DING, Dr John, Medical Director, NexusMD.ai [by video link] KOS, Dr Simon, Global Chief Medical Officer, Heidi Health [by video link] WILSON, Mr Andrew, Chief Customer Officer, NexusMD.ai [by video link] [11:30] CHAIR: Welcome. I understand that information on privilege has been provided to you. Do you have a brief opening statement? Dr Kos : This is my first time turning up to a Senate estimates committee. I'm here at your disposal. I do not have an agenda or a case to prosecute. I'm here to answer your questions. CHAIR: We can go straight to questions, then. I think Senator Ananda-Rajah has some questions for you. Senator ANANDA-RAJAH: Yes, I do. I'm interested to know about your thinking around the incorporation of AI and digital tools in health care. Health care has been, I would argue, at the vanguard of this shift. It's already happening. It's been happening for years. What do you see as the next frontier, in terms of development, and what are some of the barriers to adoption and implementation? Dr Kos : Can I provide some context to where I'm coming from and a bit of a timeline for how that's unfolded, to project out where we're going in the future? Just by background, I practised emergency critical care medicine in New South Wales back in 1999 through 2002. Even then, health care was using artificial intelligence. We had ECG machines producing automated diagnostic interpretations. So health care has been using artificial intelligence. There has been limited adoption of it, and it's mostly been in the diagnostic imaging interpretation space, because that machine learning technology is last generation. With large amounts of training data and the interpreted reports, you can train a model to do a very specific thing. We call that narrow AI. Everything changed, in 2022, when we had democratised large language models and the advent of generative AI. Globally, there have been a number of solutions but, by far and away, the leading use case has been what we call artificial intelligence scribes, AI scribes. This use case, essentially, means that when you go to see your GP they might ask you next, 'Hey, I'm using an artificial intelligence note keeper to record our conversation and it will help me do the documentation. Do you mind if we record?' For busy GPs and doctors and clinicians of all stripes, the administrative overhead has ballooned, in recent years, and the workforce constraints mean the time pressures are acute. They love this technology because, all of a sudden, they can focus their time and attention on the patient, and that low-value cognitive work of doing admin after the fact melts away. The market dynamic means that this AI technology has been rapidly adopted because of the functional benefits, without the due diligence that a technologist or a governance organisation might ask—hard questions like, 'Where's the data stored? How is it handled? Is it used for training purposes?' and the rest of it. So we see this market adoption of shadow AI, we call it, when it's in an organisational sense—we see it walking in the front door, and it's quite well penetrated within the health sector, at the moment. In the general practice space, about 40 per cent of GPs are using AI scribes. In the UK, it's 60 per cent. In Canada, it's 30 per cent. So it's broadly adopted. Now larger organisations are asking some of the questions as they're challenged with this shadow AI within their organisation: How do we catch up? How do we put the right sorts of guardrails in place? I will say that the AI scribe use case is validating the technology and creating trust, and now the clinicians are asking, 'Well, what next? This is great.' So there's an appetite out there that many organisations are struggling to keep up with. Senator ANANDA-RAJAH: Let's take the AI scribe software, because that's what your company develops. Are you saying that you've developed this, you've deployed it and there has been a catch up from the regulators? Dr Kos : Absolutely. Senator ANANDA-RAJAH: Oh really? Do you want to talk us through some of that? Dr Kos : Heidi has been in operation since 2019, but it was only in 2023 that we pivoted to the AI scribe use case, along with many other companies at the time. The adoption of that just virally occurred throughout healthcare organisations globally. To date, Heidi is one of the largest AI scribes by volume. We're doing about 2.7 million sessions per week in 190 countries and 110 languages, so this is a bit of a runaway use case. In terms of regulators, because we operate in all of those global jurisdictions, we see the most stringent and the least stringent. The most stringent would probably be the UK's MHRA, where medical scribes are a class 1 software as a medical device regulation, versus places in the US, where they have autonomous prescribing with no human in the loop of medications for certain conditions. We see the whole continuum. Here in Australia the TGA has adopted a bit of a middle ground. That means that organisations like Heidi are able to deploy and provide value, but there is a conversation going around safety and quality and how we protect consumers. Senator ANANDA-RAJAH: Where should those regulations—guardrails, I guess—sit? The TGA is one entity. What other entities should be engaged in this? Dr Kos : Yes. I think there's a privacy and security conversation. We've got the Australian Privacy Principles, and they govern where data sits and so forth. I think, from a consumer perspective, there's a consent conversation. Doctors are used to consenting patients for the capture of their medical information, but, when they're actually recording, there are different consent requirements state by state, and that's important. It's not happening consistently, and it does need to happen. When we move beyond medical scribes into regulated territory—the TGA looks after diagnosis, treatment, screening, monitoring—organisations that produce AI scribe technology are going there. That's the next frontier. Heidi started as a scribe, but we do have multi products these days that do evidence, agentic communications and even hardware. Regardless of your roots or your origin, you'll move into this regulated space, so that's an ongoing conversation with the TGA. There is Ahpra, the Australian Health Practitioner Regulation Agency not the financial one, that regulates the professional discharge of clinicians as they deliver healthcare services—and that's changing when they're using artificial intelligence technologies—and the various colleges that provide the standards and the education that goes forward. Universities have a role to play as well. I think this is a pretty broad conversation. Senator ANANDA-RAJAH: Yes. Is your product mainly in general practices or have you ventured into hospitals now? Dr Kos : We're implementing or deployed in at least 30 hospitals in Australia, including large deployments of well over a thousand clinicians. Senator ANANDA-RAJAH: We'll get to that in a minute. How do you measure success? What kind of metrics do you use? How do you capture, for example, productivity uplift? Dr Kos : For an organisation like ours that is commercial, makes a product, revenue is one measure of success. And so, as our business grows, that's a primary metric. We do benchmark our market impact and that success more holistically across the sector. We've got impact reports for Australia and New Zealand, for the UK and for Canada. They're now published and they aggregate what we're seeing across those geographies. What we find is that there are productivity savings. When you deploy Heidi AI Scribe, generally for an eight-hour shift, it gives clinicians 59 minutes back. New Zealand, for example, has deployed nationally to every hospital, and they find that in the emergency department every clinician, regardless of seniority, will see an extra patient per shift, so there are real productivity savings. We also measure satisfaction. Clinicians love it because they're doing what they were trained to do rather than the admin, which they hate. There's patient acceptance, and that's surprisingly high. There's over 99 per cent acceptance of the technology, and they like the fact that it gives them more attention rather than a doctor with their back to them, typing on the computer. We also measure things like overtime. Overtime use is rife in healthcare. They call it 'pyjama time' because they finish their shift and then they need to do their notes and often do them at home. Whether they claim that or not, it has implications for the organisation. Unclaimed overtime has resulted in class actions in several states, and claimed is a cost burden to the healthcare sector as well. Heidi impacts all of those. The one that's harder to measure is quality of notes. What we have historically found is that notes are quite cursory in summary and don't have the whole colour of the conversation. When you record a transcript, you get a more complete note as well. But it's harder to measure quality. Senator ANANDA-RAJAH: You're absolutely right. To backtrack a little bit: burnout and fatigue are obviously major problems in health care. These can lead to injuries, can lead to errors and so on. Have you captured the impact of this on those outcomes? Dr Kos : It's subjective. Clinicians report that they feel decreased cognitive load and more enjoyment at work. This is part of the debate that goes on, because burnout is quite rife across the health industry and we know the error rates are high as well. These technologies are not infallible; they do introduce some of their own risks. But, in terms of what the baseline is, we're dealing with a sector that's under strain. Senator ANANDA-RAJAH: For sure. Now, Heidi is obviously a great Australian success story, and we want to foster more companies like Heidi that come through. What does the committee need to consider to ensure we foster this pipeline of talent, ideas and entrepreneurship? Dr Kos : Thank you for the acknowledgement. It's really appreciated. We are very proud that, as an Australian company, we can deliver at global scale. That's incredibly exciting. There's a technological foundation that allows us to do that. Before the advent of cloud computing and distributed web technologies, we wouldn't have been able to do that, so that's very exciting. The regulatory posture being a comfortable middle ground between being overregulated and underregulated means it's viable to test in this Australian marketplace. Australia is a great testbed. Think about the health system we have, which is called the Westminster model and comes out of the UK. Those Westminster models of care are prominent in socialised models of health care around the world, from Singapore to Canada to India to the Nordics in Europe. So you can test in Australia, and those results are relevant to elsewhere in the world, which is fantastic. Some of the things that are a bit challenging about the Australian context include that, while having some of the smartest people in the world, we're expensive in terms of cost of labour, and access to capital for entrepreneurial startups is tight. Often we look overseas for capital-raising and funding, and that naturally pulls technology organisations especially to the US. Senator ANANDA-RAJAH: That's fascinating. We are one of the wealthiest countries in the world, and yet it's difficult to find capital for startups. Dr Kos : And our research chops are fantastic. The quality of the engineers that we that we create is brilliant, but we're not really a centre of gravity. If you're looking at volume and scale, you can quickly run through some of that talent. That becomes a constraint too. Senator ANANDA-RAJAH: Thank you. Chair, I can see NexusMD.ai online. CHAIR: Welcome. Do you have any opening comments, or do you want us to go to questions? Mr Wilson : We've got a short a short speech to read through. CHAIR: Go ahead. Mr Wilson : Thank you very much for the opportunity to appear today. Our submission was not principally about artificial intelligence; it was about productivity, workforce capacity and patient access. Health care is one of Australia's largest industries and one of the fastest-growing areas of public expenditure. The demand continues to rise while the workforce is already under sustained pressure. The practical question is not how to do the same work with fewer people; it's how to enable the people we already have to spend more of their time delivering care. We believe that Australia's healthcare productivity challenge is fundamentally an information problem. Across Australian health care, clinicians and support staff spend countless hours documenting, searching for information, re-entering data, reconciling records and navigating systems that often cannot communicate. In many hospitals, particularly private ones and those in regional settings, much of the bedside record remains paper based, so, even when information is digital, it often cannot move efficiently across the entire patient journey. Artificial intelligence matters because it can solve both problems. It can transform conversations, handwriting, scanned documents and PDFs into structured clinical information that helps information move through interoperability and workflows. Importantly, this is not about replacing clinicians; it's about returning time to clinicians, nurses, allied health professionals, coders, health information managers and administrative teams. Every hour that is recovered from avoidable administrative tasks is capacity that's returned to patient care without the years required to train another health professional. Dr Ding : As an ED physician, I'm still working on the ED floor. I would like to reinforce Andrew's points. His points are not theoretical. For every shift, I spend over time documenting, trying to get data out and trying to reconcile information from other sources. This work has been done repeatedly during the journey the patient experiences in the hospital. In my view, the AI is not going to replace me but will put me in the centre of the decision-making process, liberating me from my computer and supporting me back to the bedside, to my job to look after my patient. Mr Wilson : Our submission deliberately avoids relying on any specific vendor claims. We're using Australia's own national hospital-cost data collection. Even a one per cent productivity improvement across Australia's approximately $70 billion public hospital cost base represents about $700 million annually. This is not a forecast; it just demonstrates why there is a need for independent, real-world evaluation and scale deployment. Our recommendations are vendor neutral. They focus on interoperability, consistent risk based regulation, sovereign capability, procurement reform and moving proven tools beyond repeated pilots into safe, wide-system adoption. Australia has never lacked innovation; what we're lacking is adoption, and closing that gap is one of the most practical opportunities available to improve healthcare access, workforce sustainability and national productivity. Thank you. CHAIR: Thank you. I think we have some questions for you from Senator Ananda-Rajah. Senator ANANDA-RAJAH: I want to unpack a few things in your submission. Could you talk us through the issue of interoperability standards and how this is a barrier. It speaks to one of your recommendations, recommendation 5, where you also cite an anti-inflammation blocking rule. I have had numerous colleagues from public hospitals who are heavy users of the EMR, for example, who have complained to me about the problems they're encountering with getting data out of their EMRs, the back end, for audit, for quality improvement, for research, for compliance and so on and so forth. I'm interested in this issue of the incumbents versus new entrants like yourselves, and what the committee needs to consider in terms of what these barriers are and how we go about breaking them down. Mr Wilson : Understood. There are a couple of aspects to that question, and I'll try to address them concisely. Firstly, we do appreciate and understand that the Australian technical components have been put into an activity to improve interoperability throughout public health. There are two aspects of interoperability that are really impacted. The first one is that there has to be a data model that has to be able to be shared. Considering that the majority of private hospitals and regional hospitals are still working in paper, the interoperability capability doesn't exist for them. The only way that can be done is through fax or through handwriting. That's why one aspect of interoperability is converting from the existing method into a digital method, and that's not being addressed by the current government requirements. The second aspect around interoperability is the standards for where the interoperability exchange exists. We call it out as something that it's very important for productivity because that information is not being shared consistently in Australia. Even if there is a definitive vision for it to happen, it doesn't exist right now. You need to be fully validated with any of the public hospital systems using their US based standards for interoperability in order to be even considered by any of the public health sectors for pilots, for trials, for tenders, or for procurement. That's a real challenge because the majority of Australian public hospitals use an EMR that has originated from the US. Those standards for interoperability have not been switched on clearly for the Australian market, which means that every new vendor, start-up vendor or company in Australia that needs to have a product interoperate with any of these solutions must be validated by that solution according to its standards, not according to what the interoperability standards requirements need to happen. That's (a) cost, (b) time and (c) a moving target that can be very difficult for any new vendor to be able to work to. Senator ANANDA-RAJAH: Who has carriage of that aspect? Which agency? Mr Wilson : It's part of the Australian Digital Health Agency. It's theirs to work on. Senator ANANDA-RAJAH: We should probably call them and question them about this. Feel free to also describe to us the kind of software you've developed and give us case examples. It just makes it a bit more concrete as to where interoperability has been an issue, for example. Mr Wilson : As an example, when you are a patient in an ambulance and you arrive at an emergency department, there is no interoperability that exists between the ambulance documentation system and the clinical staff who are working on the emergency floor. There is a visual to explain where the ambulance is in terms of time for delivery, but for the staff to understand what the health issue is for that particular patient, that is done by a phone call, and that's as much as they can receive. One of the tools that we have built is the capability for an ambulance solution to be digital, with the capability for the ambulance solution to then be integrated into an onsite device at the emergency department that the emergency department can receive and read in real time. The challenge for us is that, for the emergency departments who are using EMR, unless our solution fully integrates with the EMR they do not want to consider this application for this part of emergency care. We have many fellows of the Australasian Emergency Care society talk to us—the FACEMs, which Dr Ding here can talk to in detail—that explain that knowledge of critical patients arriving gives them an opportunity to prepare and to understand resources. They need to get access to CT or surgery much quicker. Waiting until the patient arrives— Senator ANANDA-RAJAH: Can I just stop you there. The hospital has said you've got to be compatible with its EMR, noting that these EMRs are different across these hospitals—so it's Cerner or Epic or whatever. How does it work in other countries? You mentioned the US has a different approach; it's got the cures act. What can we learn from that approach? Mr Wilson : With the cures act in the US, the reason the amendment was added to that bill was the preceding act, which was the HITECH Act—which was introduced both as a productivity measure and an inducement to get a budget stimulus happening in North America—pushed the idea that we need to move to digital records. Thus, there was an extremely high uptake of electronic medical records across North America. Caveats were put in for that, but the funding was— Senator ANANDA-RAJAH: This was Obama, right? Mr Wilson : This was Obama, correct. Very good vision—a very good idea, in my opinion. The challenge was the speed to delivery and some of the requirements that were needed for the solution. It meant that hospitals needed to attest. The only way they could attest was through an entire EMR. There was a limited number of entire EMRs. Therefore, the majority of the business went to two primary companies for the very large networks. Senator ANANDA-RAJAH: Those companies are? Mr Wilson : Those two companies are Epic and Cerner. In North America, the challenge therefore became that no interoperability mandate had been required in this. If you were an existing vendor that was in these hospitals but you did not have interoperability with those solutions, you lost your business immediately—which had an impact on exactly how that department ran, because neither of these two systems, or others, can completely cover the entire digital requirement for health. Over that period of time, there were examples—I'll use an example. A patient may decide to move from one state to another state for work. When they're in the other state, they then need to go into a hospital and continue with their care. That hospital may have Epic as its solution and needs to get the data from the other hospital, which may have had Cerner as the solution. Both solutions would not interface and they would not be able to share data. The second point that the government needed to address was some patients wanting access to their own patient data, whether for medicolegal requirements or just to have their own medical record. In Australia that is being addressed through My Health Record, but with this 21st Century Cures Act amendment it was decided to allow patients a format to get access to their data and, secondly, to define an interoperability rules exchange system that had to be mandated. There were fines and requirements put into this to prevent information blocking. Senator ANANDA-RAJAH: So we could conceivably do something like this. There's nothing stopping the Commonwealth—maybe working with the states, because we don't run hospitals—legislating something like this and saying, 'You must open up your systems to allow vendors to plug in.' Mr Wilson : I would support this. All three large US vendors that work in Australia across the public health sector already have these interoperability standards built into their solution; they're just not activated here. Senator ANANDA-RAJAH: Who's the third vendor that you mentioned? Mr Wilson : The third vendor is Allscripts. It's based in South Australia. Senator ANANDA-RAJAH: That's helpful. I'm keen to understand a few things. Can I have a brief response on cost recovery models. For example, if hospitals or clinics want to adopt digital tools, ideally from Australian companies such as yourselves—start-ups and so on—what kind of cost recovery model should we be looking at, whether that's Medicare or MBS item numbers and that kind of thing? I'm interested in credentialing—who credentials all these tools, or are you having to go hospital to hospital to sell your wares? I'm also interested in the whole sandbox concept. There are three things there I'd like some commentary on. Let's go to cost recovery first—and feel free to chime in, Heidi Health. Mr Wilson : We talk a bit about a requirement for any funding to be saved to be reinvested back into the community health. If there are productivity improvements, we don't necessarily believe that they should be just cost savings because everyone measures a definition of success differently. For some markets, it's not about saving money; it's definitely about improving health care—and that should be the first measurement for everyone. We think the productivity dividend, as a definition, should be that, if we can secure productivity improvements in large networks, we should be able to realign that funding into remote and rural markets or to preventative health care—and also to Indigenous and CALD requirements, for those markets. Dr Ding : Our emphasis on that is that we don't have an abundant workforce in Australia's hospital system; we're actually short of staff. Asking to do more for less—that's not the case here. We want to do more with the same people. We're not going to use technology to replace humans and save money. The money should go back to the system to invest in the future there. That's what we're asking for. Senator ANANDA-RAJAH: My understanding is there are no MBS item numbers. If you wanted to adopt a telehealth tool, how are these things funded? Are we asking GP clinics to pay? There are capital costs that come with this. How do we want to think about funding models? Dr Ding : You can pay more through Medicare for item numbers or whatever item number you choose. I don't think it will be a particular item number you can use just for AI technology itself. If I use an ultrasound at a bedside in the ED to treat a patient, I cannot bill the item that uses ultrasound to put the needles in for synchronising—our computer is synchronising, but I can't bill it by using technology. In Medicare as a whole, if you use a robot to do the operation you can't bill because the operation is done by a robot; you still bill the same amount for that robot. That's the problem with Medicare itself. I don't think you can introduce a payment system into Medicare itself. There will be separate funding. Dr Kos : From a Heidi perspective, we've got two commercial segments in the community specialist and GP space. They can bill item numbers—so their business case needs to be made on the basis of productivity. You put in scribes, and you can either see more patients—or some of them do it as a quality of life; I don't have to do overtime afterwards, so I've got more hours back in the day. That's a productivity based business case. When it comes to the enterprise, which is hospitals, public and private can be a little different. We have enterprise cap models or usage based models—even room based models—as well as seat and enrolment based models. For them, the publics are thinking about how they de-risk their overtime; they're thinking about productivity: How do we get through more of our outpatient clinics? How do we decrease our outstanding letters? How do we substitute across our transcription costs? Digital dictation is what AI scribes replace, and sometimes that goes out to transcription services offshore, and there's a long turnaround time. Private hospitals find it a little more difficult to make the business case. They typically want to see AI scribing and the quality of note-keeping flow through to clinical coding so they can claim more per case that comes through, because commonly there's money left on the table there. Senator ANANDA-RAJAH: And could we just briefly get a comment on credentialing and sandboxing—the concept? Could you just explain to the committee what that means? Mr Wilson : I think this was mentioned when we started—about multiple pilots being undertaken. This is actually a real burden and cost for all companies in the sense that everywhere you go to talk to a hospital they don't have a tender process for this technology at this stage. They want to run a pilot, which is a large cost in time and material. But typically the only thing there to be addressed is the low-risk items. When it comes to AI, there are different gradings of AI and how it impacts with different levels of technology, which is currently regulated through the TGA, so there are no concerns about the way the TGA approaches the regulation space. But in order for us to know the effectiveness and to collect the clinical data from an Australian perspective of higher-risk AI or broader cases of AI, we want to have these regulated sandboxes. This is one of the challenges. You're probably finding today that there are multiple pilots in multiple departments collecting the same information across every different hospital in every state, when this could have been set up and defined at a national level with very clear definitions of what needs to be done by getting expert involvement, from an AI perspective and also from a clinical and an administrative perspective, and put it under the control of these defined sandboxes so that these high-risk models can be assessed now, because the only way that can be assessed is under a controlled environment. Dr Ding : If the software has been developed, and you go to an agency that is certified then you don't have to go to every hospital and assess them by themselves. Hospitals don't have expertise in AI technology. So there is constant delay, because they can't get expert opinion from outside, or take the time and effort to do it. Senator ANANDA-RAJAH: Should this be something we put to the digital health agency—this concept of credentialing all certification as well as the sandbox? Dr Ding : Yes. Senator ANANDA-RAJAH: Okay. I might stop there, because Senator Darmanin has a few questions. CHAIR: Senator Darmanin. Dr Kos : I was just going to say that it is probably a little more than that. There's a lack of credentialing. If your outside software is a medical device, the TGA doesn't have purview, and you fall back on international standards like ISO. There is an AI ISO standard, 42000, and there are some international standards that are recognised that we called out—HITECH and HIPAA and all the rest of it. The real lack is governance in organisations. So, beyond credentialing, most organisations don't have an AI policy, don't have an AI governance structure, don't have a way to manage their risk, so they recognise the upside benefit, but with the lack of being able to quantify their downside risk are conservative in the health industries. Senator DARMANIN: Thank you for that. I was actually going to ask about exactly this issue you've just raise I was not a healthcare professional in my previous life, so some of these details I'm not as familiar with. Just broadly, what are the risks if there aren't those sorts of frameworks in place? What do we need to think about from a governance perspective, but also as to the skills and adaptation of the workforce at the different levels? There's obviously that sort of administrative arm, but there are also the practitioners using the technology. How are we supporting the workforce to adapt? Dr Kos : I'd call the healthcare sector a lagging adopter of information technology—I'd put it there with the legal and public sectors—compared it to some of the more progressive adopters like banking, manufacturing and retail. And there are good reasons for that. The risks of these models are inherent in the technology. It's a new technology. We hadn't had generative-AI based AI systems in place until, really, 2023, and now we're moving rapidly into agentic AI. So there's evolution in the technology. There's a departure from the traditional confidence that we had in AI systems that were deterministic—meaning that, given the same input, you'd get the same output. Generative models are stochastic, and so there is a risk of hallucination. The models aren't always perfect. There are ways of putting guardrails in place. Most ways they mitigate it are putting humans in the loop. So AI produces recommendations; humans validate it. And that's how we try and trap errors. You called out the workforce skilling, readiness and functions. That exists at every level. At the board level, we're finding that, because these models are not static, they need adaptive governance techniques that aren't in common usage at board levels, and we're finding that our board directors don't actually have the baseline knowledge of skills or understanding of risk profiles to be able to govern effectively. At the executive level, there's a function that doesn't yet exist, the chief AI officer, which I expect to see in coming years. It's not a technology function; I actually think it's a business transformation function. At a workforce layer, we're actually seeing a whole diffusion curve of technology. There are a lot of early adopters, fast followers, and even the early majority, but that's about 40 per cent of the workforce. We're leaving behind the late majority and the laggards. I think that requires some skilling, training, and a corporate mandate. It requires policy. It requires in-situ post-market surveillance and monitoring. And these functions that we'd characterise of a mature kind of sector don't yet exist. We're trying our hardest to catch up as fast as we can. Senator DARMANIN: That is very interesting. One might argue that being a laggard in this system is not a bad thing, given the kinds of risks that are associated with things getting away without the proper structures in place. But do you think that, at the policy level—taking it outside of the hospital or the healthcare environment and the responsibility that lies there—we have the right frameworks in place, at state and federal level, to guard against things going horribly wrong? As to the question about humans in the loop, is this mandated? We've had this conversation in other inquiries around AI. What is the situation, from your perspective? And what needs to change? Dr Kos : Even though health care is a conservative sector, the genie's out of the bottle. These technologies are democratised, which means that they don't require an organisational budget. They don't require an organisational mandate. They don't require ops-down strategy. They're actually walking their way into the organisation, piggybacking on technologies that are already in there. We talked about electronic medical record vendors—all of them will be adding on their own generative capabilities. Horizontal technologies like Microsoft Copilot, Claude and OpenAI are in organisations. Then there are the individual specific functions, and we talked a little bit about AI scribes. These can be adopted and walked straight through. They bring their own devices into the organisation; they've got their own network, and they don't need people's permission. That's why we urgently need to catch up with governance. Just the fact that it's a conservative industry doesn't eliminate the risk. Senator DARMANIN: What's the urgent catch up that's required? Dr Kos : You asked about policy infrastructure and readiness in our states and territories. They don't exist. I'll call out very specifically from a scribe perspective how the different states are dealing with it. The state of Queensland is the most advanced. In the lack of a policy directive, the individual hospital and health services adopted these technologies readily with autonomy, and now they're the most advanced state. South Australia said: 'No, we can't possibly have this technology in our state without the policy infrastructure. Stop.' It hasn't stopped anything; it's just driven it underground like during the Prohibition. New South Wales, at the state level, has said, 'Stop, and we'll do a state wide approach to how we do this.' They've now got the policy infrastructure, the enterprise level guardrails and a 'go forward' plan about how to fund it. There's no common way that the different states and territories are thinking about it. Dr Ding : I would just add one more point. Permission is a concern. Medical documents are legal documents. So who puts their name on the paper and takes full responsibility for the document produced by AI? With all my documents, I've asked clearly that they need to review, edit and sign off. Otherwise, it won't be a legal document. Senator DARMANIN: So the buck stops with the individual. Dr Ding : Yes, with the individual clinician so that humans are always in the loop. In health care, the machine can never replace me on the floor because I'm the one making the final decision. AI may produce documents for me and make some suggestions for a diagnosis, but I'm the one to decide which option I'm taking. Senator DARMANIN: Given you are from specific parts of the industry, you may not have a full answer to my next question, but I'll give it a crack anyway. In terms of measuring productivity and health overall, there's been an ongoing discussion about how you measure productivity in the care sector more broadly. What are your reflections—what have you seen that works well, what are the gaps and what should we be thinking about when it comes to your particular areas of expertise? We've already talked about a productivity dividend with preventative health care. I think that generally applies in different parts of the provision of social service—this idea of prevention—but how do you count it in health care, and is there anything we can learn for other parts of the care sector? Dr Kos : I will caveat my response by saying that I am not a health economist, but I do recognise the roots of our health system. We created a hospital system to deal with battlefield trauma and infectious disease, and the burden of disease has moved to lifestyle induced chronic illness. We haven't really adapted our model. We fund these acute hospitals, and we deal with the consequences of things that are probably better managed in the community preventatively, proactively and so forth. We have a squarely fee-for-service based health system where we reimburse activity rather than the more progressive health financial systems, which is value based care, where you decide outcomes and you reward for that. As far as addressing the demand for healthcare services, it grows year on year, and it outstrips the facilities we've got. I think we would need to build a 350-bed hospital every month for the next 15 years to meet the rising costs of demand, and that's before you even talk about the workforce needed to deliver against that. Where you see it pile up is elective surgical waiting lists, emergency department wait times and demand for hospital outpatient clinics. Those are probably some of the metrics that we can talk about from a throughput perspective. That probably neglects the quality dimension as well. I think any time you talk about productivity, you need to talk about quality, and that's error rates, complication rates and the subjective patient experience. Mr Wilson : I would add that, in our submission, in section 3, we talk about measuring productivity. We have highlighted the fact that there are three sets of parameters. One is the way that the public consider productivity in health care, and that is how long they have to wait. The government determines it by how much it costs, and the hospitals determine by the speed of patient throughput. Simon mentioned value based health care and having outcomes based models, but none of this addresses what happens in the middle between the start and end. We have suggested four additional parameters around measuring for productivity improvement, which are based around clinical outcomes relative to the resources, a measurement of the administrative burden for that particular set of episodes, complexity for the needed workload for the clinicians and the medical team, and then the impact on patient access to services. If we do it just cost per episode, we're not measuring the impact of the time that's lost in the system to interoperability issues or how long it takes to build notes, the accuracy of notes and the correction of notes et cetera. We want to understand where these delays impact the way the systems are determined, because Medicare item numbers and MBS numbers don't have that parameter in there. It's a medical point and a technical component to most of those item numbers. There's nothing in the regulation either. There's nothing in the TGA that determines any of those aspects. It's a difficult one. Senator DARMANIN: Thanks, Chair. Dr Kos : Can I build on Andrew's answer just quickly? CHAIR: No worries. Dr Kos : When we talk about productivity, we talk about the law of averages. But when you look at health equity, we've got some really significant under-served populations. It's just worth calling out that, for regional, remote and culturally and linguistically diverse populations, some health conditions like mental illness and those that isolate people with mobility can mean that your access to health services and your outcomes are different—even if you live in metro areas. That probably gets lost if you just look at the aggregate statistics. CHAIR: Thank you very much for being there today and for giving your time. We really appreciate your evidence. Proceedings suspended from 12:22 to 13:15 .