochondria actually is. The definition of hypochondriac is.
Hypochondria is illness anxiety disorder. So it is a clinical disorder and it is the excessive and persistent fear of having a serious illness.
What they'll concern, I guess with AI or what we're seeing is a thing called cyberchondria, which is where people might use, in fact, they might use Google or they might use AI to search, for example, a tension headache. Because they might have a tension headache, but actually they then think I've got a tumor.
And it's that relationship and it's how can we support people who might be susceptible, if you like, to that kind of thinking?
And make sure that we put the guardrails in with AI so that therefore the information that they're getting in terms of the responses is actually the correct information.
Speaker B:
Yeah. And that kind of brings us on to catastrophizing as well.
So as Claire said, we've kind of moved from hypochondria to cyber chondria, which basically just means you use the Internet to get that, that incorrect information out.
So in my experience, what tends to happen is it's part of the human condition is that if we have a headache, we naturally go to the worst possible conclusion for that. Even without, even without the Internet, we all automatically think, well, is it a brain tumor? Is it cancer? Is it something else?
Because I think that's partly a self preservation instinct. So if we think the worst possible case, then we are going to be, we're going to maximize, trying to protect ourselves.
And I think that kind of feeds into the region beta paradox, which is all about if a condition is really, really, really bad, we will do something. If we think it's okay, and I can live with it, we tend not to.
So our brains naturally go to the worst possible scenario because that means that we will then take action. If you don't know what the region beats, paradox is, I'll put a link in the show notes to a podcast we did a while ago about that.
But it's quite an interesting little thing. But I think that again, that feeds into the cyberconjria. Back in the days when we had Google.
I know because I put in low back pain a couple of years ago and it came up with 47 million hits. If you put the question into an AI, into an AI agent and said, I have what does low back pain mean? It will Come back with one answer.
And if you know how AI agents work, then you know that there is more, there's more to it than that. But if you ask it a simple question like I have low back pain, what does that mean?
Then it could come back with an answer that's, let's say, less than reliable. It can potentially come back with an answer that says that you have low back pain, therefore.
You have cauda equina syndrome, or you have something that's really, really terrible. But because there's no context, unless you give AI context, it will give you the answer that you. That you've asked.
It'll answer a question, but it won't necessarily give you the right answer to that question. So this is one of the issues with AI.
It's a people pleaser, so it will try and give you the answer that it thinks you want, but it won't necessarily be the right answer. To get the best out of AI, when you ask it a question, you have to give it lots and lots of context.
The more context you can give it, the better the answer will be. Without that context, if you ask it a simple one line question, it could. That increases the chance of AI coming back with a.
What they call a hallucination, which basically means it's just made stuff up. So the way that you ask AI the question, how you use AI is really, really important.
One of the things you'll notice in the next couple of our podcasts is that we may veer off a little bit, but. But that's because we are really keen on giving you the information that we think you'd need that will help you to get the most out of AI.
And Claire is here to bring you back on track.
Speaker A:
I am.
Speaker B:
Where are we?
Speaker A:
Yes. So we were. This episode is really about the early tools and how we see AI growing at the moment and what it's useful for.
Speaker B:
Absolutely.
Speaker A:
Just one of the things that I see or hear a lot about are chatbots in terms of therapy. And there was one particular one that was very, very popular. I think it had over one and a half million users in August 25. And it's called Woebot.
And it was started in:
But it got shut down in June:
So there is a lot going on in this space at the moment about what is actually allowed, the dangers, the guardrails. And there is no doubt that there is actually a market for. Therapy chatbots or therapy chat.
Because unfortunately, we don't have the systems available, certainly not in the UK, to be able to support 24.7help. And in my mind, some help is better than absolutely nothing. But we've got to be very aware of, like Bob says, what we put in the questions we ask.
And we need to be able to check the actual results that we get out of AI.
Speaker B:
We do. And we will put what CBT is in the show notes, just in case you didn't realize, it means cognitive behavioural therapy. And we will cover. We'll.
We'll put a link in the show notes to give you a better idea of what that is.
But yeah, I mean, one of the big issues with AI is the fact that it can go off at a tangent unless the prompt is phrased correctly and with enough detail, it can give you answers you may not want.
Also, there's a danger with dependency because although AI can stand in as a tool to help support people's mental health, because it's available 24 7. That is not what a therapist does. So there's a risk of it building a dependency. And then I'm stealing your conversation.
Speaker A:
Well, no, I think we're going to cover in our next episode, actually, idea of the difference really between AI and therapy, what you'd expect in the two partner sessions.
Speaker B:
We will, we will. Are we done with this one, Claire?
Speaker A:
My only. I think my last thing would be are we at what would Alan Turing think of where we are now?
Because I know in the last six months, certainly ChatGPT have brought a conversational. AI. And it is incredible. Absolutely incredible.
So even though in my mind I know it's not real and we're talking to a machine, I'd be hard pressed to know that it's machine. So I think Alan Turing would be. I think it would pass the. Yes, certainly the conversational test.
Speaker B:
I don't think the Turing Test is relevant anymore, to be honest.
Speaker A:
No. No. But it does throw a light on when we are talking about prompting and asking the right questions and framing.
You don't get that opportunity in a conversation. Not so much. So this is another concern that I've got going forward, but we'll reserve that for another podcast.
Speaker B:
We will, we will. So hopefully that's given you an idea of the history of AI and given you an idea of where Claire and I see ourselves and how.
What we haven't mentioned is how beneficial AI can be. But I think it's very important to give you an idea of the risks and concerns that we have. We will come on to the benefits of AI. I use AI personally.
I'm sure Claire does as well. But, you know, I think the main danger is that people don't and don't appreciate how different it is from using your typical search engine.
So that's what we're trying to bring across. A, it's not the same as using Google, but B, there are risks, but there are also benefits, and we'll cover those in a couple of later episodes.
So thanks for watching, hope you've learned something, and go tell all your friends. Bye for now.
