Send us Fan Mail In this episode of CBmE & U, Associate Professor Charlotte Blease (Uppsala University) talks about the placebo process for validating and testing mental health chatbots. With James and Sinead, Charlotte discusses whether, if we don't have the means to measure whether things like ChatGBT and theratbot are making us better or worse off, are justified in letting people use them? And importantly, when you get those pesky or uncomfortable messages from people, should you use openAI to write them? Or are we doing something wrong turning to AI instead of people? Remember people, be savvy!
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
SPEAKER_00The center presents a new podcast, C BMENU, a podcast devoted to promoting awareness and debate around controversial ethical issues in biomedicine.
SPEAKER_02Welcome back to CB Me and You, a podcast about biomedical ethics. Today, are clinical trials really all that? When does a pill work better than a tic-tac? And how do we know that ChatGPT is truly the therapist it makes itself out to be? Today, we have Professor Charlotte Blees from uh Uppsala University and Harvard University, and will be talking with us today about the placebo effect and digital health tools. Welcome, Charlotte.
SPEAKER_04Thank you. Great to be here.
SPEAKER_03Welcome, welcome. Um, our very first question for you today is do you prefer coffee or copy?
SPEAKER_04Um, I like copy. Anything that just gives me the caffeine hit, I'm good with, but this is excellent. It's a big hit. It is a big hit. It's excellent, yeah.
SPEAKER_03I I think it's like at least two times more caffeine than a normal than a normal cup.
SPEAKER_04That's good by me, yeah.
SPEAKER_02I feel like you should warn people before you hand them that coffee cup.
SPEAKER_03Yeah, you are gonna be particularly buzzed after this.
SPEAKER_02The whole day.
SPEAKER_04Augment those placebo effects, yeah.
SPEAKER_03Nicely do that.
SPEAKER_02Um uh but then you need a control trial where you're like, oh yeah, this does nothing, but it's just nice. Yeah, it's good. And see if it works the same way.
SPEAKER_03Second question. Where do you want to travel to next?
SPEAKER_04Oh, that's a good one. Um, I would love to stay in Asia and do a bit more traveling. I remember being in Thailand, it was 25 years ago I was in Singapore. Um we went to Thailand afterwards, and I've really got a hankering after seeing Thailand again. So we'll say Thailand.
SPEAKER_03A hankering. Yeah. Hankering for Thailand.
SPEAKER_02Hankering. Well, what are you hankering for the most?
SPEAKER_04Um it was just such a friendly place. Um, and there were bits of it we had Northern Thailand we didn't visit. It was me and my sister who did the backpacking thing. Um, there was something about the friendliness and the light and the the signs of the place and the colour. It was just so colourful. I stayed in Bangkok, I really enjoyed it.
SPEAKER_03So I love Bangkok.
SPEAKER_04Yeah.
SPEAKER_03Um, I remember it was one of those uh places where as soon as I hopped out of the airport and then got in the taxi and then looked out the window, just like this, I was like, oh, this place is awesome. Yeah. And I don't know why, I don't know what it is, but Bangkok is definitely one of my favourite cities. It's fantastic, yeah. Our final and final question of the podcast in the most important, we'll be wrapping up after this.
SPEAKER_02Signature question.
SPEAKER_03What song is your favourite to sing at karaoke?
SPEAKER_04I can't honestly tell you because I would have to be very drunk to sing karaoke, so I I have no recollection of what I sing. Probably Frank Sinatra or something ridiculous. Um to the moon. Um, that's life. Um, do you know that one? Yeah. I think I sang that once very badly in in a karaoke bar in Paris.
SPEAKER_02I don't know if it'll offend you to know that the first version that I attach to of that song is the West Life version.
SPEAKER_04Oh, is it? Shame on you.
SPEAKER_02That's why I said it might offend you.
SPEAKER_03Hey, I was like seven. So so so you don't do karaoke very often?
SPEAKER_04No, I don't. Okay, okay. Yeah.
SPEAKER_03All right. With a benefit of the society.
SPEAKER_02Yeah, but you know, I think you gave a very polite answer to that question. We've had guests come on say and say, Yeah, I think I'd rather die.
SPEAKER_01Yeah.
SPEAKER_03I wouldn't go that far, but it is excruciating. Yeah. I think Julian did say something along the lines of over my dead body, would I do that again? Yeah. But he did say again. I still remember that. I still I still remember that. It means Julian's done it at least once.
SPEAKER_02I still think it's possible. Okay. So we're gonna start off with uh a kind of basic question. Yeah. What are placebos and what are nocebos, which I know you've talked about as well before, and how can AI systems bring them about? What do they have to do with placebos and nocebos?
SPEAKER_04So, first of all, I would differentiate between placebos and placebo effects. Um, so I would say a placebo can mean a couple of different things. It can mean in a sort of common vernacular that doctors and patients might talk about where it means giving somebody something, but you know that it's not going to be effective. And it might be to please patients or a save face or a whole variety of reasons. But a placebo in a clinical trial means something very distinctive, which is it's a measuring tool for measuring a treatment that's under investigation to see if it's genuinely effective.
SPEAKER_04Um, and then from that I would say a placebo effect is something very different again. Um, and it could be elicited by placebos understood in either of those senses, um, or indeed by any treatment or no treatment at all. And it's the effect that arises when an individual expects a treatment, or it could be uh behaviors from somebody who's helping them that elicit genuinely uh beneficial effects, and this has been known to be useful in reducing pain, anxiety, uh, depression.
SPEAKER_04Um, and then the flip side of that, if you like, so often characterizes the evil twin, is the nocebo effect, whereby if you expect a treatment uh to uh be harmful to you in some way, then it does actually result in some kinds of nasty side effects or you or anxieties and so on. Nocebo effects are much less investigated than placebo effects, um, but both both of them are really relevant in healthcare contexts.
SPEAKER_04And I also want to say this the evidence doesn't suggest any of these effects are curals, or they you know they're they're not useful useful for lots of conditions, but they can be meaningful effects for lots of conditions and symptoms.
SPEAKER_03So my understanding of a Nasebo effect actually comes from an episode of House MD, where most of my medical knowledge comes from. I remember there was an episode where um House and uh the big boss lady uh who heads up the hospital are on a plane, and then everyone's getting hysteria about this bacterial rash spreading throughout the plane, and everyone starts throwing up, and then he runs a quick placebo or test on them, and it turns out that they're all just a bit hysterical. Is that the same thing?
SPEAKER_04So, yeah, so the nocebo effect, um it so yeah, getting identifying when something has led to these sort of negative expectancies, it's quite difficult to do, but certainly, as you say, at a at a public health level, it could be relevant. Uh and take one example, COVID and the vaccines. So in clinical trials, when people were given the vaccine um versus a control uh CLIN injection, but also told about the potential for side effects, people in the control conditions did experience side effects.
SPEAKER_04So um it so when you that's a big area for no-sable effects, just disclosing the potential side effects of treatment. And this is what creates this ethical dilemma, uh, whereby we want our clinicians to be open and honest, but we also don't want them to uh, you know, first do no harm. We don't want them to to to to disclose things that then are sort of self-fulfilling and lead us to actually feel harm. I give an example of when I got my COVID vaccine, the nurse was very nice and she said, you know, you're gonna feel a little prick in your little um, you know, a bit sore in your arm, but she said uh it'll be okay.
SPEAKER_04But 12 hours, which is really specific, so 12 hours from now, you might get a headache and you might feel a bit shivery. And I thought, did you really need to be so specific? I looked at my watch and I, you know, I would then 12 hours later I actually did start to feel a bit, and I wondered what, even though I was aware of these um this phenomenon, I wondered, you know, was part of it attributable to just these negative expectancies? But yeah, so it it cre it's enough to create a uh dilemma in medical context, what should clinicians do or say?
SPEAKER_03So how does AI come into that space?
SPEAKER_04Yeah, so it's interesting. We've got these chatbots now, uh, you know, your chat GPTs, it's are conversational, they have the back and forth. And because people tend to anthropomorphise these tools, um they may consider them to be authoritative and confident, but also we've got a whole range of studies now which suggest that they could be as empathetic or indeed even more empathetic than clinicians. Now, what we know from placebo effects, studies in the placebo effect, is that the uh confidence of a clinician um plus empathy can augment placebo effects.
SPEAKER_04So we have this really interesting scenario now with chatbots, and should say I don't know of any studies that have demonstrated that these tools yet. Obviously, I think that's the next step, but these could be for eliciting placebo and potentially nocebo effects too. I think it more likely that they well it depends on the scenario, it depends what you're uh conversing with the tools about, but because they are so relentlessly polite, uh they respond rapidly, um, and they can be really impressively compassionate uh in inverted commas.
SPEAKER_04It would be uh we would expect, I think, that there these could be very important uh fora for eliciting these effects, given that people can use them round the clock as well, and that they have those sort of uh obsequious biases that are built into them. In other words, they're always trying to uh flatter p people, please, and so on, and that might also uh help to strengthen a relational uh dimension between the the individual and and well the chatbot isn't real you know, it doesn't feel anything.
SPEAKER_04Um but yeah, so it it could be it could be a place whereby we would expect these effects to arise.
SPEAKER_03So this is something a bit different and potentially a bit scarier than the the phenomena that most of us go on the internet go into WebMD, put in our symptoms, and then think we're all dying. So this is worse.
SPEAKER_04Um I actually think the jury is out on that so far. So we've seen a lot of headlines about the potential for harm with these tools, and I d I believe that there are harms associated with these tools, um, particularly for young people, or potentially if they displace relationships, people are using them when they're lonely. Um so these tools could then um you know exacerbate feelings of loneliness or depression and so on.
SPEAKER_04There's some evidence of that so far, but it's not causal evidence, it's correlational. Um, but I think that they we don't I mean what one of the issues with these tools is they could be better than nothing. So if you don't have access to any kind of emotional support or advice, they may be offering you something. Um I'll give one example. There was a survey by uh Mental Health, I think it was Mental Health UK, it was it was a foundation mental health foundation in the UK that did a survey that was published in November last year, and it was a representative survey across Britain.
SPEAKER_0437% of British adults said that they had used these tools for emotional support, mental health support, and that got you know as sort of staggering in a way. When you consider the lack of resources that are put into mental health care and the waiting lists, um, you know, it's it's perhaps not surprising. But what was also hidden within this study was two-thirds said that they derived benefits from these tools, and that wasn't as widely reported.
SPEAKER_04And I think we shouldn't gaslight the public either. If they're deriving some support in some circumstances, that could be a good thing. On the other hand, it's highly likely that these tools could be can be harmful as well. So it's going to be a mixed bag. They're going to be good and they're going to be bad. It's just it's teasing out the circumstances for both.
SPEAKER_02And it's a situation where the numbers are probably increasing quite rapidly.
SPEAKER_04Yes, absolutely.
SPEAKER_02At 37%, probably outdated by now.
SPEAKER_04I would say so. Um and getting into that, we did a survey, uh, we've been doing surveys every January for the last three years of doctors' uh use of these tools, GPs in the UK. And 2024 January we did it. It was the largest survey at the time undertaken among doctors, and 20% said they were using these tools, it was mostly Chat GPT, for any clinical task. We did it the following year in 2025, that went up to 25%.
SPEAKER_04Start of this year, it's not yet published, 52%. So I think you were seeing a rapid uptake. I don't know if we're yet at a tipping point, but people are incorporating these tools into their life. And we have fewer surveys of patients' adoption of these tools. Most of the survey research has been conducted among clinicians in the healthcare space, but it would be very surprising if patients weren't using these tools. With the first opinion service, second opinion services uh for emotional support, all kinds of uses.
SPEAKER_02Yeah, I I had a situation recently where a family member had a medical emergency. They're okay now, in case anybody's wondering. But you know, while they're waiting in the emergency room and I'm just sitting there twiddling my thumbs, I'm not just gonna sit there like a schmuck. Yeah, I pull out my phone, go to I think it was MetGPT or something, uh, or or Chat DXT. What you know, I can't remember the name, but you know, this is really cool. You keen all the symptoms, and I tried sounding as doctory as I could. Uh and it's just like here's what um the most likely uh diagnosis is.
SPEAKER_02It ended up being right.
SPEAKER_03Really?
SPEAKER_02Well, it it gives answers in probabilities, right? It says uh this is the most likely thing, this is the second most likely thing, and all that. But the most likely answer was it turned out to be the actual answer.
SPEAKER_03Okay, yeah, okay, so quite beneficial.
SPEAKER_04Yeah, yeah, and just uh following up on that, and the survey, this surveys that we've conducted, we asked doctors, this was again UK GPs, what they were using the tools for, the highest um percentage of of we had was a whole range of clinical tasks that we we asked them to select, but documentation was was the biggest one. So about a third said they used it for for documentation. And given that in the in England, NHS England, primary care doctors have to share their records now.
SPEAKER_04So they may be, and again, this is brings in the nocebo effects and the placebo effects, how you write the notes, and if you're using these tools to help you write the notes, that might influence those effects. But also as to what you say about diagnosis or differential diagnoses, that was sort of the next big uh use. So I think it was around 27%, so just over a quarter, it was pretty consistent across all three years. And there was another study by the Nuffield Trust which found a similar thing. Um, so about a quarter of GPs using it, which then starts to you know invite the question well, if they're using it and I'm using it, you know, is is it a case of that a doctor has more expertise to use it effectively?
SPEAKER_04Um and I mean that's an open question as well. So yeah, I I suspect people do use it for exactly that. I've used it for that purpose as well. Yeah.
SPEAKER_02It's an emotional use because it's like, uh, I just want to find out what's happening. Yeah. Rather than just sitting there and doing nothing. Yes. Yeah, exactly. Does it help anybody? Absolutely not. As in terms of like being can I help somebody of no knowing a diagnosis? No.
SPEAKER_04Yeah.
SPEAKER_02Yeah.
SPEAKER_04But they can be useful for helping you to sort through and work things out. And of course, the issue will be with increasing online record access. And lots of countries patients have access to their test results, lab results, medical records online, um, including in Scandinavia where I work, um, that they will probably be putting these into chatbots, which is a huge gift to big tech as well. Um, but using it to get more accurate uh probabilities.
SPEAKER_04Combine that with what you're feeling right now, and you might get quite decent uh more accurate outcomes than just sort of asking it directly, like you would talking to Google, so to speak.
SPEAKER_02It does sound like there's a little bit of a difference here between how you describe placebos and nocebos in the traditional sense, because you you described it as a measure, something that we use scientifically if we want to do some kind of experiment. But in this case, it does sound like uh a bit of a side effect from how naturally polite uh things like Chat GPT are. And almost it's almost like, oh, that was a nice interaction. I feel a little a little bit better about myself now. Um so it how does this difference translate into um how we might treat these things differently?
SPEAKER_02Um do you mean treating placebos or yeah, thinking about the issue of a placebo in the medical context versus a thinking about placebo in this context?
SPEAKER_04So I think we should we do, I mean, a large part of my my work has been saying we've got to treat these quite distinctively. Um and that's been an issue in a field of research called placebo studies, which is quite a broad church, it's a very interdisciplinary field, if you can call it as such, where there has been some sort of talking across purposes sometimes. So, and this is where if you've got a bit of philosophical training and you you don't don't mind rolling up your sleeves and sort of point almost acting like a sort of sub-editor, well, you've missed this, but this is no, you're not using this term consistently.
SPEAKER_04Some of it has been that's what we're here for. That's what we're here for, to tighten things up and pick. And you know, when you in an particularly in an emerging field, which placebo studies, it's it's quite a young field of research, uh, studies into the placebo effect and nocebo effect, you can do quite good work. Uh and uh so I certainly saw a role for myself alongside other philosophers like Jeremy Hayek, uh uh who is he was at Oxford for a long period of time. Um, but basically the yeah, so it's it's differentiating what we mean.
SPEAKER_04Uh and one of the things I have uh, you know, get into the nitty-gritty of it, is there's defin also a distinction between placebo responses and placebo effects as well. And there's been a lot of conflation between those in research as well. And that again sounds sort of like narcissism of small differences, but it actually makes a big difference to measuring uh to to saying what we really mean by these effects and what their potential consequences can be for patient care.
SPEAKER_03So uh what is the difference between a placebo effect and a placebo response? And please tell me how narcissism plays a role in that. That sounds quite interesting.
SPEAKER_04No, so basically, well, I mean by narcissism small differences, just that it just seems like a silly, trivial difference, but it isn't really. So when you if you take a clinical trial, a randomized controlled, placebo-randomized controlled uh uh trial, um basically what happens when you allocate people to the the treatment that's under scrutiny, or indeed the placebo, people will uh respond when they're when they're under investigation.
SPEAKER_04So there's a lot of noise basically in clinical trials. Uh natural history, some people are going to feel better anyway. Um there's such a thing as the Hawthorne effect. So if you know you're being studied, you you might behave a bit differently. Responder biases, so if somebody comes around in a white coat and a clipboard, you might respond in a more people-pleasing way. Um and then there's also genuine placebo effects whereby you start to feel a bit differently because you expect it to feel better, or nocebo effects that you think, oh my god, this thing has side effects or whatever it may be, and you start to feel worse.
SPEAKER_04So there's a lot of noise in clinical trials that needs to be screened out, and that's really the purpose of a placebo. So you're trying to mimic everything about the treatment under scrutiny except what you think makes a difference. So that's the issue. And if we don't so so, really, a clinical trial a placebo there um is trying to screen out that noise so that we can make a good inference about what is happening, if anything's happening at all.
SPEAKER_04So all of that noise can collectively be called placebo responses. Um, it's sort of undifferentiated, and it's really the response that happens after you've been given a placebo. And if you've been given the treatment under investigation, you're going to get a sort of placebo response to, as sort of this in undifferentiated uh response to the treatment. As well. So that doesn't mean that it's all necessarily as a result of expecting expectancy like placebo effects.
SPEAKER_04So that's where the difference is important. And I think what tends to happen is that researchers look at clinical trials and they see, gosh, there was a very large placebo response here. Ergo, there was a very strong placebo effect. But that's it's not the same thing. And this is where we have to be careful because teasing out and and actually conducting experiments where you can see the people genuinely start to feel a little bit better because of placebo effects, that's a pretty hard thing to study empirically.
SPEAKER_04And if you like clinical trials are a very good place for discovering the potential for these things, but not for differentiating it.
SPEAKER_02Because we're all nerdy academics, have you ever been in a situation where somebody wants to run a test on you and you're like, all right, I'm gonna freaking ace this test? This is this is uh an autism test. I'm gonna ace it anyway.
SPEAKER_04Yeah, the competitive streak. Yeah, so these things do affect how you're when you're investigative. Yeah, absolutely.
SPEAKER_03Okay, okay. Um I see uh I see how it will necessarily be not easy, but easier to test for placebo effects or responses in uh like pharmac uh pharmacogenic trials, so drug trials, essentially. You can just take out the active compound and give it identical looking pill to both the placebo group or the control group and the the test group. But uh there's more to health than than drugs. There's physiotherapy, there's um there's behavioral therapy, there are there are things that it will be very difficult to test for without doing the thing.
SPEAKER_03Like it would be very difficult to test if physiotherapy works if the physiotherapist comes into the room and then doesn't do anything. Yes. And so I was wondering about that and whether or not that's any different to AI or digital health tools.
SPEAKER_04Yeah, it's really difficult to design placebos, and this is where I think too often um clinical researchers focus on um uh even very broadly in the medical community, people focus on the placebos as a type of thing like a sugar pill. When it's it shouldn't be. If you're really uh concerned with using this as a measuring instrument, um, you really, as I say, have to copy what it is that it's gotta almost, if it's in pharmacology, it should look like the the pill under scrutiny, it should ideally taste like it as well.
SPEAKER_04There might be some similar side effects. You've got to dispense it, administer it, deliver the tool in the same way, so the same number of interactions with clinicians, those also have to be um uh measured and made sure so this should be double blinding, ideally, here, so that that uh psychosocial aspect doesn't interfere with the clinicians' expectations, subtly augment the patient's uh expectations. All of those things can make a difference. So, even in pharmacology, it there are times there's a huge debate about this with antidepressant research, whether the controls in antidepressant research were really any good.
SPEAKER_04So Irvin Kirsch was a researcher there who did a health psychologist who did a lot of work and suggested in fact antidepressants are not more beneficial than active placebos. Now, I mean that's so that's a whole body of controversial research, but you're absolutely right when it comes to other kinds of treatment modalities with psychotherapy, is one, physiotherapy, surgery, designing these in ways that are ideally don't allow the the potency, the influence of an expectation to leak in, um that's challenging to do.
SPEAKER_04Nonetheless, it's important to try to do. So if you look at, for example, psychotherapy, um there is excuse me, there are certain things you can do. So, what we're looking for, if you like, is some sort of structural equivalence. So, if you think that do particular techniques in therapy are important, um, you might say, Well, we want uh patients to be allocated to two groups, they see the uh therapists as frequently. Um aspects of the treatment are pretty similar.
SPEAKER_04Again, it's really hard to blind these things. If you're a therapist, you know what you're you're delivering, but um still nonetheless it's important when you do the more structural equivalence there is, the more that there there are is it a level of isomorphism between the treatments and where the if you like the engine of treatment, that thing that you think makes the difference, you can see even in psychotherapy that um there are it very often that it's uh how a treatment is delivered, it is patients' expectations, it is the therapeutic alliance that may actually make more of a difference in some of these specific techniques, and that's important um because we're discovering that specific techniques may not always make a key difference.
SPEAKER_04So it's harder to do, but that as I say is not really an excuse not to try to do it because ultimately we want to know when we're delivering treatments and when we're be to be ethical about disclosing the treatments about what actually works and giving a sort of informed consent, you want to be able to do that uh honestly and try and uh truthfully, so that the best way to do that is still uh deriving uh evidence using uh placebo controlled trials.
SPEAKER_02So going back to the role AI plays in placebos for a sec, um are you concerned about anything in particular when people talk about oh, this AI system is closing causing a placebo? Is there any particular way in which you know some kind of uh loose language or um conflation might be concerning?
SPEAKER_04I mean, I guess I would say uh probably we'd want to say it's causing could it could AI be causing placebo effects? So I'd want to add placebo effects on onto the end of it. Um and then that becomes a really interesting empirical question. Is there something about the interaction um between the patient and the AI, maybe given the same sort of in uh inputs and out inputs from the patient side? Do they actually start to feel, as you say, you know, do you start to feel a little bit better because you think that this tool is quite authoritative and it's knowledgeable and it somehow gets me, um, but it also gets what's going on.
SPEAKER_04Uh and that as some researchers have said is a good shorthand for understanding how placebo effects can arise. That the the doctor, the clinician gets it, it gets the disease, and so so that the competency is is seems to augment placebo effects, but also getting me empathy augments it. And these two, there's no I I'd sort of say theoretically at the outset, you'd say these tools could really be influencing those those effects. Um patients may just uh and as they use these tools more and they get more comfortable with them, potentially they this this could be happening, and more consistently so than in uh human interactions, possibly, but where it can be very difficult for doctors and other healthcare professionals to demonstrate empathy all the time.
SPEAKER_04They're multitasking, they're sometimes they're working behind computers, and also it's just empathy is sort of a skill that that people consider a soft skill, but it's really hard to get right. And it can introduce bias and so on into care too, because doctors aren't always as empathy, well, humans aren't uh empathetic to they can be biased in how they did they they deliver empathy, if you like, um, where these tools may demonstrate more consistency.
SPEAKER_03One of the things uh that's sort of coming to me at the moment is that if the purpose of therapy or mental health treatment is not necessarily measured by like a physiological outcome, like better heart pressure or sorry, blood pressure or or heart rate, it's measured by how the person feels about themselves or how they perceive the world. And so if the person just by placebo feels better, is that not the same as having an actual effect? If the goal of therapy is to just feel better, yeah.
SPEAKER_03Oversimplifying it again.
SPEAKER_04Yeah, so uh yeah, exactly. So if you you can ask people how do you feel afterwards? Um, do you feel better or worse? Um but uh even so I I think we'd still want to say that we'd and here's the point of on the rationale but behind a placebo-controlled trial is that you still want to be able to say what caused the difference. So you're right. I mean, ultimately you might say it if somebody is even given a sugar pill and they start to feel better, that's that's not uh trivial.
SPEAKER_04Um, and it might, as as some researchers have said, well, in fact, when you do these experiments and you tell people I'm giving you a placebo when it is a sugar pill, I think this will make you feel better, and in fact they do feel better, that they've argued that you know we should that there's a space for that in healthcare. But what's but where I've pushed back against that is that you know there's nothing wrong with that being the case, but actually a lot of the studies which have investigated these so-called open label placebos, for example, haven't been terribly well designed.
SPEAKER_04So the point being it's not really clear if what we say makes it that the placebo in this case actually does make a difference to care. So this is where this research gets really thorny and intricate because and it's the same in psychotherapy, you might say, look, if it makes a difference, people feel better, that's good, you know, that's all that matters. Yes, but at the same time, what also matters is that we decide at scale whether treatments are worth rolling out, and if we don't know if that they make, and this is where the design, the methodology is really critical, because it has these downstream consequences for what we recommend.
SPEAKER_04The other aspect of it too is um the investigation of harms into psychotherapy, it's not all that extensive. So there have been a few studies, and and uh Scott Lilienfeld was a uh a cognitive scientist um who he's since passed away, but he was really one of the few pioneers who looked at the potential for harms in psychological treatments, and he estimated about 10% of treatments led to patients feeling worse, but these aren't routinely uh explored.
SPEAKER_04Um and what's interesting about investigating AI as a sort of uh modality or as a uh a method of delivering potentially uh treatments is that people to tend to hold it to a higher bar. Now everybody's like, God, God forbid that people use these sorts of tools because they could be harmful. Whereas, which is a good thing, it's good to be cynical, but uh uh and keep an open mind about the benefits and harms.
SPEAKER_04But what's interesting, I think, is when we look at psychological treatments, we haven't had that same uh apathy or or worry, um, possibly because we have an assumption that talking doesn't really lead to that, it can't possibly lead to that many um uh that much scope for harm. But in fact, there is some research that it does. Not everybody does well after psychotherapy, um, and that again is going to depend on uh could be therapist factors, patient factors, and what what they consider to be useful in therapy, whether they're not convinced it's useful and they think, God, all therapy sucks if this doesn't work, or there could be a lot of things going on.
SPEAKER_04Um, so I find the parody that the the the uh the the distinct how we hold AI to a whole higher standard, I think, is interesting in itself and quite illuminating.
SPEAKER_02That's what I keep telling you. Stop trying to get me to go to therapy, man.
SPEAKER_03Like a year ago, James and I had a massive debate at indoorslash argument where I was like, no, like I do firmly believe that everyone should go to therapy at least once in their lives. Like, I do think that there are things that a therapist can help you with that it's not either your friend's responsibility or that your friend has the professional capacity or bandwidth to deal with with you. James is like, no.
SPEAKER_02I had an argument. I didn't just say no.
SPEAKER_04Yeah, it says but it is quite interesting because a lot of people drop out of therapy. Um, so something like 20% of people drop out after this the second session. Now that's inter the attrition rate is quite interesting too, and it would be really interesting to see if um comparatively, I mean, if people can tend to get it here's the issue, they may get too addicted and too dependent on chatbots as well, but people may be depending them on these tools for emotional support in ways that are similar to what you might depend on your friends for as well.
SPEAKER_04Um I have uh an article under review, I don't know what I should say, but basically the the idea that um that the that chatbots can be obsequious at times and they can be people pleasing. Friends can be people pleasing too. Friends don't always tell us there's literature on this, they're not always going to tell us, you know, their bum looks big and something, or you know, we've got you know, leave that partner you're with, they don't like him, or they people choose the timing when they tell things, so they're not always upfront and honest.
SPEAKER_04So you're right, a therap, a good therapist could tell you in a gentle way some home truths. The qu interesting question is can you use these tools to prompt you to do something similar? So there's this sort of um blurring, if you like, between what these tools can do, could you know, and how they can do things better.
SPEAKER_03This might be a little bit telling, and I wonder if this says more about me than about um about these chatbots. But I keep getting these ads on Instagram for uh these chatbots or like these tests that you can do to be like, am I the problem? Am I a narcissist? How toxic are you? And I'm like, Jesus Christ, like how do my cat reels and swimming memes like collaborate on identifying if I'm a narcissist? So yeah, I'm cognitively pointing out some home truths. Yeah.
SPEAKER_04Oh god. Well, no comment on that.
SPEAKER_03If any of the listeners are in the same boat, I get you.
SPEAKER_02I'm sure many of you will get it.
SPEAKER_03Hey James, thank you. Thank you. I feel so supported by our friendship.
SPEAKER_02Uh I try. You see, you see, you you know what uh Charlotte was saying about how you know friends are often obsequious people pleasers. You don't have that problem with me. I got you, man.
SPEAKER_03So um my next question is if it's extremely difficult to figure out if digital health tools and like these online chatbots like ChatGPT or the more sp the more specified ones like Therobot or or Rosa, for example, Rosa is a genetic counseling um uh uh support tool. If we can't really test them in that same placebo control way, how can we really be sure that they're like beneficial, that they're actually doing good?
SPEAKER_04Yeah, uh exactly. Um and I think that what one of the the an article I wrote with uh Leo Druart, uh French researcher, was that we it was in uh Journal of Medical Ethics Practical Bioethics, and we argued that in fact we really should be striving to do um placebo-controlled research here as well. Just because it's difficult doesn't mean we shouldn't try to do it. Um and the issue with Therobot was very interesting because this was um a tool that got a lot of traction in the media, um, and it was published by this one the study was published by New England Journal of Medicine AI, so really prestigious journal, but the control in this case was treatment as usual or wait list, and that's not the that's not effectively weighing out all of those things I talked about earlier, you know.
SPEAKER_04Is it the uh responder biases, is it uh natural history, is it um uh Hawthorne effects, um placebo effects as well. So, what we don't really know is does Therabot actually make a difference to pe to people um outside of a clinical trial? That's really critically important before you you start recommending these treatments at scale to people. So there are ways in which you could try, as I said, you know, to give something that was structurally equivalent there.
SPEAKER_04That you're getting patients to do that is dispensed, that they're using the tune maybe the same number of times uh per week. Uh, there's maybe some sort of homework attached to it, whatever. There's something that the patient has to do, um, where the the interface is sophisticated looking, so so that they can't break blind. Um, and of course, asking patients as well, do you think you've been allocated to the treatment or or the control at the end of it would is important as well.
SPEAKER_04Um so I don't think we should give up on that task. Uh it it is really important it in order to discern what is it about these tools that might make a difference to people's lives, or or if it's not making a difference, you know, we we've got to know that.
SPEAKER_03A follow-up from that then, I know that there's like uh a lot of ethics around withholding drugs from patients where you have good reason to believe they would be life-saving or really beneficial. Um, it's kind it's unethical to withhold a treatment that you think will improve them just for the sake of the study. Or or so so some people ask. But is that really the same case with digital health tools? Where you're right, we may not be able to know for sure whether or not it's beneficial, but it's clearly not as harmful to withhold it. Or is it?
SPEAKER_04It could be that well, this is the issue. Um so some of these tools may be harmful. Uh, and this gets into the point about we need to have a much stronger evidence base for behavioural health interventions in a whole range of guises, whether it's uh humans who are dispensing them or uh chatbots or other kind of digital fora. Uh you've I think we do need more research into the harms there as well, and we just don't have enough of it. That's just been a complete blind spot in the in uh behavioural health.
SPEAKER_03So there is a potential chance that using ChatGPT or Therabot is just turning everyone into narcissists because it's a massive people pleaser and it's like you're right, everyone else is wrong.
SPEAKER_04Could everybody's walking around with big heads after they Is that anything new though?
SPEAKER_02Got him. Okay, so like given all these problems, um, so we have it it sounds like we don't even know for the most part whether placebos work in the first place, uh, because a lot of the research, as you said, is kind of dodgy. Um not your words, mine. Uh and then on top of that, we have to figure out whether uh things like Chat GPT or other kinds of uh AI tools are good placebos when we don't even know if the original placebos work in the first place.
SPEAKER_02So, what would progress in this field uh look like in figuring out what how whether AI systems work?
SPEAKER_04Um so it would be so it takes were you taking the example, say, of chat whether Chat GPT or similar kinds of tools are beneficial to people's health in certain circumstances? Is that what you mean? Yeah, uh in psychotherapy or I mean we could do comparisons with uh we could do sort of what are called horse race studies where you could compare uh different versions of of generative AI tools to see do patients feel better or worse.
SPEAKER_04You can where you could control somehow what the output looks like. You could design it in such a way that they aren't people pleasing, um, that they don't demonstrate empathy in their outputs. I mean it's going to take somebody pretty ingenious to be able to work how we do these, but it there are ways in which we could establish if these tools do make a difference to people. Um and you we could also investigate how long you're spending online. So it could be the same tool, but if you spend too long on it, does this actually make you feel better or worse?
SPEAKER_04Um, is it displacing some real-world activities? Um, I think there's plenty of experimental work that could be done to to identify if and when these tools are useful and for whom.
SPEAKER_03What if instead of framing these tools, like Therabot clearly has the word therapy in it? What if instead of considering these tools as therapists or replacement psychologists or counsellors, and we also don't consider them as like friends, we think of them as this new space for self reflection. Because then it would seem less like you know, encouraging more people to self reflect in a greater variety of ways throughout their daily week would be probably 100% more than what some people do already.
SPEAKER_03So I wonder then if withholding these tools or trying to say, well, we're not sure. About the benefits and we're not sure about the harms. Is there a chance though that by withholding them we're also making people who could have been better off actually worse off?
SPEAKER_04It's possible. Um, and I think it's going to be really hard to, in a sense, withhold them from people. People are, you know, the sort of real politic of it is people are going to use these tools. Um, and there's very little that healthcare professionals can do about that, particularly as you're saying, it would be a bit level of hypocrisy if they're to say that you patients can't use these tools, um, meanwhile, they're doing the same thing. Um, so but it could be as you say that uh, you know, if people one of the reasons people may be using these tools is they don't have access to uh rapid access to healthcare for a whole variety of reasons.
SPEAKER_04And there were some quite interesting studies by the Kaiser Family Foundation in the US which found that it looked which looked at demographic, patient demographics and trust in these tools and whether they thought that they were useful. And in fact, it found that black Americans and Hispanics were more likely to find the tools uh trustworthy and useful. Now that's could be out of date by now than but compared with whites, that's interesting, but it might reflect a lack of access, rapid access to care, it might reflect uh some differences when you do get access to care.
SPEAKER_04You may be more sceptical about what doctors say or more anxious. So it's gonna always depend on what you've got and compared to what, um, and that's gonna be different for all kinds of patients as well. I mean, I've got one thing, I've got uh rare illness in my family, I've got a twin sister with a rare illness, she doesn't mind m me talking about it. Um, but she um when she goes to speak to her GP and she was she had a delayed diagnosis for many years, she's she knows sometimes when GPs are trying to save face with her and they don't know a lot about it.
SPEAKER_04And in that kind of scenario, but by the way, when she speaks to her specialists, she has a lot more confidence in them. She only sees her specialists once every year. Now, if you're using these tools in area, and this is one of the interesting issues, people with rare diseases, and we did a very small-scale survey on this, may be deriving benefits because these tools are able to uh keep up to date pretty rapidly. They're very rich information sources for patients. Um so, again, for some patients, there could be a lot of benefits here, or they they may just view them as even if they don't think the the tools are perfect by any stretch.
SPEAKER_04And in our study, patients and family members of those with rare diseases, they they weren't completely bowled over by the accuracy of these tools, but they did find them useful, um, and it's the difference. It just needs to be uh superior or considered inferior. That's all that matters, you know. So perfect's the enemy of the good. These tools may be offering something to people uh that we shouldn't be dismissive of.
SPEAKER_03Have either of you used one of these tools before? Like the digital health ones, like Therabot or ChatGBT to do some counselling?
SPEAKER_04I have used it for advice occasionally, um and it can be it it not extensively, but they are so polite that it it does win you over. Um but I did recently speak to somebody who said she used it to uh speak to family members um on her WhatsApp group. She said her siblings were for some reason she was too in uh her communication style wasn't being entirely responsive with her family members over some critical issue, and she used the tool and she said she she she had tears in her eyes with the level of support that she sort of garnered from the use of the tool, and it actually made a difference.
SPEAKER_04It was suggesting what she should do differently, in quite gentle terms, and she said she did it, and she said it had really made a difference. So I think there's pop that's probably illustrative of you know many people's experiences, we just don't know enough about it, and by the same token, you know, the people-pleasing thing can be an issue. These tools don't they lack context. So if they're giving you um advice that does not understand fully what is going on in your the interpersonal life and and and other it can't take on the perspectives of others, and incidentally, therapists suffer from this too, because they've only got you your side of the story.
SPEAKER_04Um this could this could be a problem for the advice they're disp dispensing, and this is where it gets tricky. Have you used it?
SPEAKER_02Uh I tried using it to teach once.
SPEAKER_04Oh yeah?
SPEAKER_02No, not I didn't use it in place of me. I was like, this is a demonstration, so explore what you know AI and trust means. Uh and then it's saying weird shit very quickly, and I got the I got weirded out, the whole class got weirded out. I think it was a mistake.
SPEAKER_01Yeah.
SPEAKER_02It was replica.
SPEAKER_03No, I I try I went to use it once because um NUS has our uh like mental health app and in there is a um like AI that you can talk to. And I got to like the third question and I was like, oh, this is just gonna take so much time. And I just I was just there was just something weird that I didn't like and I just I just shut it down and haven't used it since.
SPEAKER_04Yeah.
SPEAKER_03But I think um, I think I'm gonna start getting into using like cord and perplexity and stuff, just that whole guidance thing, like how to communicate to others when they're not really getting it. I know that my husband, um, and I'm gonna say this because I know that none of his family members listen to this, but he doesn't really, uh, he struggles to uh know what to say to his his uh grandmother who's deeply conservative and says some weird stuff sometimes, and so she'll send massive messages and he's like, I just can't be bothered. So he plugs it into the like chat GPT and it generates a reply and he copies and paste and sends it back.
SPEAKER_03You can do this with family members, you don't want to talk about it.
SPEAKER_04I don't think that's uncommon actually. Um, and I think that one of the use I've noticed like in academia or you know, with colleagues, sometimes you know, if you want to send an email that you think you don't want to be sitting over an hour working out exactly what in that sort of soft diplomatic way, well, how do I write this? They can be really useful for that um for editing. And I I've noticed a lot of my colleagues' emails have got extremely slick compared to possibly mine as well. But um, they can be very useful for for getting tone right, and I also suspect that's why healthcare professionals may be using these tools as well, when it comes particularly in the era of online record access, where you can just write letters or or explain things in an accessible, uh, sensitive way.
SPEAKER_04Um, they could be very, very useful.
SPEAKER_03Although I did see on someone's email chain, you know, after their name, their sign-off, it says, I have an existential issue with AI. So if you have used it, please let me know so I can never contact you again. I would never be doing that.
SPEAKER_04That's extreme that this individual could end up isolated quite soon.
SPEAKER_03Yeah, just gonna be by themselves. Anyway, that's all we have time for today. Thank you so much for coming on the podcast. You are welcome anytime.
SPEAKER_04Really enjoyed it, thank you.
SPEAKER_00Thanks for listening to our new podcast. Subscribe online and tune in every first of the month. In the meantime, listen to CBME and U and be savvy.
Send us Fan Mail We spend much of our lives searching for meaning. Some of us look for it at work. Some in our families and friends. Some look for it at an overpriced spiritual retreat in Bali where some dude named...
Send us Fan Mail In some insurance scheme, you might be required to pay more because you are more prone to disability. In another, you may have to pay more because you're a woman (and are statistically likely to live...
Send us Fan Mail Did you know that genomic sequencing can now screen newborns for hundreds of rare conditions right at birth? That looking too closely at our DNA might actually cause more anxiety than answers? And is...
Send us Fan Mail Did you know that over 80% of people who give birth experience an injury? That anywhere between 30 and 50% of people who give birth find it traumatic? And that up to 15% of people experience PTSD from...
Send us Fan Mail Could a blob of cells in a petri dish one day think? Feel? Have rights? In this episode, Professor Insoo Hyun (Harvard Medical School, one of the world's leading experts on the ethics of stem cell and...