Ep. 170: AI: Does it really concern neurologists?

00:00:00: Welcome to EANcast, your weekly source for education, research and updates from the European Academy of Neurology.

00:00:15: Hello and welcome to EANcast, the weekly podcast of the European Academy of Neurology.

00:00:22: My name is Rachel Wurm.

00:00:23: I'm a member of the Communication Board and the Editorial Board of EAN News, as well as a founding member of the Task Force on Artificial Intelligence.

00:00:32: Today's special topic is the quite provocatively titled AI, does it really concern the clinical neurologist?

00:00:39: And with me today is Professor Francisco Cardoso, who is calling us from a very early morning in Brazil.

00:00:47: He is a specialist in movement disorders, having served as the past president of the Movement Disorder Society and a basic clinician.

00:00:55: So he sees patients every day and we're very interested to hear his take on artificial intelligence.

00:01:01: Thank you for joining us this morning.

00:01:03: Well, thank you very much, Raphael, and also thank the European Academy of Neurology for this invitation.

00:01:11: It's a really treat to do that.

00:01:14: Perhaps I could start saying, as you have already mentioned, that despite being involved with academic life for three decades, but I remain busy as a clinician, including in the academic setting and also in the private setting, so seeing patients.

00:01:31: So my views of a potential impact of artificial intelligence, AI, they are based on these daily practice of my professional life.

00:01:49: Yeah, I think that's a very valuable point of view.

00:01:52: At the European Academy, we have founded this task force that is chaired by Alisa Karoni and Maria Chiara Malgutti.

00:02:00: And we're interested obviously also in understanding how people use it.

00:02:04: So let me just start off with the first question is, do you use any sort of artificial intelligence in your daily workflow?

00:02:11: And if so, what is that?

00:02:13: That's a good question.

00:02:14: When I was reflecting on our conversation today, I came to conclude that I can really approach AI in different settings.

00:02:27: So if I speak about my life as an investigator, as a researcher, it's extremely useful.

00:02:36: So that's really now practice to resort to AI tools.

00:02:44: So that's very useful and helpful.

00:02:47: It really speeds up process.

00:02:51: For instance, if I am preparing a presentation, if I want to use software, so it makes it much, much easier.

00:03:00: So I find it very helpful and available.

00:03:03: But a second point is in terms of seeing patients.

00:03:07: So the role as a clinician, I think it's extremely limited, at least nowadays, because I was talking to my medical students, I'm also... seeing them and liaison with medical students near the end of the medical school on a weekly basis.

00:03:28: And it remains the most difficult task for us clinicians is how to get a proper history.

00:03:36: So this is really the fundamental starting point of making a reasonable diagnosis, a hypothesis, and of course, designing management for these patients.

00:03:51: and at least right now I do not see really a major role for AI helping me or helping clinicians in general to get this history.

00:04:05: It's quite fascinating really to see the mental process of everyone, even physicians.

00:04:11: I have many patients who are physicians, of course, all of them trained really to get a proper history.

00:04:16: But even our colleagues, ourselves, whenever we are telling a physician a history, we tend to embark on hypotheses that are not really based on facts.

00:04:29: It's quite fascinating.

00:04:30: And I realize that if an AI interface takes these pieces of information on their face value, this would be very helpful.

00:04:43: It's quite likely that AI will come to a conclusion, which is not really the most appropriate for that particular person.

00:04:52: So I think that in this set, and at least presently, the role of AI is quite limited.

00:04:59: It's fantastic as a tool to retrieve a huge trove.

00:05:04: of information that probably our human brain cannot handle that in an expedite manner.

00:05:11: But whenever it's to create a hypothesis, I find it much more limited nowadays, actually.

00:05:20: You raise a very good point in that this also, I would say, belongs or pertains to the human connection that you built with your patients and you need to treat everyone differently also in the way that you take their history.

00:05:33: I wonder... Did you think about any sort of clerical or not really medical but still performed by physicians' tasks that perhaps you could outsource if you will to an artificial intelligence or something like that?

00:05:48: Oh yes, absolutely.

00:05:49: But just getting back to this point of human bondage, which I think it's really the core of being a clinician, it may well be the case that younger generations will not mind interacting with some sort of AI interface.

00:06:07: But at least speaking about myself, I wouldn't really like to have this cold, impersonal relationship.

00:06:16: As you have said, I think this is fundamental in clinical practice, despite a given disease having somehow a similar... profiling terms of symptoms, signs, and so on.

00:06:31: But every human having that particular condition behaves in a different way.

00:06:37: And this person deserves a different approach.

00:06:39: And I cannot imagine, again, at least currently, a way of having AI dealing with these subtle differences among our patients.

00:06:53: But getting back to this clerical point, absolutely, because there are so many tasks that really take up a huge amount of time, time that we could spend talking more with patients, or if we are in academia trying to develop new trials, or reasoning, or these were some things.

00:07:15: I think there are many things that could be outsourced to AI.

00:07:19: One of the issues that I deal on a daily basis, both as head of my unit and as a clinician as well, is the paperwork.

00:07:28: And many of these things, they are rather automatic in a way that we just need to replace a few pieces of information.

00:07:36: And I can foresee, as you have alluded to in your question, that this is something that it's likely that it will make our lives easier and lighter, I must say, as a physician.

00:07:52: Yeah, when I think about my own practice, I find it very curious how often I will get some information that is coming from a database.

00:08:00: So they have printed it out and then I have to enter yet again into another database.

00:08:05: So there's lots of, I would say, shuffling around data that we could certainly automate.

00:08:10: And I think the most important part is that I hope at the end of this process how I will shape out to be and I think it will certainly impact our lives, right?

00:08:19: I think there is no denying that this technology is going to enter into our daily lives and our work lives.

00:08:26: I hope that it does in fact give us more time to spend with patients and not just compress and compress so that we can see more patients.

00:08:35: And in that sense, I also think that we should not be too much afraid about this.

00:08:41: I think if we use it as a tool, sometimes I get the sentiment that this is something that's going to replace us.

00:08:48: Radiologists are alarmed.

00:08:50: Will they still be needed?

00:08:52: I think that's not the way it's going to shake out personally.

00:08:55: I think it's going to help us, it's going to support us, but in the same way that airlines are still flying with two pilots, even though the airplane is piloted like ninety percent of the time.

00:09:07: by a co-pilot.

00:09:09: Is your hospital or is your provider taking any steps towards integrating these tools already?

00:09:14: or what would you think?

00:09:16: What would you suggest that they do?

00:09:18: Yeah, that's a very good point really in terms of this somehow generalized panic perhaps.

00:09:27: It's even worse than fear that physicians will be replaced by AI and I share your feeling that this is unlikely for many reasons.

00:09:40: Probably the most important one is that we have discussed in one of the previous questions that humans crave for human contact.

00:09:50: So that's really something very important and fundamental in this discussion.

00:09:57: But the other one is that there is some time of flexibility in approaching data that currently the AI models they lack.

00:10:06: Even getting back to this example, which is somehow unrelated to neurology in general, but radiologists, I was reading in the Financial Times a few months ago that despite the effect of AI in detecting patterns in imaging studies, but the number of radiologists who have lost their jobs as a result of these technologies, basically zero right now.

00:10:38: For many reasons, one of them because once a given finding is reported, this needs to really be... Considered in the context of clinical findings in talking to colleagues and so on.

00:10:52: so i i i agree i think we should not be afraid of this technology because it can be very helpful to us.

00:11:00: as we have already discussed speaking about my own.

00:11:04: environment.

00:11:05: I actually work in an academic hospital, which is part of the Brazilian National Health System, which is public.

00:11:14: It's universal in the sense that everyone who lives in Brazil has access to that.

00:11:19: And one major step that has really made our lives much easier is the integration of some basic AI tools, but to generate requests for medications for instance because patients they receive a prescription from us but in the past they would have really to fill out using their own hands a lot of paper really to request these medications from another sector of the national health system.

00:11:51: so this is now all integrated.

00:11:53: so this is a very good example of improvement of the quality of life of physicians and also patients, they are happier with that really.

00:12:05: Yeah, I think you give a very good example of something that also if we as the professionals are approached in this conversation is I think something we should push for is give us these tools to make our day easier, right?

00:12:20: We don't need the most sophisticated algorithms that would integrate, you know, seven hundred different variables from your patient to calculate a specific risk or something.

00:12:30: I think we can do this when it's necessary and also intuitively.

00:12:34: that's our job.

00:12:35: But there's so much clerical overhead that right now is being done by physicians, I would say in all systems, that you could get so many returns on this investment if you put AI to use as this.

00:12:53: like to switch the sites now a little bit, because something that we're also interested in in the task force and I'm personally interested in finding out more about is, in your experience, did you have patients that were using AI as a first point of contact with their medical problems?

00:13:10: If so, what's your experience with that?

00:13:11: And what do you envision that will be like in the future?

00:13:16: That's something that I think it's actually more frightening, perhaps, because I can recall a recent example, someone who actually has a tardive dyskinesia, a gentleman with a long-standing history of psychiatric illness, treated with dopamine receptor blocking agents, and he has a classical tardive dyskinesia.

00:13:42: So he started actually chatting with chat GPT, but providing biased information.

00:13:52: As a result of that, and it's of course understandable, so chat GPT came up with conclusions which are completely unrelated to his condition that in fact have led to a major source of anxiety in these gentlemen.

00:14:09: One of them that somehow chat GPT developed the hypothesis that he lacked an optic chiasm, which of course we know that he has nothing to do.

00:14:23: It's a pity that you guys who are listening to our podcast, you could not see Raphael's face when I mentioned that because it's so really outlandish these hypothesis, but in any case, so that's an unfortunate example of a misuse of AI.

00:14:42: My patients right now, because even if you have a smartphone, of course, you can have access to chat GPT and to come up with ideas about diagnostic hypotheses, management and so on.

00:14:56: That's something that we must be aware of and to have perhaps a proactive attitude with our patients, warning them that it's okay to use AI in this sense, but they must really bring back to us any pieces of conclusions that they have in order to discuss, because sometimes they can be quite misleading.

00:15:22: But I suppose that with the easy access everywhere in the world, this particular scenario will become more and more frequent in our professional lives.

00:15:36: Yeah, absolutely.

00:15:37: One of the things that I suggest really everybody should do is take five minutes and get yourself in the shoes of your typical patient, whatever that is, whatever you see in your clinic.

00:15:47: And from their point of view, go to chat GPT or any other, doesn't matter, the chat box and start the conversation and see where this leads you.

00:15:56: Because I feel like it is a bit different from the way I think we all, I mean, everybody was aware of patients that would Google their symptoms, right?

00:16:03: And that usually led you down one or two paths.

00:16:07: They were usually very dramatic, but you could easily tell when somebody had been doing that before.

00:16:12: And now I feel like with the way these chatbots interact with you, they also give you so much information.

00:16:18: They'll be like, well, do you want me to prepare you for your next clinic visit?

00:16:22: And then you'll be, yes.

00:16:23: And then they'll be like, well, these are the seven tests that your physician should look for.

00:16:26: And probably they will come with an expectation of some tests that might or might not make sense.

00:16:32: And this, I think, can shift the dynamic.

00:16:34: of the whole visit quite dramatically.

00:16:38: And there's nothing we can do about that right there.

00:16:40: As you said, everybody has access to it.

00:16:43: It's free for the most part.

00:16:45: This is going to be used.

00:16:46: So I think the only way that we can prepare for this is to use it ourselves, see what happens.

00:16:51: And then, as you said, open up the conversation and be like, it's cool that you used that.

00:16:56: No problem.

00:16:56: Just tell me what did it tell you so I can give you my professional advice on that.

00:17:03: Yeah.

00:17:03: Yeah.

00:17:04: And if I can add something in that line, there is something related also Rafael to language, because the size of the database in the Portuguese language, which is the native language of all Brazilians is quite limited in contrast to English.

00:17:22: So English has really much larger and more powerful.

00:17:26: And my patients, they in fact interact with AI using Portuguese.

00:17:32: And this is a source of bias as well.

00:17:36: I can recall one interesting example, a colleague and friend of mine from the US, he came down to Mexico for a movement disorders meeting and he decided to use chat GPT in Spanish to come up with ideas about one of his talks.

00:17:56: And some of the pieces of information, they were completely really wrong.

00:18:00: because of this size of the database.

00:18:03: So when he posed the same questions in English, his native language, it really raised the standard to a very nice really type of presentation.

00:18:14: So that's something that in Europe with so many different languages, it's something that we should be mindful as well.

00:18:22: So how people are interacting in terms of the language that they use, I suppose that most of European patients will use their own language, native language rather than English to pose questions and to come up with this interesting example that you gave someone preparing themselves to the next visit.

00:18:46: Yeah, absolutely.

00:18:48: I think it's always important also for clinicians to have at least a basic understanding of how these... these machines actually work, right?

00:18:58: We can't expect the general public and we can't expect the patients to have this in-depth knowledge, but knowing that this is something that is actually just generating words out of what's the most likely next word.

00:19:11: So it will always go down the route of the more likely and probably one of the niches I think for physicians will always be to also be trained and be aware of the rare diseases, which will obviously be biased against in the databases.

00:19:25: Same with languages same.

00:19:27: we see this also with the radiology examples that when you train on specific data set in one part of the world it Does not really work as well in another part of the world even on something as simple as chest x-rays.

00:19:40: So I think for for some time being even though you know the AI makers will have you believe otherwise I would say I don't think that that this is gonna be working perfectly or even close to to perfectly for a long time.

00:19:56: This has been a very, very interesting chat.

00:19:59: I wish we could do this longer, but unfortunately our time is running a bit short.

00:20:03: Do you have any closing remarks or anything that you would like to add?

00:20:07: Well, other than thank you very much and Erwin, what you have just said, this has been a very interesting and fascinating conversation.

00:20:16: So many thanks.

00:20:17: Just one point that we didn't discuss, Rafael.

00:20:20: It's related to how to handle genetic data because we are somehow receiving a huge amount of data in terms of genetics.

00:20:31: Exome sequencing is almost common practice nowadays everywhere in the world.

00:20:37: So the prices have come down and so on.

00:20:39: And how to?

00:20:40: really, it's impossible, at least for me, I cannot handle all.

00:20:45: these pieces of information.

00:20:46: So that's a very useful, really a way of handling AI.

00:20:52: So it has helped me really a lot in defining this VOOS variant of unknown significance, if they are really relevant to my patient or not.

00:21:02: But other than that, I just would like really to thank you very much and also EAN for this kind of invitation.

00:21:12: It is now for me to thank you.

00:21:15: for sparing your time on this early morning for this very interesting talks.

00:21:18: There's two final things that I would like to add.

00:21:21: First, since we're talking about the application of AI in a clinical context, I would just like to give the general disclaimer that you should never enter identifiable patient information into any of these tools because this is the same thing as sharing this on the internet basically.

00:21:37: Everything you put in will be used to train models in future and could potentially be retrieved.

00:21:43: So do not do that.

00:21:44: It is against almost any data protection laws, no matter where you work.

00:21:49: And second of all, I want to remind our audience that we are reopening the House Courses Survey on Artificial Intelligence.

00:21:56: Some of you might have already gotten that through the newsletter.

00:21:58: We're very interested in finding out about your use cases, your views on artificial intelligence, and we would very much welcome your participation in that.

00:22:08: With this, once again, thank you very much for joining me today.

00:22:12: And I wish you a pleasant day.

00:22:21: This has been EANcast Weekly Neurology.

00:22:24: Thank you for listening.

00:22:25: Be sure to follow us on Apple Podcasts, Spotify or your preferred podcatcher for weekly updates from the European Academy of Neurology.

00:22:33: You can also listen to this and all of our previous episodes on the EAN campus to gain points and become an EAN expert in any of our twenty nine neurological specialties.

00:22:44: Simply become an individual member to gain access.

00:22:47: For more information, Visit ean.org slash membership.

00:22:51: That's ean.org backslash membership.

00:22:56: Thanks for listening.

00:22:58: EANcast weekly neurology is your unbiased and independent source for educational and research-related neurological content.

00:23:05: Although all content is provided by experts in their field, it should not be considered official medical advice.