Artificial Intelligence, RBTs, BCBAs and Treatment Plans--Special Guest Michael Gao from Alpaca Health

AI has been injected into every aspect of our lives and its been a fast and furious deployment. While the speed of the technology is enticing, it quickly raises questions about the individuality and client specificity of AI output, especially for essential artifacts like medical notes. While medical notes are a well-known nightmare to ABA providers---lengthy, confusing, seemingly meant to confound and confuse, and clearly installed to defer payment wherever possible--they also serve a very important purpose for clinicians, who truly use them to document procedure, protocol and progress. Enter Michael Gao from Alpaca Health. He joins ABA on Tap to explain his solution and innovation. Despite Mike and Dan's initial skepticism, Michael elucidates his product and presents a convincing stance on its advantages and specificity to helping ABA professionals.

This is a light, refreshing and prospectively thirst quenching brew so enjoy a couple if you must. So keep running those sessions, take good data, consider AI for your notes, and always analyze responsibly.

Transcript

SPEAKER_00: 0:00

Welcome to ABA on Tap, where our goal is to find the best recipe to brew the smoothest, coldest, and best tasting ABA around. I'm Dan Lowry with Mike Rubio, and join us on our journey as we look back into the ingredients to form the best concoction of ABA on tap. In this podcast, we will talk about the history of the ABA brew, how much to consume to achieve the optimum buzz while not getting too drunk, and the recommended pairings to bring to the table. So without further ado, sit back, relax, and always analyze responsibly. All

SPEAKER_03: 0:54

right, all right, and welcome yet again to another installment of ABA on Tap. I am your co-host, Mike Rubio. along with Mr. Daniel Lowery, as usual. Mr. Dan, how you doing, sir?

SPEAKER_00: 1:05

Good to see you, Mike. I'm doing great. Excited to hear you navigate this AI discussion today, buddy.

SPEAKER_03: 1:11

Well, you jumped right in, yeah. You're alluding to the fact that I haven't been much of a fan. I'm also not a fan of medical notes. said, I am not a fan of medical notes. In fact, I love to talk about the fact that in my 20 plus years of practice, this idea of time conversion and completing notes has been an ongoing issue and not at one place that I've worked at ever can I say that we figured it out. A lot of this now deals with the complexities, right? This idea you've got a funding source, they want to know certain things, why they want to know those things. Maybe we'll get into those motives later.

SPEAKER_00: 2:15

And they're not going to pay you to do it. Wow. Yeah, that's correct. Or it comes out of the client time. Either way, right?

SPEAKER_03: 2:21

But today, we've got a very special guest, Mr. Michael Gao from Alpaca Health, and he's got a solution for us. I think he's got this figured out, and we're going to help him figure more of this out as we go through this conversation because he's going to make our lives easier. Michael, welcome to ABA on Tap. Thank you for your time, sir.

SPEAKER_01: 2:39

Thank you, guys, for having me. I'm excited.

SPEAKER_03: 2:42

All right. Just pardon the... Pardon the Zoom interruption there. We forgot to turn that on. So we're officially starting now. All good, yeah. So you're working on integrating language models, AI, as we're commonly calling it, into making the life of professionals in ABA or maybe any other medical profession a little bit easier to make sure that those extra 15, 20, 25 minutes, depending on which funding source you're writing a note for, after you've spent a better part of two hours doing dealing with challenging behavior and really putting all of yourself into the session. And now you've got this final task, which often gets put off. It's complicated. It's important because you want to capture what you did and you want to capture aspects of the data. And then again, it's coming at the end of a rigorous situation. And a lot of us put that off and then you put that off. And before you know it, you've got a week's worth of medical notes to complete. So tell us a little bit about yourself, a little bit about your background and then what got you to this particular project please

SPEAKER_01: 3:49

Yeah, absolutely. So before Alpaca Health, I was building an online education company called Dewey Smart, where I was able to meet and work with a number of families who had kids on the spectrum in special needs programs. And we worked with all kinds of parents, and parents of all kinds are always stressed out, but parents of special needs kids seem the most stressed out. And I just remember walking through how they should handle getting a learning accommodation for a standard test or

SPEAKER_02: 4:20

trying

SPEAKER_01: 4:20

to help them deal with paperwork from school and just thought that there was a lot more that we could do. And so earlier this year, started to kind of dig more into the clinical side of helping special needs kids, the autism spectrum, found out about ABA and started talking to clinicians. And when I realized how much paperwork y'all are doing on top of spending time in sessions, watching sessions, seeing how chaotic they often were, I was like, wow, there's a lot we could do better here.

SPEAKER_00: 4:52

Well, the good thing about our paperwork is that it's unpaid. So that makes it a little bit easier to deal with.

SPEAKER_01: 4:59

That's awesome.

SPEAKER_00: 5:01

We love a lot of paperwork, right? Michael, can you just tell us maybe a little bit about yourself? I know you mentioned your previous company that you were with, but just a little bit about your background and anything that you think might be relevant for our listeners to know about you, man?

UNKNOWN: 5:15

Yeah.

SPEAKER_01: 5:16

Yeah, yeah. Well, I was born and raised in Dallas and went to a Title I school. And so I was always really interested in all of these things that rich kids got, but the Title I low-income students didn't. And that's honestly why I started the education business originally. There's ways to make this advice that is so hidden and specialized more common to more people. Went to New York for school, went to Columbia, studied computer science. Nice. I think I'm a decent programmer, although I think that as I've kind of gone through my career, other things have started to become more interesting. You graduated from

SPEAKER_03: 5:56

Columbia with a computer science degree. I'm going to say you're a good programmer. We're just going to leave it at that. Come on. Come on, man. Columbia is a nice, small, somewhat prestigious school.

SPEAKER_01: 6:08

Although I'll say ChatGPT is probably better at programming than every single engineer out

SPEAKER_00: 6:14

there. Oh, shots fired. I thought you said you weren't going to say any controversial stuff. About

SPEAKER_03: 6:21

ABA. He wasn't going to say anything controversial about ABA. Okay. Excellent. So, um, yeah, continue. So you, uh, CS at Columbia, did you do, uh, work in computer science for a little while as well?

SPEAKER_01: 6:35

Yeah. Yeah. Spent some time as an engineer, did some product manager work in, in, um, financial technology and really just found that kind of unfulfilling. Okay. And so all the while, while doing that, I was doing this education thing and really able to see, um, how we could change kiddos lives. Like, like honestly, like one of the first kids I worked with wanted to become a chemical engineer. engineer. She had no interest in chemical engineering. She hated STEM. She was like, my mom wants me to do this, so that's why I'm majoring in it. But after working with her for a few years, we got her and also the family OK with her doing a joint major between creative writing, which she actually really

SPEAKER_02: 7:13

liked,

SPEAKER_01: 7:14

and chemistry. And I think she's way happier in college for it. And so really lent myself to doing things that related more to people and seeing the impact on people rather than just pure Interesting.

SPEAKER_03: 7:30

Interesting. So what, tell us a little bit more about that. So what, you know, in a nutshell, what is the positive aspects of technology and education? Where are we using it well? Where are we not using it well, in your opinion?

SPEAKER_01: 7:45

Yeah. Well, I think this is pretty similar to how I think about it in ABA as well, but there's just so much paperwork everywhere. Like in the school system, especially for IEPs and the conversations we have with parents who are advocating for their kids, there's just so much paperwork. It's so much like letter writing, email writing to the right people at the right time, following up. And that's sort of the case with ABA too, right? It's like writing these notes, following up on these notes. There's a denial of from insurance, so writing something or readjusting your note to be more insurance compliant. And so these are the places where I think that new technology and AI has the, hopefully will have the biggest impact, right? Like humans are really smart creatures. We're social creatures. We do a lot of really cool, creative, great things. And that comes with interacting with other people. It comes from sort of human to human connections, working with others, creating ideas, teaching, learning, Not paperwork, right? And so the less we can do of the boring stuff that nobody wants to do, the more we can do the cool stuff that really does make a difference.

SPEAKER_00: 8:56

That's such a good point. And you had a sentence that I thought was really interesting because it's so true. You said readjusting the notes so it's more in compliance with what insurance is looking for. It's funny how that happens all the time. And the session didn't change because the session has already happened. But it's like, how are we going to communicate to the insurance? It's just weird how that It comes down to the note. It just shows the difference between the documentation and the session because I guess the documentation is supposed to be a representation of the session, but so often then the session... either becomes a representation of the documentation or they become two totally different things. You're like, well, this is what happened in the session, but just write this in the note.

SPEAKER_03: 9:35

Well, and it's this fear. You mentioned about something getting kicked back or whatever the case may be when you submit it to the insurance. But it's this fear now to make them align, which then runs the risk of your documentation not being accurate, not actually capturing what you did. And then now maybe it's compliant with the insurance, but for the clinician, it doesn't really capture the history or the important stuff that you wanted. To put this out there, there are plenty of colleagues who might say, yeah, I've got my notes templated already, so I just cut and paste at the end. And it makes it easy. And it's like, oh, great. But does that really then meet the purpose of the documentation? So we're always just kind of trying to appease the insurance, I guess. To be fair and not to be disparaging, we are, and that can be a big problem. This is where you come in. Tell us a little bit about learning about being compliant with insurance, how that's affected your approach in creating this product toward quicker, easier, accurate medical notes.

SPEAKER_01: 10:39

Yeah, absolutely. So Alpaca Health's first product is our AI note taker for BCBAs. We listen to parent interviews, parent trainings, and also BT supervision sessions, and automatically take notes, write summaries that hopefully you can copy and paste as the session summary for documentation purposes, and also call out important highlights that you might need to know to work with the client in the future. When I think about documentation, I think the big reason why people do it is because they need to get paid by insurance but the other piece is also that taking notes helps you remember what happens so that when you need to go back and think about how was Joey doing two months ago you actually have a note summarizing it and I think we kind of forget about that second reason why we do documentation that is equally if not more important and so what I think about this tool is if you have a listener in two sessions then you can create the best summary possible because it's literally right there listening to the entire session that you just conducted. And it can create a summary that captures all the important details that you would need for clinical interactions going forward, but also write that and spin that in a way that is insurance compliant, which understanding what that means has been a whole journey in itself, right? Templates, requirements, every insurer has a different medical necessity brochure that seems to change randomly, and it's not clear which document is out of date on their website until you call them, or just because you know the payer guy. This is how we've come to discover what insurance requires, is just having... really great run-ins at conferences sometimes.

SPEAKER_02: 12:23

And

SPEAKER_01: 12:25

so I can't imagine being like a solo BCBA who's running a practice, managing BTs, intaking new clients, at the same time trying to understand what

SPEAKER_03: 12:35

insurance wants. We appreciate your empathy tremendously, man. Everything you just mentioned is preach it. Sing it, brother. Sing it. Amen. Hallelujah. It's quite a terrain. Go ahead, Mr.

SPEAKER_00: 12:48

Dan. No, I was going to say, Mike, you've always had an interesting idea and it's actually created some friction with colleagues in the field the idea of we've had an interesting relationship with data and I'll make data and documentation synonymous although they are a little bit different I think documentation the way that we think about it now is more of what are we submitting to the insurance the data is more of what are we referencing as BCBAs but we've had an interesting relationship with data and we understand its importance it absolutely is we cannot run effective ABA without data that because And you brought up an interesting point that, you know, Michael's software really does help with is this idea of we can be doing one thing at a time, taking data or interacting with the client. And whenever, like Michael said, with the parent interview piece, if we're writing down what the parents talking about, we're not paying attention to what the parents talking about. So we could have conversations or we could take that or we could do these things and then do it on the back end, which is really intriguing. That means our interactions can be that much more vibrant.

SPEAKER_03: 13:53

I like that a lot. The idea that I could... you know, break down a three-part contingency or do ABC data as I'm talking to a parent and then reference and back ops, you know, so often in that conversation, I'm coming up with three, four different antecedents or consequence strategies that, you know, in that dynamic conversation. So to sit there, I have to stop myself and write them all down or remember them. I'm not capturing it fully. What you're saying now is that would be captured fully and then I just call it up however I want to and then I print it up, you know, or edit the text however I need to and then hand it to the parent. And I like the way you pitched that, Dan. I had a very rocky relationship with data as a developmentalist. You're not quite divorced

SPEAKER_00: 14:38

from

SPEAKER_03: 14:38

data yet, but you're in counseling. I love data. I just don't like when it's clingy or always on me. I need some space once in a while. But as a developmentalist, more looking at developmental theory or the idea of narratives and how you can extract data from, that's always been much more my approach to that interaction, this idea that a very clear or very truncated trial, ABC data point, ABC data point, I've always found that a very disjointed way to interact with clients. So what you're talking about sounds fantastic. Even looking into the future, the notion that this could be capturing certain trial sets, and then you just call up your verbal consequence to count how many times there was a correct response versus an incorrect response. So I mean, this Hopefully I'm not stretching this too far, but this opens up a lot of possibilities where I see clinicians being able to free up their hands and their attention toward the interaction and the implementation of protocols. And now the data is still being captured. Just a matter of calling it back up after the fact.

SPEAKER_01: 15:46

Yeah. Research shows that humans don't multitask. We don't. We become good at it by training it over time, kind of flipping back from one thing to another. but we don't actually multitask. We don't actually have two things going on in our mind at the same time. We're just thinking back between those things really quickly. And so, you know, like more experienced BCBAs tell me like, oh, I got note taking down. This is actually super easy for me. And that's great for you because you've had to do it this way, right, for so long. And you've trained yourself to multitask really quickly to switch between the clinical interaction writing notes. But for everybody else, it's still hard. And even for a lot of experienced folks, it is hard as well. And so that's sort of of our thesis with AI and just better technology is how can we just do the note-taking, the documentation, the data collection on behalf of clinicians so they can truly have 100% focus on the clinical interaction. One of the places that we're taking the product, and today everything's audio-based. You record parent interviews, you record caregiver training sessions, you record BT supervision sessions, but we know that audio is only just one of the things that happens in a session. A lot of ABA is physical. A lot of autistic kids that we work with are nonverbal. And so what we are working on is the video-based approach to data collection. Stick a video camera in the corner of a session, record what's happening, and we'll take all that behavioral data for your BT or for you in a direct session so that you can get the most granular pieces of data. You can go back in the video record and see what happened. You can automatically pull out out ABC data that maybe your BT can't pull out because they're not experienced trained professionals like you are and all have that done while the BT focuses 100% of the clinical interaction because there's a bit of camera and AI watching the video.

SPEAKER_00: 17:44

That's always been the challenge with ABA is that we are lab focused. We're evidence based. So we tell this laboratory basis, for lack of a better term. But when things are actually going on in the lab, you have somebody interacting with the person and then an independent observer taking data. The person interacting isn't the person taking data. So when you have that person become the same person, you run into a lot of issues. So I like what Michael's saying. I didn't know that you were going to bring that up, Michael. We still will talk about the audio piece, but that allows us to kind of run both sides. It allows us to have the internal validity of the lab with the external validity of not having to take the data while we're interacting with the person.

SPEAKER_03: 18:31

It takes training to a whole new level. Being able to sit with your BT and re-watch a session. The idea of reliability between data sets. Now that doesn't have to happen live. It can happen more comfortably after the fact during a supervision session. Hey, let's just, let's see what AI punched out. Let's see what you and I punch out in terms of rewatching these three minutes of a session, you know, whatever that, yeah, that really changes the whole game. That's exciting. I like to, my little analogy here, people see the law says if you're texting who's driving, That's the way I feel about data and the clinical interaction, right? I mean, if you're really dividing your attention between even just a push button on a tablet, I understand that's very minimal, but it's like talking to somebody and then them going to do text. There's a shift in that conversation and that interaction, and it's probably not modeling the best that we can model, especially for our clients, in terms of something socially significant. So, yeah, what you're presenting here changes the whole game. That's exciting.

SPEAKER_01: 19:37

Absolutely. I mean, I know that there are some BCBAs who've started typing during parent interviews because they're like, I want to write everything down. I know I need this in my computer to write this initial treatment plan later. And I understand why they're doing it. But I also think if this is your first time meeting with the parent of an autistic kid who just got diagnosed, who's extremely freaked out about everything, who just wants, honestly, a chance to vent because you're the first person they're talking to who actually understands what's going on beyond just like, oh, sorry, autism diagnosis, hoo-hoo for you, can actually have a real conversation with you about your kid, how you're parenting the kid, then it's like, that's a very human conversation. And to have a computer screen typing in between is just really jarring as an experience. It's like we all know the doctor who spends the doctor's appointments with us looking at the computer screen and not us. And it's frustrating, but it's also not the worst thing in the world because I just came in with a strep throat and I just need antibiotics. Parent interviews, parent trainings are just a whole different ballgame. So to have a computer screen kind of blocking that interaction feels not really great. And that was one of the pieces that really stood out to me of like, wow, we need to find a better way to do this.

SPEAKER_00: 20:58

Wow, that's a good point. Yeah, that becomes very like... What's the word? Sterile, I guess would be a good way to put it. Like you have this person typing, and yeah, I think that sometimes they do just want that conversation.

SPEAKER_03: 21:10

And you're right. I mean, the typing is almost a necessary evil. You want to capture that information. It's important information, and then you're absolutely right. Again, now you're texting. You're no longer driving. So it's going to change the game in terms of how actively we will be listening with this tool, but then how we will emote and present that active listenership to the parent. which I think is amazing. It's going to go a long way. I mean, we think about it in terms of ease and convenience, but in terms of focus and being able to pay attention now to those details that a very concerned or frustrated parent or somebody that's going through their steps of grief, they just got some really tough news, just how much more empathic and focused we may be able to be by knowing that everything we want to hear and all the questions we're asking are being documented, recorded without us having to lift one finger. That's amazing.

SPEAKER_00: 22:07

100%. Let me take a step back and get into your current product. I'm assuming you're with a team or maybe it was just you. What was your impetus of like, all right, this is Where I see, did you conduct outreach to BCBAs or is it just kind of what you've seen? Why did you kind of get to where you are now with the product that you're offering, if that makes sense?

SPEAKER_01: 22:36

At this point, we've talked to over 100 BCBAs and even more VTs and ABA operations folks. I went to ABA International over Memorial Day in Philly. I was able to talk to a lot of BCBAs there, listen in on sessions. And I think just like the common theme among all of them was this complaint around documentation, initial treatment plans, and also the accuracy of data collections that VTs were doing. doing. And so we really started there. There's a lot of other things that we've heard about, like scheduling. phone all the time, things that we're definitely working on. But we've started here with documentation and paperwork because we've heard it's sort of the biggest challenge from clinicians. And I guess taking a step back and thinking about the field and where it is, it makes a lot of sense, right? Like ABA only became insurance covered a decade ago, 2014. So we're still, as an industry, payers as a funding source, we're still trying to understand how to properly track ABA efficacy, what is value, How do I make sure things are actually happening? And so, of course, things aren't that standardized. And of course, it's still a little bit of the wild west of documentation because the field is still tough. But we don't have to let clinicians and clients be the victim of that. We can help them deal with all this while still being able to have fidelity to the core human elements of clinical ABA.

SPEAKER_00: 24:06

So with your initial product, some of the cool features that you showed us when we met prior, maybe you could talk a little bit about that. So you mentioned that it records and can transcribe parent trainings, parent interviews, things like that. I feel like a lot of software can do that. Yours does some really cool stuff in addition to just the transcription. Can you talk about some of the cool features that make your product unique? Because I think those are pretty awesome.

UNKNOWN: 24:35

Yeah.

SPEAKER_01: 24:36

Yeah. So we are building everything for our ABA specifically. And so when we're summarizing these sessions, we're doing it in a way where the AI knows what is important out of an ABA sessions. It knows the language of ABCs. It knows the language of reinforcers. It knows the questions that you're intending to ask in a parent interview. It knows what's important and also knows what's not. So when the parent goes on a long tangent, they'll sort of ignore it. keep it in the transcript, but keep the summary focused on what is important to you as a clinician. for parent training sessions and BT supervision sessions. It creates a summary that does the same thing, focuses on the clinical pieces of the interaction. It also summarizes the action items that you have for the parent, the BT, and also for yourself so that after, say, a parent training session, you can easily just put that in an email or a text message to the parent and say, today, here's what we talked about, summary, here are your action items for the upcoming week. Oh, I love that. so that no parent can tell you, oh, you didn't tell me to do that because there is a written record of what you were supposed to do in your inbox at any time. And I think a similar thing would be really helpful for PTs. Then the last piece that's publicly available that anybody can sign up and try is we've gone deep into how treatment planning should work. It just feels like a lot of treatment planning is document summarization, document review, rewriting it in a Word document. When really what matters is writing your goals in a smart way and creating a behavior support plan that now your BTs can reference and use when working in sessions. And so I'd rather you use the four or eight hours you were authorized doing great definitions and great behavior support plans instead of trying to comb through an 80-page IEP and summarize it for one section of your treatment plan. And so what we've done is after you have recorded and summarized your parent interview, you can go back to our platform, upload your medical Sure. Essentially, you can write the 75% of the treatment plan that is boring document summarization so that you can focus the other 25% on the really cool stuff, the interesting stuff, the complicated stuff, the human stuff that I think every BCBA would rather be spending their time on.

SPEAKER_03: 27:27

Hell yeah, man. That's amazing in the sense that as you were talking, one of the things that I talk to younger clinicians about in terms of intake appointments is saying, hey, we get this whole lot of collateral. And then we've got our own in-house forms that start up our FBA, whatever it is, that are going to ask the very same questions that these parents have already answered a thousand times. And the fact that you're asking the same questions again to fill out this stupid form that we need for us in-house means that you didn't read the collateral. And now you're just some other clinician asking the same questions and means you don't know that child. What you're talking about opens up a whole lot of opportunity to go in fully well-versed, to go in from that collateral documentation having already been uploaded and used to fill out whatever in-home form you need. And then now your conversation is heartfelt, very human, very clinically sound, very concerned with the right amount of focus. So that's super exciting. I can't wait to learn more about that as you start using it. That reminds me of

SPEAKER_00: 28:38

when you You call the phone and they ask you, like, what's your account number? Enter your social and then whatever. And then as soon as the representative comes on the line, they're like, please confirm your account number, your social, all the stuff you just entered in like 10 seconds ago.

SPEAKER_03: 28:51

And by the way, Dan, make sure you listen to all the options because they may have changed recently. And wait times are longer than normal as well.

SPEAKER_00: 28:58

Right, wait times are longer than usual. But yeah, no, that's what you're talking about, Michael. It's just allowing us to become more efficient with our time. Now, let me ask you this. Well, in

SPEAKER_03: 29:10

time for documentation, just to be clear, it's more efficient and having more time, clinically speaking, to give to the client. So I think that's what you meant.

SPEAKER_00: 29:19

Yeah. So on the other end, kind of the skeptical end of it, how can we be confident that the software will pull out all of the information that we want to the extent that we won't need to go back and see what was missed and then read the whole 80 pages over again, if that makes sense.

SPEAKER_01: 29:38

Yeah, definitely. AI technology is still new, so it's not going to be perfect, and that's why... They'll never replace humans and never replace BCBAs. But what we've done that ChachiBT and other AI tools don't is we have trained our AI on what is important to a BCBA in an ABA three-tier model. And organizations can go in and customize our AI and tell it, here are the things I really care about in a treatment plan. Here are the things I really care about in a parent training. Here are the things I care about in a BT supervision session so it knows what things to look for and what things to pull out. And so generally speaking, the summary can be as customized as you want it to be if you're an organization partnered with Alpaca Health. From there, if it misses something, if you're like, hey, the client said something about cats and I forget what about cats they were talking about. You can go into the transcript and instead of like command F, type in cats, try to figure out what part of the transcript it's in, you can chat with our AI and it'll tell you the place in the transcript to watch and pull out an answer for you. And so I would say it's like instead of having having to refer to a transcript or your really messy handwritten notes to figure out what happened, you can just chat with an AI who will help direct you to the right part of the transcript and help give you a little bit of information to start out with.

SPEAKER_00: 31:05

Wow. That's awesome. AI is so abstract to me right now because I don't know exactly what it is. In terms of all of the uses, that just seems amazing.

SPEAKER_03: 31:18

Go ahead, Mike. I'm going to go back to something you said. There's a lot of people that are fearful of AI in terms of it replacing certain jobs, and that's certainly a possibility. You made an interesting statement, particularly to ABA, I think, particularly to medicine. In my opinion, it would be pretty hard to replace the human element. Just give us a little bit more about your insight on that. Why, in particular at ABA, why would it not be possible for a language model or AI to replace, say, a BCBA?

SPEAKER_01: 31:52

Yeah. Well, I think about this a lot, which is like what if everything goes right in ABA therapy, how is a family's life transformed? How is a client's life transformed? And almost always it has something to do with the way the client communicates with their family and the rest of the world. It has to do with the social skills of the client. It has to do with how the family can now do regular family outings with their child with autism. All of those things are fundamentally human and fundamentally social tasks, right? And so it's just sort of like, almost impossible to think of a world where you could use ABA to help a kid communicate better without having another human in the room to teach them how to communicate better. How is that possible? Imagine the most future world where you have some robot that's able to interact with a human, with a kid. That's not even what we want. We want to teach our kids to be able to communicate with other kids, other adults, other human beings, not how to communicate with AI. There's no way There's no conceivable way that you can replace humans out of that interaction. Now, if you're a doctor and all you do is prescribe antibiotics for strep throats, then maybe you're in a little bit more trouble. But I think the world of client-facing interactions in ADA will always be safe because they're the most human and the most important parts of what we do.

SPEAKER_00: 33:20

I'm

SPEAKER_03: 33:21

okay with that. Yeah,

SPEAKER_00: 33:22

that's actually interesting. I didn't even think about it until you just brought it up. Have there been applications, because a lot of the individuals we work with are nonverbal, which really creates a huge obstacle for them to communicate with others. Have there been applications of AI explored in terms of speech devices for individuals on the spectrum to allow them, I don't even know if it's possible, because again, AI is very abstract to me. I'm very ignorant to the AI field. But almost as another communication model to allow individuals that are nonverbal alternative ways to communicate. Sorry, I'm on a tangent.

SPEAKER_01: 33:55

No, that could be really cool. I mean, I could imagine sort of like your typical, like, you know, instead of saying words, they like have cards that they use to communicate. Maybe there's some digital version of that where instead of it just being like, oh, he gave me the milk card, so I'm going to give him milk, that maybe like it speaks something to the parent or something like that. Like I can imagine interesting ways that that could be applied. I don't know of anybody now, but...

SPEAKER_00: 34:25

They have Proloquo, which is like Pex was the pictures. That was kind of the initial one. And then Proloquo was like the iPad electronic version of that. Like you press car and it says car. You say press I want car and it says that. I bet there's some interesting applications of, like you were saying, not communicating with robots. I totally agree with that. However, I wonder if there's a way that AI could allow us to get on the same wavelength. You're

SPEAKER_03: 34:52

making me think here, and Michael, you'd be the person to actually make this happen, but yeah, this technology actually changes the face of voice output or augmentative alternative communication devices, right? So the idea that now it could be a much more dynamic screen that's listening and then gives the learner an array of options that are specific to that response as opposed to the whole screen that they have to sort through. Now, that would then take away some of the learning, the semantic part it but it would get to that voice output a lot more accurately and faster and still I mean you could still narrow it down to a you know three or more array to make sure that the learners having to choose between options but yeah I never thought of that what you're talking about Dan I'm not sure if anybody's doing that but this really changes the whole face of voice output devices in terms of now that device listening and narrowing down options for the for the learner to respond with I don't know if that makes any sense

SPEAKER_01: 35:47

Yeah, no, it definitely does. This is totally outside the world of autism and ABA, but there's really interesting and just mind-boggling early research and case studies of human brain interaction, where if you were a quadriplegic, hadn't been able to speak very easily before, now we can sort of hook up to the electrical signals in your brain and actually what you're thinking your brain gets communicated out in Wow. It's super early on, right? And it's AI plus biology plus medicine plus a whole bunch of other disciplines coming together to make that happen. But yeah, I think a lot of cool things are happening that are going to be really game-changing for a lot of people who need help.

SPEAKER_00: 36:38

Yeah, just maybe another thing to explore eventually. Yeah, sorry, we go off on tangents. This can be one of the ones like, you know, how the AI can skip over the cat disguise If y'all are using AI to listen to this, you can skip over our tangents. I thought that was a good one. I thought that was very relevant. No, I did too. We'll talk about that moving forward. Ways of giving people voices. The new Android, I think it's Android 14 or whatever, you can actively real-time translate with somebody as opposed to, I'm saying it, we can have a conversation and it translates it in real-time. Kind of like some way of I don't know whether it translates it into a visual, some way to get people on the same wavelength. Going back to what currently you have out in that intake assessment, parent training piece, one thing that's useful too is can it automatically take some of the things that people say and goal plan or develop goals from that kind of automatically? Is that correct?

UNKNOWN: 37:42

Yeah.

SPEAKER_01: 37:43

We've started to work on that. We've started to work on that where it'll actually suggest baseline goals and definitions of goals. We haven't released that yet because we're still testing it. That feels like a piece that we want to get really right into the world. But I think whatever AI does there, it'll never be enough. I think this is the piece where you want a BCBA to go through and use that clinical judgment to adjust goals and make sure it makes sense for the kid, the BT, the interaction, all those things. But we're certainly working on that piece as well. But I do think that's one of the places where you absolutely still want to be CBA in the room, in the loop, to make sure that it's accurate.

SPEAKER_00: 38:25

The cool thing about that, though, if it's generated from the AI transcript from the parent discussion, is that it would still be individualized. One thing that I think concerns us is that as things become more templated, they become less individualized. So there's a lot of times, oh, you've got a three-year-old. These are the 10 goals you're going to work on with a three-year-old. It doesn't matter what the parent said in the intake interview. These are your 10 three-year-old goals. And then when they master these, these are your next progressions because it fits within whatever the template of whatever electronic processor, or it could even just be a template. It sounds exciting that if it was derived from the AI discussion, it would then be individualized. It wouldn't be these are your 10 three-year-old goals. It would be these are your 10 goals that the discussion of you and Susie's mom talked about.

SPEAKER_01: 39:11

Yeah. And I would make the argument that AI is peak AI. individualism, individualized content and treatment planning, as opposed to what is happening today with templates. Like I put template here, I guess listeners, I gesticulate a lot. I would put templates to the left and I would put AI to the right. And I think templates are actually pretty anti-individualism. I see a lot of, especially larger organizations, having these cookie cutter templates where clinicians are rewriting three words in an entire paragraph. And I, whenever I see that, I don't want to say it in the moment. I'm like, how could this be as individualized as you think ABA should be? And AI be the scary thing that like prints out the same generic thing for every kid. Like that is what's happened on a large organization.

SPEAKER_02: 40:08

have

SPEAKER_01: 40:09

very robust templates for treatment plans. What AI lets you do is it lets you take like the core things you need to include in a treatment plan, a rubric of items, and then dynamically generates that the words, the sentences are going to be different for every single output for every single kid based off of the documents that you've uploaded, the conversations that you've had, and the information you've shared with the platform about the kid. Once that generates, you can go in there and make any changes you want. And so maybe this is a controversial take, but I think AI is actually much better than templates at providing individualized care experiences for clients.

SPEAKER_00: 40:48

That makes sense. I think we were kind of anti-template at our previous company. Templates do give you the opportunity to have access to a gold bank, but what people run into so often is you just control H client and control H the kid's name in there, and now you've got Your 15 goals just from the template with no individualization. I really like what you said, Michael. Go ahead, Mike.

SPEAKER_03: 41:08

No, no. Again, this is super exciting. Even thinking about activities, I think the same thing happens with learning activities. So traditionally in ABA, one of the challenges, depending on which model you're using, is you define your goal and now you've got this one particular set of stimuli that you're using to teach toward that goal. AI sounds like it's going to be able to generate a whole bunch of different ideas now, not based on what you're generically making it. your office across clients, but now more specified to the client's environment. The idea of a preference assessment becomes a lot easier here. Talk to us a little bit about that and kind of your thought process behind that or how you've seen people use this already successfully.

SPEAKER_01: 41:49

This is something that we're really actively thinking about is how can we make protocol generation and modification more tailored to clients? Because it seems like what's happening today is that you have this like massive program book. You sort of like, like, okay, we're going to work with the patient. I know that the BT Susie has really done this program before. I'm going to take this page out and like put it in the treatment plan and give it to Susie and put it in the behavior support plan and call it a day. What if we take that base protocol and use AI as an individualization layer to make that protocol custom for Susie, for the BT, for the family, the changes could be as simple as use examples because this is a child who really likes cars, so talk about different colored cars rather than different colored blocks, for instance. It could go deep as this kid is someone who struggles with object identification. So we're going to work on some other type of communication first that I can customize. But the kind of like basic point is we can take base kind of starting points that used to call templates and really individualize that using AI based off of all the other information the AI knows about that kid. And so another reason why I think AI is truly peak individualization for our client.

SPEAKER_00: 43:21

That's awesome. That is exciting. And I want to preface this too. I think all the listeners know we're not going to show for any random company. Michael got in touch with us and we spoke with him for about an hour prior to this meeting because we were very skeptical of AI. A lot of technology we've seen be very helpful. Some of the billing programs and stuff and the stuff that we have in there. We use central reach a lot, love central reach. But what we found is that a lot of times, What happens is these programs are designed to enhance the efficiency, which they do, but then all of a sudden the treatment revolves around, can I fit it into the way that this program wants to dissect the data, right? So I'm going to program in a way that I can graph it onto this Central Reach platform. Again, I'm shouting out to Central Reach, and no way am I disparaging them. I think they have a great product. But they went from, okay, let me help you, to now I feel like, For max efficiency, people are programming around the way to get it into their platform so they can distribute it to the insurance. What's really cool, and maybe you could speak to it too, Michael, is it seems like your goal is not necessarily just let me make a way for treatment to be built around us, but you're, again, focusing on the individualization piece of AI. So we're not going to need to organize our treatment around you. The AI will organize its responses around us. Is that correct?

SPEAKER_01: 44:50

Exactly. Yeah. And so in our platform, if you're an organization that partners with us, you can go in there and customize every single thing, right? Like, what do you want the AI to summarize? What do you want in the summary? Right? There's an action item section. What are the action items that might be important for AI to really look into and pull out? And so, like, we have a base starting point, right? Sure. Just to kind of bolt somewhere to use. But as you start using it, you're like, hey, like, the AI seems to be consistently missing the, you know, like... Preferences, you know, the caregiver talking about why data collection is hard during parent training. Let me make sure that the AI knows that that's something that I want pulled out explicitly, because I know that my family's tend to have a really hard time with data collection. You can go in there, type it out in natural language, and then the AI will get smart on what you want. And so that's, I think, something that's really interesting about using this new type of AI is that you can talk to it as if it was your human executive assistant, right? Like if you had a human following you around in all of your parent trainings, typing out notes for you, you would be able to kind of coach that assistant of yours like, hey, remember to keep in mind X, Y, Z things. Or if they say something like this, that's really important. Please write it out. Having a human executive assistant for every BCBA would be pretty expensive and probably would be slightly awkward during parent interactions. And now you can have all of that on an AI on your phone in every single session that you conduct.

SPEAKER_00: 46:24

That's all. So AI, instead of artificial intelligence, we could call it active individualization. I like that. Oh, look at you. Michael's going to take that one. Take it. Take it. Take it. Yeah, no, that's awesome. Because it can change such real time, right? Again. I'm just bringing up Central Reach or whatever. For them to change the way that their program, that's going to take a lot of coding and a lot of, like, this could change real time, which is so, so exciting. Go ahead, Mike.

SPEAKER_03: 46:52

Well, you've changed my mind. I think I told you this the first time we spoke, and I'm just not experienced enough yet with these models, but that was my fear, and I think that's the fear I get when people go a little too far with the, oh, this is going to replace humans. I agree with you for the most part that there's going to be a human element that's needed for everything, but you've certainly changed changed my mind in terms of my fear of how things were just going to be templated and duplicated. Obviously, part of that was me fearing how people are going to use this technology, which I think is still an ongoing concern. And I think we need to get better at using it. But more than anything, it was just more that fear that the model was just going to be spitting out these generic things. But what you're talking about can become client-specific. And then if you use the system wisely enough, you could probably collapse that information or that system learning across all your clients toward a more clinical practice or your individualized standard of practice. So that sounds, is that something that's possible?

SPEAKER_01: 47:53

Yeah, yeah. Customize and individualize the client based on client information, but also customize and individualize your practice, your templates, your way of doing documentation.

SPEAKER_02: 48:03

Wow.

SPEAKER_01: 48:04

But yeah, I guess for you guys, I'd be curious to know, like, what concerns do you still have? What concerns do you think other people have? I

SPEAKER_00: 48:14

think you answered one of them. My concern would be... We were running into some tech issues yesterday on Zoom and our whole setup. It seems like when AI gets involved, they were like, have you done these 10 steps? And you were like, yes, we've done these 10 steps. But we had to suffice all of those 10 steps which you'd already done to get to step 11. So my... And that way was actually inefficient because it was like a protocol of like, well, you start here. Okay, we've done that. We're already way past that. No, you got to go here next. You got to go here. But it sounds like in what you're saying that it could be like individualized so specifically that I wouldn't need to go through the previous 10 steps if I didn't want to go through the 10 steps. I could just say we're going to start with step 11. Does that make sense? Yeah.

SPEAKER_01: 49:08

Yeah, it does, right? Based off of all the information you've given us, what you've done in the platform so far, we can jump you to kind of like the thing that you think we think you should be doing next. And I think like that's just what AI native software is able to do, right? Like I do think like people like Central Reach, Rethink, they've been around for a long time. They have a platform that has a lot of like switches and toggles and different things that you, that they've built with a lot of difficulty over time, coding every single one of those toggles. But now what AI native software lets you do is if you want to make a change, you just type it in. And you do that, like not a programmer. Those toggles become

SPEAKER_03: 49:49

fixed in SNC. You now have to deal– that's the constraints that Dan was talking about. So great software, but now it's fixed, and its automation, if it has anything, is going to almost be restrictive sometimes. I don't want it to do this, but it did it anyway. Okay.

SPEAKER_01: 50:04

Exactly. It's restricted to the toggles that were in the developer-coded platform, right? Like what if you want a different toggle? You want the toggle to be slightly different. Well, now you can do that in AI-native software like us just to not– So I think that's the biggest difference between old software and new software, software 2.0 and AI-enabled software 3.0, if you want to use some tech people, San Francisco language. Oh, snap. Well, you're in San Francisco right now. It is. I am.

SPEAKER_00: 50:37

So that would be my first concern, but I think you've satisfied that. That was our initial AI concern. My second one would... again, this is going to sound kind of pessimistic for the field, that what I've seen is that when these companies, like, again, I'll use Central Reach, they have some such good strategies, like they can automatically graph it. We used to have to put everything in Excel and it would take forever to graph it. And they can just graph things and make things such more efficient with the premise that, hey, if we can take these efficiencies from you, you can spend more time from the client. What I've seen, unfortunately, in the field is when this becomes more efficient, Now they just spread the BCBAs thinner. So instead of having to graph all your stuff, and now you can spend that time actually reviewing your graphs because Central Reach does that. The BCBAs aren't even reviewing their graphs because Central Reach are auto-progress. Probably 8 out of 10 of them, unfortunately, from what I've heard. If you ask them where any client is at any given time, a lot of them won't even know because they're just letting Central Reach progress through the graphs and things like that so they can go see more clients. So yes, it's more efficient. but the client-facing services haven't necessarily improved, if that makes sense. The efficiency has. The amount of hours that I can bill has improved, but the client-facing service hasn't improved. So that would be my second concern, personally, just from what I've seen, is that how can we make sure that it's not, cool, we got this AI that's going to spit out all the programs, so go run a bunch more programs as opposed to monitor the programs. Does that make sense?

SPEAKER_01: 52:11

No, it makes total sense, right? If you think of your ABA clinic as a factory and now you've made one part of the factory slightly more efficient, now you're like, oh, how much more can I squeeze in and out of the factory?

SPEAKER_00: 52:26

Exactly, exactly. How many more clients can I give this BCBA now that we have versus your BCBA was already behind. They can maybe just catch up on the clients that they have. No, we got AI. They can see 15 clients now instead of 10.

SPEAKER_01: 52:39

Yeah, or can a BCBA finally not work after 7 or whatever time they actually get off work?

SPEAKER_00: 52:47

Can they not work Sundays?

SPEAKER_01: 52:48

Yeah, not get burnt out. So many BCBAs I talk to are like, I fall asleep to documentation. I was writing a treatment plan on Friday and then I actually spilled my tea over my computer. All this crazy stuff. Yeah. I think the challenge for me is I don't run an ABA agency. I don't run the ABA agencies of people who choose to use our software or not. I think you can use technology very poorly and you can be one of those places who treats their clinic like an input-output factory. We will all eventually kind of turn on those places. Insurance companies will start to realize that the care there is actually not that cost effective and start to give them lower rates and start to maybe stop pushing clients to them. I think providers, clinicians, BCBAs are starting to understand that they have a lot of power, right? There's a massive shortage of BCBAs out there. And so you as a BCBA actually do have a choice about who you work for and what they do. My hope is that technology like Alpaca Health helps more. more clinicians go independent, start their own practices, continue running their practices, grow their small practices, because now we take the back office scale of one of these private equity roll ups and give it to every single BCBA and every single clinician owned and operated business. So that's really my hope for what this technology can do is it sort of powers people who are really clinically minded, care about their clients, care about their clinician to run their own agencies the way they want to run them rather than whatever blueprint playbook private equity has printed out and given to all of its little companies.

SPEAKER_00: 54:40

That's so cool. I love it, man. We vetted, Mike. I want to say this. We have plenty of people reach out with similar stuff, and a lot of times we're just like, ah, nah. We really do believe you're genuine in what you're saying, and that's why we wanted to have you on, and we're very excited to promote your product because we do feel like it comes from that part of how can we make services better.

SPEAKER_03: 55:01

Yeah, this is really awesome. As I said in the beginning, the hour flies by. We're at our 55-minute mark. We've got a ton more to talk about. I'm sure this doesn't have to be the first and last time you appear. In fact, I can't wait to start using this product, and we're at the perfect time to start using this and learning more and then have you on again to give you feedback and actually learn more from you about how to better improve our use of this product. Because, again, I think as long as we stay smart about it and don't try to, you know, squeeze in more time or more clients. As long as we keep the status quo in terms of ethical practice and manageable caseloads, but then use this to improve the quality, not increase the quantity, it stands to have a lot of power. I really appreciate that you've brought this to our field. We're excited. I

SPEAKER_00: 55:54

guess in conclusion, Michael, kind of wrapping up for you, let's say a BCBA is listening, a company is listening, anybody like that, what would you say is kind of your... Perfect.

SPEAKER_01: 56:28

Yeah. I think in short, turn the time you spend on paperwork to time spent with people. All that documentation time, let that go away because an alpaca is a system. In terms of how to evaluate whether it actually works, I think with a lot of these AI things, proof is in the pudding, right? Give it a shot. You like it, awesome. If not, tell us why, let us fix it, please. But also no sweat there. And so we actually have alpaca assistant totally free for anybody to sign up and give it a shot. You don't even need to use it with a real client. We have a sample client client records. We have sample transcripts that you could just read aloud so that you can test what Alpaca Assistant will do. on your own time without a client and without having to worry about any of those things. And so that way, when you use it yourself and you like it and you want to start using it in clinical interactions, you also know how to leverage the product the best. It's totally HIPAA compliant. Every user who signs up with us signs a business associates agreement with Alpaca Health, the company. And so folks are covered when it comes to a HIPAA and privacy confidentiality compliance point of view. So my ask for everybody who's made it us this far into the hour is to give it a try. It's free. You can use it without clients and it's HIPAA compliant. And you can go to alpacahealth.io and just start using it. And if you like it or if you have feedback, please reach out. I'm very active on LinkedIn. My email's on the website. So we'd love to continue the conversation.

SPEAKER_00: 58:07

We'll have the website. If you don't mind, we'll maybe put your email or something for people to contact you in the description below. We're very excited. We've always been... I feel like on the forefront of trying to innovate in the ABA field, that's why we're kind of where we're at right now. And yeah, I'm really excited to see what Michael and Alpaca Health Assistant can provide.

SPEAKER_03: 58:31

You've convinced me. I will actually now use AI, and it's because of you, Mr. Gao. Thank you so much for reaching out. Thank you so much for creating this product. I'd like to end it with a little synopsis here, and I'm going to have to quote you. You said, take the time you're spending on paperwork and spend it with people. I love that. And then like we like to say here on ABA on tap, always analyze responsibly. Cheers. Cheers. Thanks a lot, Michael.

SPEAKER_01: 59:01

Thank you guys.

SPEAKER_03: 59:03

ABA on tap is recorded live and unfiltered. We're done for the day. You don't have to go home, but you can't stay here. See you next time.