Click on your preferred podcast player link to listen wherever you enjoy podcasts.
Welcome to the show notes for Episode 147 of the OT Schoolhouse Podcast.
Do you think AI can revolutionize school-based OT?
In this episode, join expert guest Teresa Fair-Field as she explores innovative uses of artificial intelligence that are reshaping the way OTP's support students and clinical decision making.
Discover the collaborative efforts between clinicians and tech experts in developing predictive AI, and learn how these advancements can generate treatment plans and improve communication.
This episode is great for any school-based OT who wants to stay up to date, and utilize technology to enhance evidence-based practice and ultimately enrich student outcomes.
Listen now to learn the following objectives:
Learners will recognize the ethical considerations of AI use.
Learners will understand the role of occupational therapy in supporting neurodivergent students in physical education settings.
Learners will identify strategies for integrating AI into clinical decision-making, IEP development, and treatment planning.
Guest Bio
Teresa Fair-Field, OTD, OTR/L
Clinical Assistant Professor of Occupational Therapy
Baylor University
Teresa Fair-Field, OTD, OTR/L, Clinical Assistant Professor of Occupational Therapy at Baylor University since 2021, affiliated with Early Markers, a Seattle-based startup focusing on machine learning and infant motor development. Their NIH-funded work aims to create an app for AI-supported early screening, parent coaching, and education. Teresa specializes in neurological development across the lifespan, including early intervention, infant motor skills, and neuromotor conditions.
Quotes
"AI is a family of computer systems... aimed at solving problems that could include learning, problem solving, decision making, pattern recognition."
– Teresa Fair-Field, OTD, OTR/L
“All of the ethical responsibility is on the user to sort of normalize. I use this tool in the same way that you would have used Microsoft Word embedded grammar checks.”
–Teresa Fair-Field, OTD, OTR/L
“You kind of have to assume if you are entering details, that they are going on a computer somewhere that you don't have control over. So that's why it's really important that you're sort of cutting and pasting things and that those things don't have explicit information in them.”
–Teresa Fair-Field, OTD, OTR/L
“That's the biggest hurdle. It's not the integration. It's connecting the existing systems we already have.”
–Teresa Fair-Field, OTD, OTR/L
“Evidence based practice is reading the research, learning about the research, taking what you know about the patient, what you know about the environment, the school…and deciding for yourself whether or not that research actually makes sense for your case.”
– Jayson Davies, MA, OTR/L
Resources
👉 Chat GPT
Episode Transcript
Expand to view the full episode transcript.
Jayson Davies
Hello and welcome to the second episode of where we are diving into school based occupational therapy and artificial intelligence. In this episode, I am joined by Teresa Fairfield, an occupational therapist who is on the forefront of incorporating AI into the profession of OT. As the Chief Clinical Officer of early markers, a company that is using AI to develop a home based mobile app to detect early risk indicators for cerebral palsy. It is pretty crazy, I definitely recommend checking it out. You can learn more about it over at early markers.com. In addition to her role at early markers, Theresa has an extensive background in pediatric OT. And as a Clinical Assistant Professor of Occupational Therapy at Baylor University. I've known Teresa for almost 10 years now, it's hard to say that like 10 years ago feels like yesterday. But she has done so much. And I've just watched her career developed over the last 10 years. And when I found out that she was working with early markers, I was just kind of blown away because I never heard of an OT working alongside of a quote unquote, tech startup. So very cool to have her joining us today. This episode is a very special one in the sense that it was originally recorded live on YouTube back in March of 2024. This year, if you're listening to it recently, because of that, you might hear us refer to some listener questions near the end of the episode. If you would like to watch the episode on YouTube, we have added a link to the show notes over at OTSchoolHouse.com slash episode 147. You're getting the same content here. It'll be a little bit more edited here because it's not live. But if you want to have that picture with the audio, you can head on over to YouTube and find it there. And one last thing before we get started, did you know that you can earn professional development hours for listening to the OT schoolhouse podcast? Yep, that's right. Members of our OT school house collaborative community can earn unlimited certificates of completion by listening to the OT school health podcast and displaying their learned knowledge with a simple quiz. To learn more about otisco House collaborative and just start earning your professional development hours for the podcast episodes. You already listened to head on over to OTSchoolHouse.com slash collab that OTSchoolHouse.com slash co LL. A B. And with that, let's dive into how AI may revolutionize school based OT in the future and the impact it is already having today. Stay tuned.
Amazing Narrator
Hello, and welcome to the OT schoolhouse podcast, your source for school based occupational therapy tips, interviews and professional development. Now to get the conversation started. Here's your host Jayson Davies. PLAs is officially in session.
Jayson Davies
All right, I see the chat is working. The chat is definitely popping up. Hello, hello, I am refreshing the YouTube page just to make sure we are good over here it is good. And we are just gonna go ahead and get started. I see so many people here. I think we had 40 people or 48 people waiting and just before we go, but I see Billy I see March. Thank you for being here. Christy KALLIOPI, a member of the OT School Health Collaborative, thank you so much for being here. Really appreciate you all. And as you note today is all about school based occupational therapy and the use of AI and school based occupational therapy. This is a very special Live episode of the OT school health podcast. And in case you're wondering, a replay will be available both right here on YouTube where you're watching it now. And we will also be releasing this as a podcast episode for the OT schoolhouse podcast as its own separate thing probably in about a month or so. So stay tuned for that if you want to catch it again, in audio format in between schools number one, two and three. So with that, let me go ahead and we're going to dive right into it. I'm going to introduce my fellow colleague here. And I'm actually going to bring her on the screen right now. So we can bring her on you can all meet Theresa and the threes with Ella was not I actually met Teresa several years ago now in the middle of a sensory integration and practice test course. Going through the USC WPS program when it was still a thing. She was just a friendly, friendly hello friendly face and we went through like four classes together over the span of a year. got lunch several times you haven't just had a good time learning together just like we are doing today. And so if I can really briefly introduce Teresa to you all. Teresa is a Clinical Assistant Professor of Occupational Therapy at Baylor University. She's been in that role since 2021 and is affiliated with early markers, a Seattle based company pewter vision startup whose work in machine learning and infant motor development is building an app based tool to provide AI supported early screening together with parent coaching and education. So and kind of see why we have her on today. She is like with this, she's working alongside this AI base program to basically support infants and infant motor development. The smartphone app is poised to improve screening access to areas underserved by skilled early interventionist, and over her nearly 30 year career. Teresa has specialized in neurological development through the lifespan to include early intervention, and infant motor skills, as well as neuro motor conditions, adulthood and neurodegenerative diseases. And she's also certified. Once again, that's where I met her as well. So, Teresa, how are you doing today?
Teresa Fairfield
Oh, very excellent. It's nice to have so many people that are interested in the AI space. And you know, it's an exciting topic. So let's get started. Yeah,
Jayson Davies
you know, and this is definitely something that is exciting. It's extremely new, we're kind of like you're saying like the wild west of AI, it's, it's a OTA has not even had time to kind of publish any sort of guidelines. As many of you know, I participated in an EO TA, forum, chat, whatever you wanna call live with them about a month ago. And that was just so awesome. We had four of us on, on there to just kind of talk about what we're seeing. And the world of AI and I kind of was the pseudo school based OT, therefore, that it's just amazing, we're seeing so many different use cases of AI. And I think that's just kind of the best place to start here. Today, we'll get into some q&a, everyone, for if you're listening, we'll do some q&a later. But we're going to start this off by just letting Teresa share a little bit about what she has seen within the realm of AI. So with that threesomes share a little bit, just kind of what you've seen out there with AI and occupational therapy in general.
Teresa Fairfield
Absolutely. So to begin with, let's just talk about what AI is, and maybe places that were already using AI and might not even know about it, that's probably a good place to start. So AI is a family of theater systems really, that's aimed at solving problems that could include learning, problem solving, decision making all kinds of pattern recognition on a on a high level. And it includes things like branches of AI like predictive AI, which is where you're using tools to actually predict an outcome or such as, you know, scanning data and seeing, you know, can we predict a specific outcome based on these large format data streams, or what what we as consumers are probably more familiar with, which is generative AI, and that is AI that is, has been trained on resources that generates data that is similar to but not identical to those resources. And those are things like chat TPT, and some things that I know that you've played around with. So in that sense, we have both sides sort of represented, and that, you know, my research in early intervention is really the predictive AI can we use an AI tool to identify those infants that would benefit from a full occupational therapy assessment or a full team assessment by early interventionists? So, you know, it's it's in nobody's interest to replace the team with AI assessment bots. However, we do know that we have a pipeline issue in the field and, you know, identifying the infants that you know, would benefit that are missing their milestones that have a certain number of motor markers that would likely qualify could help really unclog that access issue. And when we started out this work, it was shortly around the time when I met you, Jason around 2014 2015. We were using tools like Well, at that time, it was the camera that was on your Microsoft Xbox that was you know, detecting humans in play and and our team used some of those same tools to you know, begin research and say, Okay, can you know that was designed with a certain use case being adults being, you know, in the gaming world, and can we use that same technology and see if it can Identify an infant or a child and because that was sort of, you know, not a use case it was designed for and turns out, it performed beautifully. But it didn't resolve the issue of access, because of course, not everyone had a at that time, not everyone had an X Box in their home. And but, you know, the technology that was running on that planet, nobody has an Xbox, you know, Kinect in their home now, because, you know, the technology that was available then is now available on everybody's smartphone, even older smartphones are capable of, you know, producing body pose imaging, that can, you know, detect an infant and what the infant is doing with hot with very high accuracy. So, so that being said, there's, you know, there's this space of AI, but there's also this space of digital health, which is, you know, related can be aI informed, but you know, not necessarily as explicitly research or documentation, but how do we use data to solve some of Healthcare's challenges? So that's where it's really exciting and really fun to talk about.
Jayson Davies
Yeah, and I think most of us when we think of AI, right, we think about what you mentioned that I've talked about, which is mostly chat, GPT. Maybe what's it, what's it called copilot I think, is the one that Microsoft is officially going with, or Gemini for Google, we think about, you know, interacting with a chatbot. But you've opened the door. Like, there's so much more to that with machine learning, I want to I think we're gonna spend most of today talking more about like chatGPT, because I think that's what's accessible. But before we dive really into that, the other side of things, are you seeing any of that in school based OT as of right now, or in any pediatric OT for that sense, other than the project that you're working on?
Teresa Fairfield
Well, not necessarily in the clinical space, because again, we're working on validation. So the tool is built, we've looked at hundreds of infants, we're doing what's called supervised machine learning. And that means that you and and we had OTs be early intervention OTs involved with what's called setting the ground truth. So you observe 1000s of video snippets, you label them, and then essentially train the computer, what you as a clinician are observing. So in that sense, you can label it both quantitatively as well as qualitatively if we think about aspects of motor skills, we have both quantitative and qualitative data points in any motor activity. And so by training that label set, then you can deliver it to videos and content that it has not seen before. And it can, you know, detect those same movements, those same motor markers on the untrained data set. So but really, when we talk about that predictive AI, I think the important part for clinicians is to understand that clinicians are the ones doing that work in association with the programmers and software engineers, because, you know, they know the code, but they don't know what they're looking at. So it's very much a collaborative exercise. And really, all the things that we'll be talking about have even the even the GPT so a GPT is a generative stands for generative pre trained transformer. So you know, you are you are again, you are giving it data to learn from and then asking it to generate something new from that. And that's where the and there's many GPUs that do that, including ones that are specifically focused on medical health issues like Hippocratic AI is a health GPT. And it's been trained using actual nursing and and health care providers. So that it's not just programmers that are that are delivering this content. It's not just and I think it's important to kind of pull back the curtain and make sure that people know that a there's seats at the table if you're interested in this work, because they need us to produce a product that's relevant. And they need our insights in terms of understanding what the problem is that we face.
Jayson Davies
Yeah, yeah. And I'm just trying to think of some like use cases and throw some out for me like obviously, I think people understand the GPT is because we can use those to generate goals. We can use this to generate we'll talk more about this but goals or treatment plans evaluations help us to write up notes and all that good stuff but the others side of it, what do you see as some pie in the sky 10 years from now activities that that will be supporting us on a daily, weekly basis. Just you don't need to go into detail, but just, you know, things that we know of already today?
Teresa Fairfield
Absolutely, I think, you know, I think that what it makes possible is for us to be spending our time on not only where we want to be spending our time, which is clinical decision making, and you know, but it's a nice decision support tool, in terms of, you know, how do I triage this issue, and that, you know, kind of materially comes in to some of those workload caseload discussions that I know you have on your platform quite a bit in that, you know, not just what can we take off of our plate, as you know, we're spending more time doing activities that are not value added direct to the client, but also, what becomes possible in terms of things that we haven't thought about yet. And that's some of the conversations that I've had in the digital health community is, you know, we don't have you know, calls going out to families, we don't have like, there's a lot of sort of connection between school and home or clinic and home or environment at home, like there's a lot of connection that doesn't exist, because we don't have the ability to, you know, perform that ourselves. And that's where we can look at a chatbot. And, and, you know, the language support is really, really taking a giant leap forward. And, you know, we can expect that, you know, all of those gaps in communication could potentially be closed.
Jayson Davies
All right, I have a hypothetical for you, how long would it take me potentially today? Or I don't know, if it's not possible today, to potentially create something where I upload a student's writing sample and AI is able to, quote unquote, graded on legibility, based upon maybe letter size, spacing, placement, all that good stuff. Is that something that could probably be produced with a GPT? today?
Teresa Fairfield
Yeah, yes, and certainly with, you know, certainly with image detection, now, your GPS that you're using, don't have that sort of image databank. But, you know, again, if you train it here is the, the range of typical, then whether it's handwriting or motor milestones, a computer's going to be able to, you know, very closely identify, here is your normative band. And we already have that with our OT assessments that, you know, they're moving towards, you know, look at the PDMS. Three, it's, you know, it's all online scoring. So, you know, and the, and the bot three coming out soon, near you within the next few years near you, is all computerized scoring. And so it's really a very tiny jump before you'd be able to just scan in sheets and, and receive data. And in that sense, like, the, you know, I saw a chat, I saw a comment pop up about clinical reasoning. And, of course, we're going to spend a lot of time talking about that, as well as the ethics of some of these scenarios. But, you know, again, we already have the it's not the actual delivery of the counting the number of below line letters like that is not where our skill is necessary. You know, that is a very mechanical tasks that can be performed by, you know, a trained individual of any variety to say, Okay, how many, you know, mixed case letter formations do we have in this student sample, that's not where the skill exists, the skill is in, you know, all of the aspects that affect handwriting, everything from you know, seating, postural control, you know, all of those things, and, and similarly to some of the conversations, you know, even looking at vestibular equipment, like the swing is not the OT, the OT brings the OT, this swing is just a tool. And you know, anybody can put the kid in a swing and they're not performing OT. So I think it I think it requires us to really focus in on our clinical reasoning skills as well as perfect our ability to communicate them.
Jayson Davies
Yeah. Yeah, that's funny that you bring that up the whole idea that like, yeah, AI can score an assessment, but so can just about anyone who can read the manual and go through, like, especially standardized assessment and tells you exactly what to say, right? Like, I mean, we can put a ruler over the circle and see if the circle is cut like, you don't need to be an OT to determine if it's cut within one quarter inch of a line. And that can save us a lot of time time that we could be spending with teachers spending with parents spending with kids, giving them the information that they need, or, or walking them off a cliff because AI told them something wrong. We need to clarify a little bit. But yeah, there's so many generative are so many, so many ways to utilize this to give us more time, honestly, which is which is very important. Let's talk about ethics, not just necessarily related to school based OT, but just in general, what are some things that maybe we should be very aware of when it comes to AI as it currently is today?
Teresa Fairfield
Absolutely, certainly, I think the and as an educator, there's a lot of and talking with other educators, there's a lot of initially sort of reactive response to how do we keep our students from using this. And that's really not a sustainable approach. Because it's a tool, it has the ability to improve your quality of life as a clinician, and as a student, and, and again, like, equity in the space is important too. And so, you know, I feel like AI is is has the ability to bring us to bring so many good ideas to the forefront. But but really the the onus is on us, as individuals to declare that we've used AI and you can, you know, you can submit a manuscript to a peer reviewed journal, you can give it a draft, you can and you can indicate and are expected to indicate where AI has been used in the development of that manuscript. So that's already happened in the American Medical Association Journal, where, you know, it is assumed that with these large language models, we have the ability to improve our sentence structure, we have the ability to, you know, look for sort of holes in our argument, and then buttress them. And it's important to understand that our results are still our results. Those are things that, you know, we achieved based on our human in invention and innovation. And if we use a large language model for refinement of our results section, and we declare that, you know, that is widely acceptable within within evidence based practice. So again, the AI is not generating data, that's, that's outside of its role. But it does have the ability to, you know, as, as Amy's mentioning, in the chat, it does have the ability to improve the professional writing voice and you know, assist you as the author from taking something from potentially unpublishable, to to publishable levels. And it's, it's a declaration and again, we as all of the ethical responsibility is on this on the user to sort of normalize, you know, I use this tool in the same way that you know, you would have used Microsoft Word embedded grammar checks 10 years ago.
Jayson Davies
So but what if we get out of the idea of formal manuscripts and whatnot, what if we're talking about our daily treatment notes or an evaluation that we're completing, whether it's school based or in a hospital, we use chat GPT, just because we gave it the bot scores, and it was able to summarize that a little bit quicker, honestly, than we could do you think at that point? Is it becoming a norm to say, you know, what, this report includes AI generated content, or is this separate from manuscripts?
Teresa Fairfield
Well, I think I think that's where we probably should go in the sense that if you and again 510 years ago, if you were using the online scoring software of the bot, it produced a report for you. Yeah, from your data alone. So you know, and we did that all the time. You could print short form reports, you could print long form reports after only having having given it the bot data. Now, the issue is, it's very clear that you didn't write that, right that's coming from the bot A company from the assessment company. So and I think the challenge, I never really used a lot of those reports because they weren't parent friendly. They weren't parent centric, they weren't at an appropriate reading level for the parent. And again, you can use the large language module model AI to, to set whatever reading level you specify, which has, you know, an incredible opportunity for health literacy. But it's still the onus is still on the therapists to ensure understanding to provide education to provide generalization and relevance. But again, it really starts to shape. What is the skilled service? The skilled service is the assessment? It's not necessarily the manual scoring. And it's the interpretation of the results.
Jayson Davies
Yeah, absolutely. And, you know, this is something that this is why we're having the discussion, because this isn't set in stone yet. AI has only I mean, chat GPT opened the doors like a year and a half ago. So this is all new, a OTA is still trying to figure out where to go with guidance. At the OT school house, I'm definitely using AI. And I know many OTs are someone mentioned in the chat earlier, one of the tools that they use as magic school.ai, which is its AI designed for teachers really, and related service providers, but it's really more geared to our teachers and admin, but I know that OTs are using it for the same purpose. I don't think anyone fully knows how to quote unquote, site AI out of maybe the more formal manuscripts, like people are still trying to figure that out. The same way that everyone tried to figure out what to do when Google took over the road. Like, you know, this is a tool that we are trying to use. Now, we obviously need to talk a little bit about HIPAA. In the schools, we also have FERPA which is basically just the educational version of HIPAA. But what are some things that we should know in relationship to AI? When it comes to HIPAA and or FERPA? Could they kind of go together?
Teresa Fairfield
Absolutely, I think it's important to understand how large language models work and the computer is, if you test it, it does not give you any revealing information about how it has learned what it has learned. So in that sense, and that's why there's some gray area around some of the lawsuits that have already come into chat GPT and other large language models, because it is like if you ask it, if it knows this, or if it knows a particular thing or a particular person, you know, there's filters in place that say I you know, I am not able to do that I do not provide that information. If you ask it about you know Winnie the Pooh characters, or to write a story involving Winnie the Pooh, you can detect that it's probably heard of Winnie the Pooh before, that it's been trained on a Winnie the Pooh story. But it's not in the sense that you can Google a specific name and sort of deep dive and find out more information. It doesn't give you specific information about an individual a case. That being said, does it have the official stamp of HIPAA or FERPA approval? I don't think so. But you're you know, you shouldn't be feeding it client specific data anyway. What you should be using it for is describing the scenario. I am a school based occupational therapist, I am working with a client that's six years old with the following challenges. Here's what I have, in terms of findings, you know, helped me write a summary paragraph or a concluding paragraph helped me develop notes for my IEP meeting. Like it can do all of those things. But you have to set up the scenario. It's not so easy. And I think one of the challenges is that we go into Google and we put four or five search terms up there, and we expect magic to come out. And you know, it's a garbage in garbage out system. So if you, you know, the more time you have sort of describe the scenario and what you what you want it to do, and the better better your outcome better your output.
Jayson Davies
Yeah, yeah, right. We'll get into that crystal. We're going to get into goals in a little bit. We're trying to do more of an overview first, and then we're going to really dive into the special Perfect. So I really do want to ask you, because I know the HIPAA FERPA ethical areas is the main. I mean, it's the main holdup for a lot of reasons, right? Like, I think that OT practitioners are the most ethical people in the world, to be honest. And like, we really want to make sure that this is safe for us and our clients. And so when we are using AI, say, I'm using chat gptc, I'm using copilot What should I assume? is the default to them keeping data? I mean, like, if I type in a student's name, is that name going to be in there forever? Is it going to learn off that name? Do we not know, do we know? Like, what should we assume I guess when we're using AI, as far as data collection,
Teresa Fairfield
I think anytime you put something into your system, you should assume that you have lost control of that information. I think that's just sort of standard practice now. And that's why that's why a lot of like, I just listened to a Male Clinic speaker, and they have set up and they're working a lot in the area of mobile health. But they have identified that in order to do that. And they're they're working on relationships around the world with clinics on every single continent. But as a result, they are saying we are only using our own data, our own internal data, they are not sending it out, they are not sharing it, they have locked it down and what they call a behind the glass model. So you know, you can look it on their data, but it's not coming. And it's not leaving the system at all. So I think you know, I think it's reasonable to expect. And again, I have a presence on the Internet, you can Google me and things pop up. And you know, I think I think a lot of us do, particularly when we're working in these innovative spaces, but you have to kind of assume that if you're entering details that they are going on a computer somewhere that you don't have control over. So that's why it's really important that you're sort of cutting and pasting things, and that those things don't have explicit information in them.
Jayson Davies
Yeah, and I mean, I go as far as just using pseudonyms, I just default to using John for everything. Like John is a fourth grader who can do all this, please help me with John. So that's one way to definitely do it. I know some programs, I'm assuming magic AI or magic school AI but I know ever, ever ability I believe it's called, they were able to basically partner with Chet GPT and certify that they have HIPAA compliance. And what that meant was that they can basically guarantee that the input that you the inputs you give it are not being trained extra. I don't want to say it wrong, but I can guarantee stored Yeah, it's basically been deleted. And so there is a way to go about making sure that you do use a HIPAA compliant or a FERPA compliant, I believe FERPA is more difficult because I don't think there are like FERPA, it's a government agency that just doesn't do a whole lot. But if something's HIPAA compliant, you can almost kind of assume that it's FERPA.
Teresa Fairfield
The higher, stringent, yeah, and no, and one way around that and one way, and then of course, in our research, we're using videos of infants. So we have been up one side and down the other with the IRB around security of the data safety of the information. And one way that you can do that is that all of the the identifying information stays local on the device, ie local on your laptop or your smartphone, and none of it actually gets transmitted to the cloud, or it's encrypted by the time it goes to the cloud. So that in our situation, when the parent takes the video of the infant, the video stays on the parents phone, along with you know, 4000 other videos, the parent has taken up their infant, but when they're what they're uploading to, in our case, Amazon Web Services is just the data, just the data around the movement. There's no video showing the child's image, there's no protected health information of the child's name or date of birth. So it's all sort of, you know, locally processed, and that's another thing that mobile health companies are focused on is how to do more sort of processing on the device itself. So that the only thing that goes out is encrypted data that cannot be He tracked back to an individual.
Jayson Davies
Yeah, I mean, I know that's I think that's part of the reason that I know, what's his name the same ultimate? No, I don't know, the guy who owns Chet GPT, like he is trying to mass produce a bunch of the chips that are required for AI, because it's just computers are going to are going to need those chips in order for us to process AI, I think in the future, which, yeah, it sounds crazy. But anyways, we've mentioned ethics, we mentioned HIPAA, FERPA, all that good stuff. Are there others? Or are you seeing other challenges or misconceptions surrounding AI that are maybe preventing? Not just OTs, but just in general healthcare use of AI? And is there anything that we can work to overcome?
Teresa Fairfield
I think, you know, all of the people that are with us today live, all of the people that are going to be watching this recording are, I think, absolutely in the right place to be on the forefront of this because, you know, you're here to find out more. And certainly, there's early adopters, and then there's, and then there's people that we're bringing along behind us that you know, when this is on everybody's phone, then they'll begin to start and pay attention. But But I feel like, again, there needs to be more presence in OT there needs to be. But there's so much sort of interdisciplinary focus on these issues that I think learning more keeping your ear out. One thing that's interesting is that, when you're thinking about what is possible in the healthcare system, one thing to be aware of is that it might be possible, but one of the biggest challenges within the system is how they communicate with each other. And currently, they don't. And so that's one of the biggest issues is that if you show up at an ER in another country, they don't have access to your Fitbit data, your provider doesn't have a way of moving your, you know, heart rate information easily into your medical chart. The EMR companies are not speaking with each other. And so we tend to think that these things are possible. And they are but it's still extremely siloed, where these things are taking place. And so a world where all of this data is available to improve our lives as practitioners as well as consumers of health care. We're just not there yet. And that's the biggest hurdle. It's not the it's not the innovation. It's connecting the existing systems we already have.
Jayson Davies
Interesting. Interesting. All right. All right. Let's dive into being more specific about school based IoT we've kind of touched the surface, but let's dive into it here. I'm using I'm using AI I know many other people have already mentioned they're using AI, what are some of the most common uses that you're hearing of school based OT pediatric coachees? Using AI? I'm assuming primarily GPS. But yeah, what I just list them off.
Teresa Fairfield
Absolutely, I would say primarily the large language models, because again, that is some of the heaviest lifting of our job is writing goals, updating materials, some of the EMRs. And the and the IEP software have really restrictive sort of textboxes and really restrictive. And so you know, it's really a helpful tool in that regard. There isn't a lot of innovation in pediatrics in some of the predictive modeling. Again, ours is really unique focused on infants, but there's not a lot of it, that is looking at, like you described, just automatic scoring of, you know, fine motor work or work products of, of the child. And again, that's only one tiny little thing that OTs are doing, but gosh, if it were easier, it would be a big game changer, right. But, you know, they're not being designed with a pediatric population in mind. And so, I think the, I think the text generating solutions are going to be the most high value because again, think about how many times you would even just like having an outline. I need an outline to provide a teacher some, you know, self regulating activities. And, you know, all these scenarios in which, you know, you could produce really customized sorts of content for teachers, for parents for, for other, you know, conference submissions, you're, you're still needing to provide all of the data that goes into those. But but, you know, I think that's really going to alleviate some of our biggest pain points.
Jayson Davies
Yeah. I mean, to be to be like, I don't know, I'm waiting for the day that WPS says, Here's a $10 monitor, it's probably closer to $100 month subscription to access our WPS bot that has all the data on all of our assessments. And you can ask it anything like I know it's coming. I'm sure that in the works of it, it's got to be. But yeah, I think that like for me, we have what's called the OT school house, goal bank, and I'm absolutely using AI to take my goals. And based upon my goals that I have used over time, and even goals that others are submitting, were saying, Alright, AI, here's the goal, please turn it into a template. And it's turning that into a template. So now we're using it as a goal being and from there, I can say, Okay, well, if this is my goal, what might a baseline on an IEP look like for this for this goal? And so now it's creating a draft baseline. And from there, I can ask it, okay, well, what about accommodations? What accommodations might be able to support a student with this goal? What type of services might be able to support a student with those goals? And, you know, just like with everything else, I mean, you think about evidence based practice, right? Evidence Based Practice is not going out, finding the research and replicating the research, that's research, evidence based practices, reading the research, learning about the research, taking what you know about the patient, what you know, about the environmental school and all that, and deciding for yourself whether or not that research that evidence actually makes sense for your case. And I think we all obviously, you have to read anything that comes back from Ai, with a grain of salt for the exact reason you said, right, like, it doesn't have all the knowledge that we think it might like, it hasn't read everything on the world. I know, it's read a lot. But it hasn't read everything just quite yet. So yeah, that's just one example that I can think of, for school base, Oh, Jesus, the goals.
Teresa Fairfield
Absolutely. And and very similarly, like, if you have any ethical quandaries, about using an LLM to generate your goals, it's the same process as using a goal bank of you know, here's all and, you know, I did that for years, here's goals I've already written, and saving myself some time, you know, you reflect or you, you know, copy and paste, and then modify, you know, like those sorts of old school techniques we've been using for years. And do you need as a therapist to modify, if you're taking this description of what a sensory seeker means and, and sort of translating it for another child, and another scenario and another SPM score? Absolutely. But it's a very similar process of, you know, taking a goal bank or, or, you know, finishing a document that you felt really proud of, and saving it for later as a template. And, you know, this is very much like that, where you're, you know, needing to sort of reduce your sort of cognitive load, if you will, as a school based therapist, because, you know, the things you need to be responsible for are tremendous.
Jayson Davies
yeah, yeah. So, so, when you're using when OTs Kobe Sochi, right, they just had a full day of seeing clients and whatnot, and how can they remain that maybe they're jumping on chat up to you to ask you for some support with treatment strategies, they want to plan their next week treatment plan? How can they ensure that interventions are remaining client centered and individualized for each student?
Teresa Fairfield
Um, I think that's, you know, I saw I'm, I'm paying as much attention as I'm able to the chat, but that was one that was one thing that I saw earlier is you can't just give it demographic information. You can't just say you have a child and a diagnosis and expect it to know what that child looks like. Because, you know, again, there's other aspects that are client centered that it will not provide. And so you know, the more sort of context that you can add or that you can come up with, including, you know, in preparation for this. Today, I gave chat GPT for some, some prompts about coming up with a treatment activity of a four student group focused on social skills. And, you know, here's the reading level, I want it to be about three minutes. And so, you know, we're focused on basic social skills with, you know, around the prompt of what are you doing this weekend. And, you know, it produced line by line script that, you know, the students could practice with each other in sort of a skit sort of way in terms of roleplay. And I think, you know, even just taking that activity, which I just kind of threw up there to play with, it just shows you, you know, things that you could produce, I've seen a lot of things happening in the social stories, absolutely. Like, you know, the that content, it's easy to, it's actually the most fun part of social story building, to which I've done a lot of to do, you know, take the pictures or the video snippets to actually work with the child and family over what those what social stories are necessary or, or with the teacher and in school base cases, and then have have the large language model just punch out what you know, what script needs to accompany that. And the, you know, language generating tools are incredibly they're improving at an incredible rate. They currently like all of the text to speech still still sounds a little bit robotic. But I just listened to some this weekend down at Stanford, and it's pretty undetectable.
Jayson Davies
Yeah, yeah, it's, it's pretty amazing where we're going. And so I do want to definitely keep time for some q&a. And so I kind of want to get into some quote unquote, maybe rapid fire. If you were working on this right now. And you had chat CPT open, what would you do? Alright, so situation, you're working on an evaluation, you've got some data in front of you, you are working on maybe your interpretations at the end of your result? How are you going to ask Chet GPT? For some support?
Teresa Fairfield
Me? Are you asking the chat?
Jayson Davies
No, you I want to I want to know this. Yeah. So what what might you I mean, everyone can, everyone can play a long Feel free?
Teresa Fairfield
Well, what I would what I would do, first of all, is I would upload a de identified sample of something that I had already written. Right, this is me, this is this is how Theresa sounds when she writes a report. And you know, these are some of the contents you can and then you can you ask it to use that as the model. So then it takes everything that is learned, but it's specifically just learned from you individually. And I think that's an important part. Since you know, you want it to sound like you. And then you I think a good place to go would be to describe what you're trying to achieve. And it's the same thing I would tell my students actually, who's the audience for the work? Yeah. Is it a teacher? Is it a parent? What's the reading level? What's the vocabulary? Are there any terms that need defining you could kind of assume that that chat would know a lot of these terms because of the model that's been trained on. But if there's any, any sorts of vocabulary or jargon that you're using? Obviously, you can't give it any direct quotes, you'd be doing some of that editing. afterwards. I find when I'm doing that sort of thing, even just having a text to start with, it's much faster, it's much easier to edit something, then start with a blank page.
Jayson Davies
Yeah, I have found that especially I don't know, I use chat. GPT. Mostly I don't use copilot or Gemini, but I found that chat. GPT likes to use the word Delve, I delve into something so if you see the word Delve, I would question if it's a I like that's just my personal I don't know, maybe your kitchen something else. But either way you need to edit whatever it gives you you need to edit it, you need to make sure it actually sounds more like you what Theresa said was a great idea by uploading your own work to it, it is being trained on you or on your words. But still, usually you got to edit it just a little bit. Alright, so that was evaluation.
Teresa Fairfield
But do you want to do you want the prompt I actually gave Chad GPT?
Jayson Davies
I mean, if you want to go for it.
Teresa Fairfield
It says I am an occupational therapist working in the school system. I'm working with a group of 15 year old students with Level Two autism that have difficulty with appropriate social communication. I'm interested in setting up a roleplay activity where peers will read a script that models the appropriate interactions between students. I'd like you to provide the script for a group of four teenage students with a third grade reading level. The scenario should involve the four teens sitting at the table at lunchtime, asking each other what they like to do on the weekend. The scenario should take about three minutes, please provide the script.
Jayson Davies
Yeah. And I just, I just also put up a chat comment up here from Heather. And it should be easier to read now that I change the formatting. But this is basically what Theresa just did. She gave it a role. She gave it context, she gave it action, she gave it formats and constraints, you really want to do that. I mean, a lot of us just want to give it a quick sentence and and expect it to give us something amazing. And that's just not the way it happens. I mean, like, it's almost like trying to tell you to do an evaluation as an 18 year old before you have any OTs experience, and then telling you to do an evaluation, once you have the training, what's your taking OT 5015025035040 Those classes now you have much more information available to us. So yeah, absolutely got it. Gotta make sure we prompt it. Alright, so that was a little bit about evaluations, and also a little bit of treatment planning there. What if you wanted to let's see, what if you wanted to find out a support that might be helpful for a student? How might you interact with AI?
Teresa Fairfield
In that case, I would provide some similar to, you know, what just got posted? What's the what's the context? What's the length? What's the complexity? How many steps? For example, you know, one thing that I haven't tried, but I am excited to see some folks in the in the chat that have already hopped over and tried it for the first time, How fantastic is that? Glad to have you jumping in the pool. And, you know, again, you can just use name, you can just use a placeholder, you you can you know, the but again, the ideas come from you. And that's where if you give sort of an idea, it could come up with some ideas, the relevance of those ideas may not be that helpful. But I think for like student centered, like, here's where you can use it. to buttress your, you know, let's say you're not so skilled on the pokey Milan characters. You could ask the you could ask the chat to generate a game involving five Pokemon characters, and, you know, three rules that could be played in 10 minutes or less. And again, you're providing the bumpers on that activity. But you know, think about your time savings if if pokimane is not your jam, by asking sort of those specific sorts of questions.
Jayson Davies
Yeah. All right. We're gonna dive into some q&a Right after this last question. But I just kind of want to prompt everyone. Thank you so much for being here. First of all, we are going to actually post into the chat a link to get a certificate of attendance for being here. So stay tuned for that. But I want to ask you for one thing, and that is to please let me know in the chat, what you most want to learn about AI now that you've attended today, like what is the one thing that you wish that you could snap your fingers, and tomorrow, you would just know how to do please leave that in the chat for us. And with that, I'm going to ask Teresa, if you could snap your fingers. And tomorrow a tool was built for you related to AI. What would that be? Hopefully related to OT but who knows?
Teresa Fairfield
Yes, I think you know, again, my vision for the project that we're working on is is so huge. And we've done so many focus groups with parents and early intervention. I know some of you are working with littles as well. And there's so much opportunity for also sort of supporting the clinician and the parent with like a micro konsult. And so my vision is that, you know, that parental sort of anxiety or caregiver anxiety around development is resolved with a combination of machine learning insights, a beautiful parent education platform, and an entire, you know, team at armada of occupational therapists that could, you know, provide sort of micro consults support that puts in a meaningful way to really resolve some of our, you know, equity and access issues around early intervention. And, you know, as we know, even as school based therapists you're, you're seeing kiddos later on, after that early intervention window is closed. And, you know, what, what if? What if they had access to all three years of Birth to Three services? Like, what would their school experience look like?
Jayson Davies
Yeah, yeah, I kind of already mentioned mine earlier. So I won't go too much into it. But it'd be like a tool that just literally can tell me the exact legibility of this writing, and then also some strategies to move forward. But to whoever just gave us a thumbs down and then change it to a thumbs up. Thank you, I appreciate you, and to anyone else who wants to give us a thumbs up. Thank you so much, I appreciate it. Alright, so I've got some questions here. And feel free to also add a few more to the chat, we might have time. The first one that I want to cover is, which AI platform might you recommend for getting started, we've talked a lot about chat, it's up to you is that is that your go to?
Teresa Fairfield
That's my go to I have decided personally, that chat for subscription is is worth my investment. As a clinician, I feel like you know, certainly try around with the free tools and see you see what might be possible. And then you know, there are a you can have in terms of a free trial, you can compare what three and 3.5 gives you what four gives you. And keep in mind, there's also a bandwidth issue, sometimes you'll come to it and it will sound really impaired in terms of this response. And then And then, you know, and again, keep in mind that this is essentially a crowdsourcing issue, we're all going to one well, and when more of us are trying to get our problem solved, it can get a little murky in there, but but play around and again, keep in mind, it is a computer, the more you give it in terms of guidance, the more it's going to help you out.
Jayson Davies
Yeah, exactly. I found that I mean, you really have to learn how to prompt and a lot of people are asking for prompt goals or for prompting help. So we'll have to come up with something for that. But crystal asks, Does it write goals? And I want to get your take on? I don't know if you've used it much to write goals. I obviously talked a little bit about earlier, but your concept on that?
Teresa Fairfield
Yeah, I think I think it can write goals. Again, if you give it parameters, and you know, maybe maybe a nice supplement. Jason, you and I are talking about more ways to intersect on this content. So I think something that would be fantastic is to provide some more sort of prompt specific support. And maybe you know, some case study resources that you can put out to your, your members and subscribers about you know how to begin shaping this tool for themselves.
Jayson Davies
Yeah, yeah, definitely. I've got another question here from Amy. Amy. Thanks for joining us, a alumni of the OT school house Podcast. I'm trying to put it on the screen right now self determined kids shout out finding research articles that I need on a particular topic. Have you explored anything with that? I've used consensus, which is an app out there. What have you used?
Teresa Fairfield
Yeah, I think that's I think it's great. The issue is that there's still these products still need funding. So you know, all of the all of the places that we get our articles still need financial support. So libraries are getting them and they're paying for them. And you know, you still have to go to your library and chat GPT will not help you get out behind a paywall. It is absolutely blocked from providing, can it help write code, it can absolutely help write code, but it will not help you write code to get behind a paywall. So it does have some ethical boundaries for itself as well.
Jayson Davies
You know, that's interesting, because I'm a big daily listener of the podcast, and they've had journalists talk about this exact topic. And while they couldn't get it to, like, recreate the article, well, I should take that back. Someone did, but then they shut it off because of that. But one way that people were getting around it was saying write me an article about this topic in the voice of said journalist, and it didn't know that journalist voice and it knew what that journalist knew. So then it just recreated a journal that was very, very similar, but yeah, very interested in how things are working.
Teresa Fairfield
I would say though, that anything it would create does not have the peer reviewed rigor of something that's actually published in the journal and that's, you know, the names the the hallmark of evidence based practice is the peer review process, not just the writing one.
Jayson Davies
Yeah, yeah, absolutely. All right, well, it is 359 My time Here in California, I think we're just about to wrap up unless there was any pieces of the comments that maybe you wanted to address really quickly before we say goodbye to everybody.
Teresa Fairfield
Um
Jayson Davies
I know they were going by fast. So
Teresa Fairfield
I only got a sense of them. But you know, what I love is that it's an active community, and they're generating great ideas and support for each other. That's absolutely fantastic. And, you know, I hope to be back again at some other point in the conversation.
Jayson Davies
Yeah, thank you so much. And Marcia, to answer your question about consensus. It's basically AI for research, you type in a question, and then it searches research for your answers, and will cite the research that is giving you an answer based off of so yeah, there are so many tools out there. Like I feel like we just need to put together a cheat sheet on different tools. Maybe we can get that together in the near future. But yeah, thank you all everyone so much for being here. Really appreciate it. We had over 400 people registered for this 200 People nearly that attended today. Obviously AI and school based OT AI and OT in general, it's just a huge thing we are on or we are in the very beginning phases of this. Be sure to look out for information for myself for information from Teresa about a I really quickly Theresa, if anyone wants to get in touch with you learn more about you find you somewhere on the interweb without asking chat GPT where can they do that?
Teresa Fairfield
You can you can Google me Like I said, I am known to the internet. And I am available at Teresa underscore fair hyphen field@baylor.edu.
Jayson Davies
Yes. Geralyn called you TFF. I don't think people will find you that way. But maybe we are calling you TFF or T fields. I like T fields. All right. Anyways, Teresa, thank you so much for joining us. Thank you, everyone, for joining us so much, really appreciate it. And we'll see you next time on the T schoolhouse podcast. Take care, everyone. Good night. All right, and that is going to wrap up this very special episode with Theresa Fairfield. Thank you all so much for listening, I really hope you appreciate this. I hope that this episode helped you to kind of understand AI as a tool as opposed to something that we should fear and is going to take off or take over all of our jobs. In fact, I was just looking up an article. And it's actually in the Canadian Journal of Occupational Therapy from 2018. So several years ago before AI was really a thing. And they can talk about how how AI isn't really a threat to overtake the role of occupational therapy practitioners. However, AI is going to make it harder for many of our clients that people that we work with on a daily basis to keep up with some of the work demands. And I think this is going to be huge for our youth today, they're really going to have to learn how to use AI, the same way that so many of us had to learn how to use Microsoft Word, learn how to use Google Drive, and how to use Google search terms of find what we wanted to find what we needed to find to get through our daily lives over the last 20 years. And I really think that that is important. We can't just you know, ignore it. We need to embrace it and figure out how we are going to use AI in practice the same way that we might use Google or we might use Google Drive Microsoft Word Canva all those tools that we do use every day. So thank you again, so much for listening. And if you would like to earn your certificate of completion for listen to this podcast, be sure to check out the OT schoolhouse collaborative. That's at OTSchoolHouse.com slash Code Lab. Thank you so much, and we'll see you next time. Take care.
Amazing Narrator
Thank you for listening to the OT schoolhouse podcast. For more ways to help you and your students succeed right now. Head on over to OT schoolhouse.com Until next time, class is dismissed.
Click on the file below to download the transcript to your device.
Be sure to subscribe to the OT Schoolhouse email list & get access to our free downloads of Gray-Space paper and the Occupational Profile for school-based OTs. Subscribe now!
Thanks for visiting the podcast show notes! If you enjoyed this episode, be sure to subscribe on Apple Podcasts, Google Podcast, Spotify, or wherever you listen to podcasts
