 I to every one of you, it is a pleasure to be with you today and to join the virtual conference for computational audiology and to share some of my thoughts around the emerging revolution of AI chatbots in hearing health care. My name is David Swanapool and I'm a professor of audiology at the University of Pretoria in South Africa and I have an adjunct position at the University of Colorado. I'd just like to acknowledge two of my colleagues as we set out on this presentation, Vinay Manchaya and Jan Willem-Vasman, our co-authors with me on this brief article we published earlier this year in the hearing journal. And much of what I'll share is based on this summary of how AI chatbots can play a role in the hearing health care space into the future. So before we take a look into the future of AI chatbots perhaps it's worth just taking a view back into the past. This is a episode of Star Trek episode 53 that came out in 1968, which was called the ultimate computer. In fact this was episode was one of the top ten episodes identified by Forbes a few years ago. In this episode it really deals with the effects of AI. The ultimate computer is a computer that allows, well that can actually fly and handle the entire enterprise and that also handles its security and safety and weapons. And of course it starts out really well but Captain Kirk is concerned about having a computer, an AI run the enterprise and then of course things start out really good but it ends out bad. So it does kind of serve as a metaphor for people and the fears that people have even with a new technology like chat GPT that has come out. The concern about people losing their jobs due to mechanization and digitalization. So there's always the potential and the benefits but there's also the fears that come with it. We did see the dawn of a new era also start in November 2022 when open AI launched chat GPT publicly. Chat GPT is an AI chat bot that can engage in conversations in response to questions from the user. These are so called prompts and they can help generate responses to users questions that are almost indistinguishable from those of humans. So the launch of chat GPT certainly has represented a major technological revolution. One that could change the face of health care as we know it including hearing health care. I think it's important also to note that chat GPT is not an isolated example but it is part of a global race of who can develop the most compelling AI chat bots. Beside open AI, which was mainly financed by Microsoft, we've also seen other large corporations like Meta, Google with their BARD launch which came soon afterwards and even Tencent that have launched similar proprietary products based on large language models. So it is a whole new era that is rapidly expanding, growing and changing. AI chat bots have a very wide application but there's also a particular application for AI chat bots in health care. Mainly the broad trend has been that AI chat bots can increase accessibility to medical knowledge but also potentially affordability of care. Chat bots can provide 24-7 access to health care advice and support and it can reduce the need for in-person consultations and potentially improve patient outcomes. Additionally AI chat bots could potentially provide very valuable insights and data to health care professionals that allows them then to make more informed decisions about patient care. We certainly also need to realise it's still very early days, we need more transparency on the data that these chat bots have access to and that they use to produce their output so that we can be sure over time that we have systems that are reliable and trustworthy in the applications that they provide in health care in general but also in hearing health care. So there was an interesting piece that recently came out in the New England Journal of Medicine catalyst journal which really in essence was an interview with chat GPT about health care in general. But I think it does set a helpful framework to understand the potential opportunities that these large language models offer in health care in general but potentially also in hearing health care. One of the questions was what are the big opportunities for using chat GPT in health care. So there were several opportunities identified. The first is virtual assistance for patient care so chat GPT could be used to create a virtual assistant that can help patients schedule appointments, answer medical questions and that can provide them with information on symptoms and treatments. A second application is clinical documentation, chat GPT could be used to automate the process of creating clinical documentation such as medical charts and progress notes. Also in terms of patient engagement, chat GPT could be used to create chat bot based applications for patient engagement for example providing patients with real time feedback and support of self management and self management care. And then of course there's also the whole field of medical research and education where these tools are very powerful to support both analysis of data but also to allow for interactive educational tools for medical students and also potentially for patient education. If we take a little closer look to hearing health care and how patients and AI chat bots can benefit the field, there is a variety of different potential applications. For example initial screening, they could be a role for the AI chat bots to interview patients initially about their symptoms and their hearing history and then extract and summarize the most pertinent aspects. Recommendations, they could also provide initial self management recommendations for the symptoms or recommend further evaluation or treatment based on those responses potentially. And there's also the educational resources, these AI chat bots are excellent in providing general information but also in providing hearing health information and prevention tips to patients. So it could be an early advocacy and education tool. There's also self management applications where they could provide advice on managing hearing conditions, the use and the management even of hearing devices so the troubleshooting and support particularly in the early periods of time after a hearing aid fitting. Also potentially reminders and follow up communications from the AI system to remind patients to do certain exercises or training modules. Cognitive behavioral therapy programs with specified prompts is another potential tool that could help self management for patients. In terms of co-mobilities and social needs, these AI tools could also be useful to provide additional secondary screening for co-mobilities and evaluation of social needs so that personalized care can be offered and also informed. Teleodeology, ideal of course remote synchronous but also asynchronous support that can be provided by these tools 24-7. And then intervention, want someone has received a hearing device for example oral rehabilitation programs in a way that the AI provides a conversational companion of sorts are all very realistic potential ways in which it could play a role for patients. In providing hearing care. Here is just an example of recent research findings that demonstrate some of the ways in which AI chatbots can engage with patients. So this was a study that looked at and can an artificial intelligence chatbot assistant provide responses to patient questions that are of comparable quality and empathy to those written by physicians. So the findings of the study show that in the cross sectional study of 195 randomly drawn patient questions from a social media forum. A team of licensed healthcare professionals compared physicians and chatbots responses to the patients questions asked publicly on a public social media forum. So the chatbot responses were actually preferred over the physician responses and they were rated significantly higher for both quality and empathy. Which I think demonstrates some of the strengths that these chatbots have in terms of language use and access to information. So what do these findings mean? It just means that the results indicate that artificial intelligence assistants may be able to aid in drafting responses to patient questions. Supporting this self management and also the asynchronous telehealth provision of information and recommendations. And here is just our own example of a virtual audiologist. So Jan Willem-Vasman put this on the computational audiology network website. And you can ask Alan any question about audiology and it will provide you with an answer. So this is an example of how such an AI chatbot could work. Of course this one in reality is just a chat GPT integration so it's still a general model. It's not specific to audiology or hearing healthcare but it does illustrate the potential. And if you ask Alan a question about audiology it's actually surprisingly accurate in its responses. So we've looked at some of the potential applications for the target users being patients. But there's obviously also risks that we need to take into consideration. And this is not a comprehensive list but of course what comes to mind immediately is the accuracy of the information. And are there potential risks of inaccurate and misleading information that's provided by these tools. And what kind of governance and oversight would there be over something like that. And there's also the risk of over reliance on chatbots for decision making. Which is also in healthcare a very important factor to consider. There's also concerns about the potential loss of human touch and emotional support. Which I think is usually something that is raised as a potential concern even in telehealth. Although I think lots of the evidence is showing that that can be overcome. And then there's also of course with any digital technological service there's the risk of technical issues or difficulties with communication. And there's also important research priorities that have to be developed and also pursued to ensure the safe and judicious use of these chatbots. So things like looking at the efficacy of the chatbots in providing accurate and reliable information. Looking at the impact of AI chatbots on patient care outcomes and the satisfaction with the care. Also looking at the effectiveness of chatbots in improving adherence to treatment plans. And things like the feasibility and acceptability of tele audiology services that are provided by chatbots. This brings us to the target group of clinicians in hearing health care and how AI chatbots could provide a role in how clinicians engage in their tasks. So clinicians can benefit from AI chatbots in various ways. Chatbots can for example assist with data collection and analysis by collecting data on patients hearing health. Like self reported symptoms or hearing aid usage. Chatbots can also provide a summary report or visualizations that can help clinicians make treatment decisions like providing a summary report of a patient's hearing test results or highlighting areas of concern and providing recommendations for future evaluation and treatment. Another potential application is assisting with decision making and treatment planning. So for medical applications Google and Deep Mind for example develop MedPalm which is a large language model that incorporates clinical knowledge that has been evaluated using newly developed benchmarks. So chatbots can unlock clinical knowledge or that can unlock clinical knowledge and they could suggest treatment options based on a patient's hearing health history and their symptoms and they could provide information on the benefits and the risks of each option. For example chatbots could suggest a specific type of treatment based on a patient's hearing test results and their preferences. And then finally chatbots can also support clinicians in their communication of information so that they can provide this in more accessible and even person centered ways. So just some recent examples of how these AI chatbots can actually influence the work of clinicians and improve it. Here's an example that was released in March this year of an open AI powered app from Microsoft that will instantly transcribe the patient notes during your doctor's visit or during the audiology visit. So this initiative will allow clinical notes to be made for healthcare workers powered by artificial intelligence. And the whole goal here is to reduce the clinicians administrative burdens by automatically generating a draft of the clinical notes after a patient visit. So you can imagine how that can streamline the engagement and future applications could be that it extracts pertinent information that can help the clinician to double check some of their own clinical judgments after the engagement. And here's another interesting recent paper that came out that looked at how AI chatbots chat GPT in particular can translate radiology reports in this instance into plain language. And they provide some of the results and the potential they discovered in this work. So I think it could just as well be relevant for audiology and audiological reports. So according to the evaluation by radiologists chat GPT could successfully translate radiology reports into plain language with an average score of 4.27 in the 5 point rating system. So very high rating levels and in terms of the suggestions provided by chat GPT they were generally relevant such that they were keeping follow up appointments with doctors and closely monitoring any symptoms. And for 37% of the 138 cases in total chat GPT offered specific suggestions that were based on the findings in the report. So this report makes the conclusion that there is great potential and high effectiveness for chat GPT to translate and simplify advanced radiology reports. So back to our New England Journal of Medicine catalyst article the interview with chat GPT. So if we look at the clinicians they also asked an interesting question in this paper. They asked chat GPT about the crisis in clinician burnout in the US for doctors nurses and other clinicians. I imagine audiologists as well. So how can chat GPT help them? Automation of administrative tasks. So chat GPT can be used to automate routine administrative tasks like scheduling, charting en data entry freeing up clinicians to focus on the more important tasks. Clinical decision supports chat GPT can be used to provide clinicians with real time clinical decision support helping them to make more informed decisions and reducing the risk of medical errors. Potentially in terms of patient education chat GPT could be used to provide patients with information about their condition and treatment options. Helping to reduce the number of questions clinicians need to answer and reduce the burden on clinician time. Is also potential for remote consultations chat GPT could assist with remote consultations allowing clinicians to provide medical advice and treatment to patients regardless of their location. This could help to reduce the burden of travel on clinicians and increase access to care for patients in remote areas. And then finally even emotional support. So chat GPT could potentially be used to provide emotional support to clinicians to help them cope with the stress of their job and providing them with a safe and perhaps non judgmental space to talk about their feelings. So that of course is out of the mouth of chat GPT itself. Most of these things need to be tested, but certainly the data is starting to come in to show some of these applications truly are possible. There's also of course the potential risks. There's a long list of these. I'm just providing some of these risks for us. So there's a risk of a loss of empathy and understanding in patient care that can result in when we're over relying on these AI chat bots. Miss diagnosis is a very real risk or delay diagnosis due to chat bot errors. There's also potentially that incomplete patient information can lead to improper triaging and referral pathways. So significant research priorities are necessary to pursue to validate some of these applications. So integration of AI chat bots with existing health care systems and evaluating those integrations. Also looking at the accuracy and reliability of chat bots for specific functions in health care and in hearing health care. And then also looking at feasibility and effectiveness of these chat bots in really providing an improving patient triage and referral pathways. So I do want to spend a little bit more time looking at the priorities because yes the potential of these AI chat bots are enormous. But there is a massive need for research and validation. So to specify current priorities in terms of that for clinical applications in particular is very important. So here is just some of the thoughts around current priorities. So a first priority would be clinical applications to investigate the potential of these AI chat bots in hearing health care in different applications. But also guideline development for the use of AI chat bots in hearing health research and clinical practice. We do need guidance and oversight and perhaps audiological societies and patient organizations are good resources to support that. There is also the validity and reliability to evaluate how these tools can collect and analyze hearing health data. Also looking at the accuracy and the usability so evaluating the accuracy of hearing health applications. But also the usability and accessibility of these tools. There is also the question around cultural sensitivity and appropriateness for all populations. And to make sure that we are actually getting a good distribution of representation and person centredness in these applications. Ethical considerations obviously abound. There is major issues around informed consent data privacy and security that need to be addressed. And certainly this is an overarching priority across all health care fields. But also in general there is lots of concerns being raised and also efforts to look into how to regulate some of that. There is also long term outcomes to assess the impact on outcomes on quality of life satisfaction and adherence. And then guidelines for usage as we've mentioned earlier are very important. And then finally avoiding misuse and bias. So we know there is a risk of these models being biased because of the information they are trained on. We don't have full transparencies on what data was used to train these models. So we certainly do need work around ensuring that we are having representative well curated responses that are actually built upon reliable data. So maybe in conclusion a last question from this catalyst paper. What will it take before chat GPT can be applied in everyday health care? So data privacy and security as we mentioned. The sensitive nature of health care information requires strong data privacy and security measures. Regulations so health care organizations will need to comply with a range of regulations. Things like HIPAA to ensure safe and appropriate use of chat GPT and other AI chat bots in health care. Data quality and quantity to ensure that these AI chat bots can provide accurate and relevant responses. It will require access to high quality data sets that are diverse and large especially in health care. And then integration with existing systems so to be effective these AI chat bots will need to be integrated with existing health care systems such as electronic health records etc. And then of course we also need human oversight while chat GPT can automate certain tasks. It is important to have human oversight to ensure that the output of the model is correct and to correct any errors that occur. And then finally we mentioned this already ethical considerations and potential for bias and accountability. All those things need to be addressed. So in conclusion AI chat bots represent a major technological advancement for many different fields for health care in particular but also for hearing health care. And it does have the potential to revolutionize hearing health care in many different aspects. But it's important to remember also that it is early days. This technology is still in its infancy in a very real sense it's still being trialed. It's showing the potential but a lot of additional work needs to be done. But at the same time that also demonstrates the future potential that will still be and could still be unlocked by this technology. So we do need rigorous and responsive research in our fields to see how these tools can be utilized for the benefit of patients and clinicians and also research and teaching and education purposes. It's also important that we start to look at how professional and patient organizations can provide guidance for the use of these AI chat bot technologies. In a way that is accountable and that ensures appropriate care and usage across these different user groups. And then finally we do see the major advantages of these technologies to really play into the real need in hearing health care to decentralize hearing health care. To make access more affordable but also to open it up to all different types of populations. And this is not just a regional priority this is a global priority that's also included in the call by the world hearing forum and the world report on hearing that came out recently. So lots of exciting things to look forward to and I hope to continue this discussion and conversation with many of you into the future. Thank you.