Integrating AI into Your Business with Mai Ling & James

James Berges 00:01 
If you haven't noticed, AI is taking over the world. And one of the most promising areas is in the world of accessibility. And so today, Mei Ling and I are going to dive into the new and exciting world of artificial intelligence or AI to explore its potential and nuances in speech therapy and special education. 

James Berges 00:16 
We're going to dive into what AI is, the critical ethical considerations in its application, which there are many, and explore how it can revolutionize your therapy as well as your workload with specific applications and speech therapy, because that's Mei Ling and I's bread and butter as speech pathologists. 

James Berges 00:35 
And by the way, we're posting this epic AI resource presentation on our website. So don't miss out on getting that full download. You can get it for free at exceptionalleaders .com. And you just go there and enter your email, and you can have this resource for all things AI. 

James Berges 00:52 
So you can future proof your practice and just be ahead of the curve. And so the amazing thing about chat GPT and generative AI is you can type in like, hey, I have a student who's this profile without giving away too much information about who they are. 

James Berges 01:13 
Let's say they're working on the R sound in the medial position. You know, give me, give me non -words, give me a list of non -words we can practice that are multi -syllabic. 

Mai Ling Chan 01:23 
Welcome to the Exceptional Leaders Podcast. I'm Mai Ling Chan, and together with James Berges, we're getting you top tips and resources for building and scaling your disability -focused offerings straight from the forefront of disability advocacy and leadership. 

Mai Ling Chan 01:56 
Yeah, I love that. I just did a big presentation for speechtherapypd .com plug. You can't get the course now unless you subscribe. So that means you can, but it's behind this paywall now. But that is a deep dive into the specifics of AI and ethics and speech therapy. 

Mai Ling Chan 02:12 
So I welcome our listener to take a look at that if that interests you. But to answer James at this point is that we've been moving towards AI for years. And I would say back since like the 1940s and 50s, we had these theoretical concepts about can machines actually think. 

Mai Ling Chan 02:28 
And then we had it coined by John McCarthy of the title of artificial intelligence in 1956. And then came this rise of machine learning and the ability for it to problem solve and to process big data. 

Mai Ling Chan 02:40 
And that's kind of where we are now, which is what we're calling generative AI, where computers can create actual new and original content. 

James Berges 02:50 
Yeah, I love it. I have gone crazy with chat GPT. I think we're kind of at that point now where AI said booms and busts where we, you know, we get excited about this new technology and it's almost like a gold rush. 

James Berges 03:01 
And with generative AI and chat GPT, I mean, it's been around for over a year now. And I remember, like the hype around it was so big, but now we're entering that phase of like, okay, what can we actually do with this thing that's practical, we're starting to see its limitations. 

James Berges 03:16 
And so that's on the consumer side, but then there's also the entrepreneurial side of helping developers code faster, and language models, which we'll get into a speech therapy that help not just create new things, but to actually recognize patterns, right? 

James Berges 03:31 
It's all pattern recognition. So we'll dive into that. 

Mai Ling Chan 03:34 
Yeah. And for our listener, I just want to remind myself and James that James and I have been using this and he's saying, you know, it's been out for a year already. I got started right after James said he was like, hey, have you heard about this new thing? 

Mai Ling Chan 03:45 
Chat GPT. So we've been playing around with it for years and obviously I'm now presenting on it. But for a lot of the population, people are still just going, hmm, I don't know what that is and I don't know if I need it. 

Mai Ling Chan 03:56 
And so that's why we thought we would do this episode to kind of give you some broad strokes on ways that we're using it and ways that it might be able to help you in your clinical endeavors. If you're a therapist or a service provider, but then also in your entrepreneurial endeavors. 

Mai Ling Chan 04:11 
And I think that's the area that you guys should really take some notes because this will really reduce your workload. And I could see James is nodding his head. So for our audio only listeners, I mean, we are not even sharing how much we were using Chat GPT, not that it's the end all for everything. 

Mai Ling Chan 04:26 
I mean, honestly, James, sometimes it takes longer for me to get through like putting together the prompt, you know, taking a look for accuracy, deciding on what I want than if I had really started from scratch. 

Mai Ling Chan 04:36 
So sometimes I do have to really consider like, was this really worth it to go, you know, on this whole journey? But other times it's amazing. 

James Berges 04:43 
Yeah, absolutely. So we'll go into pros and cons. And first let's start with some ethical considerations because I think we need to know what to look out for so we don't abuse it. After all, it's a tool and any tool can be used for a good or evil. 

James Berges 04:57 
So what should we consider ethically, mainly? 

Mai Ling Chan 05:00 
Yeah, it's stuff that we've already known. I mean, those of us that are out there in healthcare, obviously, like the HIPAA, Health Insurance Portability and Accountability Act information, you know, so data, privacy, security, you know, how are you, how are you uploading the data into these systems and how public is that information? 

Mai Ling Chan 05:17 
It's just, it's always going to be there and it's important for us to keep that in mind, you know, that this isn't your information, this is your customer's information, your client's information, your patient's information, and you are the shepherd of that data. 

Mai Ling Chan 05:29 
And so that is absolutely going to apply to AI. 

James Berges 05:33 
Yeah, absolutely. And so HIPAA compliance, privacy, those are always ongoing issues with any software. But I think of even more broadly, I'm sort of thinking of AI as corrective lenses as this analogy. 

James Berges 05:47 
When we're thinking about accessibility, for example, you're wearing glasses, if you're watching on YouTube, I'm wearing contacts. So we're both using this layer on top of our normal, they're not even normal vision, but just on our organic vision to help us augment. 

James Berges 06:04 
But AI can be like that. It's like an interface that can help take whatever ability level you're at to see the world more clearly. But ethically, that can be a problem when you're talking about these large language models. 

James Berges 06:16 
Basically, we're feeding so much information into them. But where does it get this information to form a model of the world? And so you have this overrepresentation of neurotypical people filling it people without disabilities. 

James Berges 06:32 
And so the issue is how can we get the most representative data so that things like voice to text, text to voice, alt text are all representative of all types of neurotypes and people with different ability levels. 

Mai Ling Chan 06:45 
And actually, what I think you're really talking about is the bias section. And so one of the things that I want to spotlight is that AI only knows what you've taught it, right? So it's all of this data that we are feeding it. 

Mai Ling Chan 06:57 
And if you use one of these popular software like ChatGPT 3 .5 or ChatGPT 4, they are only taught up until a certain point in time. So the first one is, it's the free version, ChatGPT 3 .5 only knows up till September, 2021. 

Mai Ling Chan 07:12 
And then ChatGPT 4 goes up to April, 2023. But with ChatGPT 4, you can upload PDFs, Word docs, spreadsheets, information, and you can give it URL links where it can go out to the internet and get more updated information. 

Mai Ling Chan 07:29 
So the bias piece of it is like, what has it learned up until this time? That's all it knows versus what you're going to continue to train it on. And then like you're saying, if we're looking at neurodiverse people, are we teaching it to be more flexible where we're not looking for, let's say, eye contact all the time or a certain type of like a motive response because you've trained it that way. 

Mai Ling Chan 07:54 
And that's where a lot of like employment, HR companies now are trying to really get past those things that we're putting in quotes as typical. But if we're putting it into an AI program, what if we've already trained it to look for those? 

Mai Ling Chan 08:06 
Does that make sense? 

James Berges 08:08 
Yeah, absolutely. I almost feel like there's going to be different levels of specific LLMs, large language models, basically the thing you feed for the machine learning for every type of issue. We'll highlight specific companies in a bit, but thinking about voice recognition as an example for speech therapy. 

James Berges 08:28 
I mean, you have chat GPT voice now, which is pretty amazing. It's conversational. You have Siri, but it's pretty garbage when it comes to understanding non -standard speech. So how can we train these large language models to recognize stuttering or a lisp or even an accent and dyspraxia, dysarthria? 

James Berges 08:50 
And so even beyond that, how can you get it to understand someone's specific way of talking or thinking? So those are all issues. Another issue is even just with alt text. So as we're going through, you're on a website, let's say you're visually impaired and you're looking at a chart, not looking, you're having the screen reader read out the image to you. 

James Berges 09:12 
These biases can lead to misinterpretations. So let's say an image is of someone in a wheelchair, the alt text, if it's using AI to generate the alt text automatically might say, this is an image of someone sitting and it's missing the context that it's actually someone in a, in a wheelchair race, like in a competition. 

James Berges 09:35 
And so that happens when you're missing the diversity of data and input in, into these large language models. 

Mai Ling Chan 09:42 
Exactly. And in my presentations, what I've been saying is we need pioneers to be using the system and pointing out the errors, pointing out the strengths and helping us to shape what is being created out there. 

Mai Ling Chan 09:53 
So for our listeners and people watching, I really, really invite you to just start using stuff. You know, there are so many great AI tools that are out there that will say powered by AI, you know, Bing has a search feature now, Microsoft also has one, or you can go with the chat GPT versions, but I would say don't be afraid, you know, join us in pioneering this new technology and poke holes, you know, find the places that, oh, this could be better or this is absolutely wrong and reach out to the people who are creating them. 

Mai Ling Chan 10:22 
You know, there's always a contact us. We need your feedback and we need to find out how can we continue to shape this to be better, more efficient, more protective and secure and more non -biased, of course, but they need to hear from us. 

Mai Ling Chan 10:34 
Yes. 

James Berges 10:35 
That's absolutely a good call to action mailing. It can be hard to find these data sets. For example, someone using AAC could be tiring to type out 100 sentences in a text to speech application, but I think with the internet and technology where we can collect samples at large, it's cool that we can crowdsource these things. 

James Berges 10:59 
Google's doing something like that with Project Euphonia to collect different speech samples from around the world. So just thinking of how can, instead of spending maybe seven hours in a clinical setting where someone's collecting language samples, can we crowdsource these data sets and feed them into a better language model? 

James Berges 11:16 
I love that. 

Mai Ling Chan 11:16 
Yeah. So many great ideas out there. And actually, if you're in those speech therapy chat groups, they're really cool for coming up with ideas. What are ideas for AI prompts? And I'm saying speech therapy because that's our area of strength. 

Mai Ling Chan 11:28 
But if you're in a certain profession, deep dive into those chat groups because it's amazing what people are freely sharing. And I'm sure that you will learn from them how to better set up your prompts and how to better utilize these tools in your specific systems. 

James Berges 11:44 
Yeah, and again, we're going to put a free resource at ExceptionalLeaders .com. You enter your email there. We'll have some examples of how you can actually practically use these things with a better holistic understanding. 

James Berges 11:56 
So, why don't we dive into some applications? We've talked about the ethics and some biases and issues here. Let's look at some examples of people who are transforming speech therapy and accessibility. 

James Berges 12:07 
I'll start with a really simple one that I wish I had when I was doing more clinical work in the schools, and that is voice -to -voice interface. So, depending on your impairment or your abilities, you know, you might be better at reading comprehension and have a harder time with voice recognition. 

James Berges 12:29 
You might have an easier time the other way around. But one example that comes to my mind is I had a student who was in high school. I worked in the schools, and he could carry a conversation and learn really well when you're talking things out with him. 

James Berges 12:44 
It was just really hard for him to read and comprehend that way. So, what we would do is use these kind of janky text -to -speech, you know, and read it out and that's standard when we think of like the Stephen Hawking voice almost. 

James Berges 12:59 
But now, if you go to the App Store, if you have an iPhone, I think Android too, you can get the chat GPT app for free. And I think this comes with a free version. Don't quote me on that. But the pro version is only $20 a month. 

James Berges 13:13 
It's worth it. And you can do voice -to -voice, and it sounds so human. It's like, okay, I can say explain quantum physics to me like I'm a fifth grader. And, you know, it takes this complex topic. It breathes, which is a little bit creepy. 

James Berges 13:28 
It adds in these vocal inflections, but it just makes it way easier to digest information than if that student had to go back and read or type it into an interface and then have it press a button to do voice -to -text or text -to -voice. 

James Berges 13:43 
So, I love that. 

Mai Ling Chan 13:44 
I love that, and then for those of us that are in the speech and hearing realm, Voice It is another great technology that I love where you can bank your voice or you can utilize a different voice. If you have a degenerative disorder or you're using AAC or you're a child and now you don't want to use all these child voices, it's just amazing what they can do now with the technology. 

James Berges 14:08 
Yeah. And we have a whole interview with the founders of voice it. So if you just search voice it and exceptional leaders, uh, exceptional leaders .com, you'll see that one. Really cool to, to learn about that firsthand. 

Mai Ling Chan 14:22 
If you're like me, you can't get enough of books, podcasts, blogs, and other ways to find out how to create, grow, and scale. That's why I brought together 43 disability -focused leaders to give you more of what you're looking for. 

Mai Ling Chan 14:34 
You will hear their stories in three best -selling books, which focus on general offerings, augmentative and alternative communication, and speech -language pathology. I invite you to search for Becoming an Exceptional Leader on Amazon, so you can learn intimate start -up pearls of wisdom and keep growing your brilliant ideas. 

James Berges 14:55 
Another part of this is just multilingualism. Something that seems so, I mean, I kind of took for granted till I was traveling the world in South America and realized my Spanish is un poco malo, not that great. 

James Berges 15:10 
But, but when you think about, you know, we have Google translate, it's okay. All these things are not that great, but especially for people who use augmentative alternative communication devices, like their options are pretty limited. 

James Berges 15:23 
They have these symbols, they're sort of janky symbols, maybe 2D. All right. Not to anyways, a lot of them are accessible and they've been researched, so I shouldn't say that. But as far as multilingualism, it's still lacking. 

James Berges 15:37 
So one of our other past guests, Alan Vu has come up with an amazing software called flex speak, and you can get that on the app store now too. I'm not, he's not asking me to promote it, but I'm just saying it here because it's really cool that this simple problem of like if a student in Vietnam or a student who has to go to Malaysia or wherever else in the world and before they didn't have access to multilingual AAC, now they can communicate with their whole family back home using AI to really take that like input and expand it out faster than it could before in the right language. 

Mai Ling Chan 16:11 
Yeah, and I just want to share, you can Google this translate video and you can record yourself saying something one time and then you can have it immediately transcribed by AI into any language. I mean, you just name the language that's available, but your mouth is actually speaking the articulation of the words. 

Mai Ling Chan 16:29 
It's incredible. I don't know if you've seen that, but a lot of companies are now using it for like, let's say, healthcare instructions. You know, say you have a family who speaks Farsi and you don't. 

Mai Ling Chan 16:37 
It's just incredible. 

James Berges 16:40 
Yeah, I love that I was thinking of these deep fakes that are kind of creepy where you can make it seem like, you know, a politician is saying something they never said by matching their mouth to it. 

James Berges 16:49 
So again, there's always going to be ethical dark sides to these things. But yeah, as far as just clearly communicating in other languages, it's going to be amazing. I'm super excited for it. And then another use that's just kind of practical for clinicians is creating materials, right? 

James Berges 17:06 
So as a speech therapist or occupational therapist or anyone who's working in in allied health, part of it is lesson planning. And I'm sure that takes up a lot of your time, as I remember it does. And so, you know, there's teacher pay teachers, there's these different websites you can go to to download premade materials. 

James Berges 17:26 
But often, they're not exactly personalized exactly to your student. And so the amazing thing about chat GPT and generative AI is you can type in, like, hey, I have a student who's this profile, without giving away too much information about who they are, you just say, they're working on these targets. 

James Berges 17:44 
Let's say they're working on the R sound in the medial position, you know, give me give me non words, give me a list of non words we can practice that are multi syllabic, or give me even images. So not just words, but you can create a whole world you can create, you know, personalized if they're into Shrek, you're like, give me our words and make a story with Shrek. 

James Berges 18:07 
So nice. Yeah. And that's all like free as long as you're you have chat GPT. So you don't have to scour the web for other people's materials, necessarily, you can really make it your own. 

Mai Ling Chan 18:20 
Yeah. And then for our listeners who are also entrepreneurs like us, there are so many workarounds and activities that you can ask chat GPT to do. Like you were just saying, um, you know, create something for you. 

Mai Ling Chan 18:31 
So we actually use it to help create our, uh, episode transcriptions or our descriptions, major points. It can tell you how to create an actual social media post, um, including emojis it's just incredible, but you definitely have to deep dive a little bit on how to set up the prompts. 

Mai Ling Chan 18:48 
So James is an expert in that and taught me a lot about, uh, setting up like you are a marketing guru, you know, specifically in healthcare. And then from there, you know, giving it what you want. So I definitely would look into that if you were the one man band like James and I usually are. 

James Berges 19:05 
Yeah, it's a game changer for like anyone who's a solo clinician provider slash entrepreneurial person where you're wearing many hats. The caveat I would give is always go through whatever it produces with a fine tooth comb because sometimes it confabulates a little bit or it uses wordy language that maybe would sound good in an essay in college, but it doesn't really translate to real practical information, but play around with the prompts. 

James Berges 19:30 
I think that's some of the most fun I have with it is like tweaking the prompts and seeing what works and what doesn't and then iterate on it, you know, it doesn't all have to be in one shot. What they call one shot prompting was where you put all the information in one prompt. 

James Berges 19:43 
If you're seeing that it's kind of missing some context, you can go back into the chat and say, hey, take out the fluff. I've said that a few times. No fluffy words or like here are a list of words I don't want you to use or even right in the style of insert favorite author or insert favorite copywriter. 

James Berges 20:02 
And so that's really fun. 

Mai Ling Chan 20:04 
That's incredible. 

James Berges 20:05 
But yeah, you got to make sure it's not making stuff up. 

Mai Ling Chan 20:08 
So that brings us to this very important point that I've also been sharing when I present about AI is it makes mistakes and even like 40% of the time, not always, but it can. And it is really not accurate for research for like, if you're doing large research, you have to be very, very niche. 

Mai Ling Chan 20:26 
And that's because it can't get behind those paywalls for the research that's available there. So just be really careful. And that's what we've been saying, especially for like our students, we're using it for, you know, education or using it in graduate programs, definitely check and recheck and recheck. 

Mai Ling Chan 20:42 
And again, that might take you a little bit longer, but I think the overall workload time will be brought down, but it really is your responsibility to infuse your personal expertise and experience and problem solving skills, you know, before you settle on that final output of content. 

James Berges 21:00 
So I've been talking a lot about chat GPT, but there's also Gemini from Google. There's a cloud, which is by Quora, but all of these are improving at rapid paces, so, you know, I just got to like play around with it now. 

James Berges 21:12 
Don't be overwhelmed because it's going to keep getting better and better. Um, and beyond, beyond generative AI. So this is all on the consumer side where you're typing things into chat GPT and checking your work, right? 

James Berges 21:25 
There's also report writing and administration as may saying, you know, as the more complex you get with expertise, these things are not going to replace research, although it can access research articles. 

James Berges 21:39 
If you go through certain chat GPs that are custom made for research. And that's, if you play for chat GPT plus different people are making specific use case versions of chat GPT. So that's exciting, but it's still really not there yet. 

James Berges 21:54 
And maybe that's a good thing for our clinical expertise. We won't lose our jobs yet, but hopefully it can make certain things faster. Like I don't love report writing. I don't know if you ever loved, you know, it gives a clear picture. 

James Berges 22:08 
A lot of this stuff is redundant information and kind of tedious. So language sample analysis, for example, I have tried putting a whole transcript of a student that I interviewed and saying segment this and code it for mean length of utterance, MLU, chat GPT is not quite there with that. 

James Berges 22:29 
It will give you an answer, but it won't be completely accurate. And so what about the generating reports? Well, you can probably make a template, say, here's sort of like what I write for my reports, fill in the information I give you. 

James Berges 22:43 
And I don't know, May, do you want to speak to we have a guest coming up about that for better speech and hearing? Yeah. 

Mai Ling Chan 22:48 
Totally, I was going to say, or you can just give it up to Michelle Baubert, who is the founder of Easy Report Pro. She's a speech language pathologist, and she is doing everything that we need to make sure that we can create speech reports ethically, and that's what her system does. 

Mai Ling Chan 23:06 
Instead of just giving you all of this rote information, it is customizing it, which is really interesting. The software highlights the section that they put in there that is AI created with red text, so it really stands out, and it's like, hey James, make sure you take a look at this before you move forward with it, so that you can go through and do that double check, critical thinking problem solving that we talked about, and your clinical expertise, and say, oh yeah, this is what I needed, or I just want to do a little more tweak to this before it goes into the final report, but I highly recommend it. 

Mai Ling Chan 23:40 
Michelle's going to be coming on the show in May for better speech and hearing month, so please pop in for that if you're interested at all in how we can use AI to help with report writing. 

James Berges 23:51 
Yeah, I actually a recruiter from some legal company just reached out to me to help with marketing. I'm not going to do it, but it's amazing the potential for the specific AI use cases for report writing and their whole business, which is making millions of dollars, is just filing legal reports about accidents. 

James Berges 24:11 
So like these ambulance chasing type situations where they're like, I don't know, lawyers who search documents who focus on that are going to have a run for their money with these types of things because yeah, the report writing is only going to get better at synthesizing information. 

James Berges 24:29 
I did want to highlight one other AI episode that we had recently with the founders of Mentra, Jilika Kumar and Connor Reinhardt, which are their rock stars who are changing the game in recruiting and matching neurodiverse talent in tech jobs. 

James Berges 24:48 
And so it's a little bit above my pay grade at this point about exactly how that AI algorithm works, but the basis of it is that they're using AI to match people who are autistic or dyslexic, whatever, neurodiverse in different ways to jobs and companies that match their profile of strengths and all that. 

James Berges 25:09 
So it's amazing for job seekers and companies. 

Mai Ling Chan 25:12 
I love that. And I'm thinking is that it's just taking away all that bias that we were talking about, you know, it's like, so what if James doesn't look at me while I'm doing the interview, it doesn't matter because he's amazing at his job. 

Mai Ling Chan 25:22 
And that's where we need to get is really look past of all all of these traditional typical, you know, things that we're looking for and instead really look at what is this? What are these people's strengths and skill set? 

Mai Ling Chan 25:33 
And, you know, what can they bring to our organization? So that that interview is fantastic. I really hope that you are listeners and our audience has a chance to go back and listen to that. Yeah. So I think in summary, what we're saying is that there is so much opportunity and potential for the software that is out there. 

Mai Ling Chan 25:49 
We've covered chat GPT, we've covered a lot of these other new AI tools that are there. But I think we're still in the new days, the early days, and you still got to use your noodle, you know, make sure that you are double checking, triple checking before your name goes on something on a piece of content on a report, you know, I mean, it can be so innocuous as an email because it does that too. 

Mai Ling Chan 26:08 
But I would definitely double check it before you go. I just want to remind you that we are giving away an epic, I'm calling an epic AI resource for speech therapy. It's 18 pages. It actually also deep dives into the ethics related to our ASHA code of ethics. 

Mai Ling Chan 26:22 
So if you're interested in how, you know, that kind of overlaps where, you know, we're talking about protecting our clients, you know, all the things that, that we have already agreed to, and we're committed to, you know, as practitioners. 

Mai Ling Chan 26:34 
But I think this also, the resources that are in there would also be thought provoking to people who are not a speech therapist, but maybe in health science, just to get you thinking about different ways to use AI. 

Mai Ling Chan 26:46 
So hopefully you guys grab that download. Thanks for watching. 

James Berges 26:48 
Yeah, go to ExceptionalLeaders .com, get the free download, and obviously we're just scratching the surface here. These things are developing at lightning speed. So let us know if you like these types of episodes or just Mei Ling and I unpacking a topic that's relevant and topical. 

James Berges 27:03 
Let us know, we're on social media at Exceptional Leaders and we'd love to hear from you, but stay tuned, the future is gonna be weird. 

Mai Ling Chan 27:15 
Awesome. We hope you enjoyed this episode and invite you to leave us a review on Apple Podcasts and Spotify and share the show with people you think will find value from it. This helps the show a lot. 

Mai Ling Chan 27:27 
Or have a great guest referral, reach out to us at xleaders at gmail .com. 

James Berges 27:33 
And if you want exclusive tips on becoming an exceptional leader, deliver straight to your inbox, just go to ExceptionalLeaders .com and sign up for our mailing list. Thanks for listening.