Dr. Jared Powell, PhD:
In 2016, famed computer scientist Jeffrey Hinton, the man now commonly called the godfather of AI, said that we should stop training radiologists. Within five years, he stated confidently the machines would read scans better. But here we are, 10 years on, and we now have more radiologists, not fewer. Demand for imaging has gone up as had demand for human radiologists. So was Hinton wrong or was he just early and wrong about how it would happen?
Dr. Jared Powell, PhD:
To be honest, I hate radiology. So let's get to musculoskeletal rehabilitation quickly. How will our profession look in five years? Have you thought about it? Most of what I hear and read is that musculoskeletal clinicians think they will be unaffected by generative AI. A loud minority reckon our profession is toast. My position is that both of these views are wrong. Outside of a few impassioned voices, I'm most concerned about our general apathy as a profession to the rising threat of AI. So I'm going to do something about it. So this episode marks the launch of my new course, The Irreplaceable Clinician. I'll tell you all about it at the end. But firstly, I want to run you through my thesis and introduce the arguments and facts that we should all be acutely aware of. So let's start with an exploration of what this thing called AI is.
Dr. Jared Powell, PhD:
So every technology in living memory has come for someone's job. The loom came for the weaver. The tractor came for the farm hand. The spreadsheet came for the bookkeeper and the filing clerk. But each wave of technological innovation stops short of a special kind of work, the work or job of knowing stuff, of exercising judgment and advising someone what to do. In other words, professional work, which is our work. It's your work, it's my work. Richard and Daniel Suskind, who are formidable father and son duo, talk and write about AI. They saw this coming a decade ago in their book, The Future of the Professions. Their argument was simple, but no one really listened. The professions exist because expertise is scarce. Professionals gatekeep a body of knowledge that lay people can't easily access or understand. What happens when we make this knowledge cheap and abundant?
Dr. Jared Powell, PhD:
The gate breaks down and knowledge becomes democratized. Richard Suskind's more recent book, which is titled How to Think About AI calls AI the defining challenge of our age. I tend to agree, no disrespect to the plethora of other challenges that we as a species are currently facing. But before we go any further, I want to clarify that AI in its current form that is gradually getting better every single day is the massive concern. We need not wait for the arrival of true superintelligence or AGI or the singularity to become appropriately worried. But right now, with the functionality of today's models, and to be clear, I'm talking about large language models, ChatGPT, Claude, and the like. This is the scenario I care about, and it's the one already impacting our work. I don't believe this is science fiction anymore. It's real life, and we must be prepared.
Dr. Jared Powell, PhD:
To give you a sense at how fast this is all moving, read Sebastian Malaby's book called The Infinity Machine, which is based on Demis Hassabas, the founder of Google DeepMind, which is Google's AI division. Hassabas's lab took protein folding, which was a grand challenge biology had chewed on for about 50 years and largely cracked it in the space of four years. He went on to win the Nobel Prize for his work. This is old news now. It's about four years ago, and we now have AI models that are solving previously intractable mathematical problems seemingly every other week. Terrence Tao, widely regarded as the greatest living mathematician in the world, and also the fellow Aussie, shout out Terrence Tao, is terrified at this prospect and is urging AI labs to slow down. Ethan Mollick, who is a popular writer on AI, puts the practical version best in his book, Co-Intelligence.
Dr. Jared Powell, PhD:
Assume the AI you're using today is the worst you'll ever use. Guys, we must confront this technology and come up with a plan on what to do about it. So how is our profession responding? Mostly with one of two stories. The first story is business as usual. It goes something like this. AI can't touch a shoulder or a knee or a back. It can't build rapport. Patients will always want a human. This is called not us thinking. Every profession agrees AI will transform the other professions. Lawyers think it's coming for accountants. Accountants think it's coming for radiologists, and we think it's coming for everyone except the people with their hands on patients. Daniel Suskin makes a deeper point from another of his books called A World Without Work. And he says that we keep making the same mistake. We assume a machine has to do a task the way we do it in order for that machine to replace us, but it doesn't.
Dr. Jared Powell, PhD:
It only has to get the patient or the customer or the student or whatever it may be to the particular outcome. So this has significant ramifications for manual therapy. Now, manual therapy is a genuine moat or protective layer around physiotherapy, chiropractic, osteopathy, or whatever, and I can see that fully. AI can't do it and won't for a long time until the science of robotics catch up. But all the evidence that we have today reports that manual therapy doesn't reliably beat exercise therapy and often simple advice and education in terms of outcomes. So this means that AI doesn't need to cross the moat or break down the protective layer. It only needs to help the patient get there, get to the outcome via a different route. The second story is doom. Quotes like, "It's over, retrain, retreat to the woods, join a cult. The profession won't exist in 10 years," are also popular in some circles.
Dr. Jared Powell, PhD:
This is the sentiment of Jeffrey Hinton in 2016, and it mistakes a capability for an outcome. A machine that can read a scan didn't remove the radiologist because the job was never just reading scans. The job is a bundle of tasks wrapped in professional accountability, regulation, and trust. Both of these stories assume you're passive and can't do anything about your future. If everything's lost, there's nothing you can do. I really don't agree with this stance. So this is a podcast that has gained popularity through its commitment to evidence-based medicine or evidence-based practice, and that ain't gonna change. So let's take a look at the evidence. There are some findings from current research groups that should make us seriously concerned. There are also some findings that should give you hope. Okay, number one, the machine alone beat the doctors with the machine. In a randomized controlled trial published in JAMA in 2024, GO and colleagues GOH, GO and colleagues gave physicians tricky diagnostic cases.
Dr. Jared Powell, PhD:
So half of these physicians got GPT-4, which is the machine that powers ChatGPT-4 to help. The other half used conventional resources such as books and internet sources. Physicians who had the AI barely improved, so the AI barely helped physicians at all. However, when GPT-4 in this experiment was working alone, it scored well above both groups. So listen to that again. GPT-4 working alone scored well above the physician using AI and the physician using conventional resources such as books and internet sources. So humans using AI were worse than AI alone. We are holding AI back according to this study. Number two, the machine was rated as more empathetic. In 2023, AIRS and colleagues in JAMA again, they took real patient questions from an online forum, answered by doctors and a chatbot, and then gave these answers to blind raters. They found that blinded raters preferred the chatbot's answers in nearly four out of five cases and rated them as more empathetic far more often.
Dr. Jared Powell, PhD:
So these blinded raters were rating AI chatbot answers as more empathetic than human answers. So if your plan is to out war more out empathy the machine, perhaps think again. Three, humans plus AI isn't automatically better. A 2024 meta-analysis published in Nature by Vacaro and colleagues pulled over a hundred different experiments. On average, they found that human plus AI teams did worse than the better of the human or the AI working alone, especially on decision tasks. There is a lovely phrase for this called the jagged frontier. So we know that AI can be superhuman at some tasks and pretty much hopeless or useless at other tasks. I know we've all shared this frustration before, but the problem is that we can't often see the edge of this jagged frontier. We know that AI often gives confident and fluent wrong answers, and these can be hard to catch.
Dr. Jared Powell, PhD:
This is called AI explaining, a paper published just this year in BMJ Open Stress Tested Chatbots on Health Questions. Now, worryingly, half of the answers were problematic and almost none of the answers from these chatbots came with any hesitation or uncertainty. They were confidently wrong about half the time. Okay, so they're the worrying cases. Now let's get to the one hopeful case. Owens and colleagues again in JAMA this year, 2026, asked patients how they felt about AI drafted messages from their care team. Now, patients didn't necessarily reject the intervention or contribution of AI drafted messages, but they did want a clinician, a human clinician who had read the answer, who had verified the answer, and who was accountable for the answer. Now, that is an important point. So if you put these cases together, I think it starts to paint a compelling picture. The AI is often more knowledgeable than us.
Dr. Jared Powell, PhD:
It can, in fact, sound kinder than us in some experiments, and simply bolting it into our practice doesn't seem to help much. And yet, despite all this, patients still want a human being who owns the decision, but this only works for clinicians who are competent enough to police or fact-check the AI, and we'll get to this later. Let's zoom out for a minute and borrow a way of thinking from economics. So let's ask the question the way an economist would. Let's delete the question, will physios still have jobs in five or 10 years time? But rather ask, when something becomes cheap, what becomes scarce? And this is what economists think about all the time. So Alex Emas is a behavioral economist at Chicago Booth University in the US, and he has written extensively on this. His starting point is that economics is the study of scarcity, and scarcity doesn't disappear in the age of AI, but it will relocate.
Dr. Jared Powell, PhD:
So as AI pushes the cost of anything mass producible towards zero, including a lot of thinking, value or scarcity will migrate to what he calls the relational sector. That's work where a human being in the loop is part of what you're paying for. His own experiments show this. When people learn AI was involved in making something, the premium they'll pay for its exclusivity drops sharply. I think the same. When I find out that something is AI produced, , whether it be a film or a book or a song, I think less of that particular product. What do you think? Do you share the same sentiment? Alexi Mass also says something about doctors that I think applies pretty directly to us as musculoskeletal clinicians. Right now, clinical jobs are part relational and part automatable. As the automatable half gets absorbed or perhaps encroached by AI, the job won't necessarily vanish.
Dr. Jared Powell, PhD:
Instead, the job will become more relational. Another one of my favorite books is called Messy Jobs, and it's just been published this year in 2026. Now, the main thesis of the book is that a job is a bundle of tasks rather than one simple task. The clean, single task jobs are the exposed minority of jobs that are exposed to AI, but most jobs are messy. They mix thinking and judgment with presence, trust, and accountability, and someone has to own the outcome of the job. I think a physio consultation is about as messy as a job gets. Does this put physio out of reach of AI? Maybe, maybe not. There's a neuroscience and philosophy angle here too. Michael Blundy said, "We know more than we can tell." And this is about tacit knowledge. Now, much of what makes an experienced clinician good is not exactly written down, so it can't be scraped for AI training data.
Dr. Jared Powell, PhD:
Is this another moat, this concept of tacit knowledge? Again, maybe. So the economics says our sector or our profession survives, but that value inside it will strateg - will stratify, will relocate. The generic middle will get squeezed. This is the clinician whose main offer is, you know, basic information or a standard exercise program or a pleasant manner. This is basically someone doing the bare minimum and they're in trouble. This could lead to a slow drain of value away from clinicians to AI, and I'm not sure if this is exactly bad. We have way too many clinicians doing the bare minimum. Now we've got some competition, and I think the cream will rise to the top. So as I see it, we have three options. Number one, carry on. We just keep doing what we've always done and trust that touch, rapport, and tradition will hold. I think this is wrong.
Dr. Jared Powell, PhD:
Two, get out, retrain, or disengage, or move to the hills and become a farmer. Again, I don't think this is a move, although you can if you want. Farming is a really cool and peaceful job. And three, we evolve. We accept that what patients can get from AI has now permanently changed our job, and we must deliberately change what we offer in response. I choose the third option, and I urge you to do the same. This is, in essence, Daniel Suskin's answer in his new book, What Should My Children Do? He wrote it for parents, but it also reads like a manual for our profession. We don't just ban the AI tools. We don't cling to inherited ways of working. You know I hate dogma, but we learn to master the new technology without letting it erode your skills and competencies underneath. That's the key.
Dr. Jared Powell, PhD:
If you hand your clinical reasoning and critical thinking to the machine, the skill that lets you check the machine will wither away. I see this as being an important part of our job in the age of AI, of checking the machine. So we should use AI, absolutely, if it aligns with your values and your preferences, and if it's within its strengths in the drag - jagged frontier. But please don't use it in a way that hollows out the judgment you'll need to oversee it. Patients will want us to oversee it. You need to be a competent clinician to effectively do so. So a human in the loop only adds value if that human is competent enough to catch what the machine gets wrong. Otherwise, it's just theatrics. So what does evolving look like? So this is the argument at the heart of my new course, the irreplaceable clinician.
Dr. Jared Powell, PhD:
And let me break it down for you. First, you need to understand that clinical practice, your job is knowledge work. So explicit knowledge, the stuff in textbooks and guidelines is exactly what AI is best at. If that's your main offer, you're in direct competition with something that's free, tireless, and available at 2:00 AM. Second, you need to move to where the gray is. Now, musculoskeletal rehab, for the most part, operates in an evidential gray zone. If you ask three different AI models about a patient with six weeks of knee pain or shoulder pain or back pain, you'll get three plausible and probably different answers. However, the reality is, if you ask three physios or three osteopaths the same thing, you'll get the same, three different answers. Consistency is not really the comparative advantage that we have. AI can suggest reasonable diagnoses and treatment pathways, but it's not great at advising what this person, the person who is sitting opposite you, should do next, because often, there is no single right answer.
Dr. Jared Powell, PhD:
So operating in this gray zone is where the good clinicians will thrive and will continue to do so in the future. Third, you need to make sure your human care is worth seeking. Simply being a human is a pretty low bar. Your job is to make sure your human care is worth seeking. What will underpin this is competence and attention, an explanation that the patient can use and checking they understood it, a plan that survives contact with this person's messy life and gets revised efficiently and effectively when it doesn't. The free AI answer is, is strong at the start of care, but it's weakest at follow through. Three weeks later, when things haven't gone to plan, the AI gets less and less effective. This is where continuity of care is important, and maybe this is where you come into your own. Fourth, you need to build a practice around it.
Dr. Jared Powell, PhD:
So use AI tools inside your service where it helps and be clear who's accountable when it's wrong. And to be clear, it's you. You also must be knowable and findable by the right people with a track record they can check. So when content is abundant and even people can be faked, which is a very real and very scary thought, a verifiable record of productive and effective work becomes a scarce and therefore valuable asset. There are sadly no shortcuts to developing deep expertise or trust. This is exactly why these skills will hold their value and will remain irreplaceable in the age of AI. The Irreplaceable Clinician is five modules for musculoskeletal clinicians on expertise and judgment in the age of AI. I created this course after immersing myself in all the books and trials I've talked about in this episode and many, many more, been a wild couple of years.
Dr. Jared Powell, PhD:
There is an enormous volume of commentary and literature out there about AI, with almost none of it penetrating our day-to-day echo chamber of arguing over treatments and making reels about each other on Instagram. This course crucially runs on a case study that ripples throughout each module. You see the answer that this fictitious person got when she asked AI about her pain, and then you'll follow her throughout two consultations, the first appointment and the follow-up review three weeks later. Step by step, you'll see what a clinician adds that the AI can't answer or struggles with. You'll also see where AI does it better because we must consider that too. The course will also change what you do in your practice. It's much more than just a way of thinking. You will need to proactively take steps to make sure that you and your clinical practice will still be here in five or 10 years time, and you must start thinking about it.
Dr. Jared Powell, PhD:
So the course is now live. It only launched yesterday. Again, you will get lifetime access, as with all of my courses, including all upgrades, and you'll get a free month in the complete clinician mentorship upon purchase. The link is in the show notes if you can. Are you ready for physical therapy 2.0? You better be. Okay. Thanks for listening. I'll chat to you soon. The Shoulder Physio Podcast would like to acknowledge that this episode was recorded from the lands of the Terribalang people. I also acknowledge the traditional custodians of the lands on which each of you are living, learning, and working from every day. I pay my respects to elders past, present, and emerging and celebrate the diversity of Aboriginal and Torres Strait Islander peoples and their ongoing cultures and connections to the lands and waters of Australia.
