TRANSCRIPT 🔗
WEBVTT
00:01:01.340 --> 00:01:04.379
The following is a site to practice production.
00:01:14.379 --> 00:01:20.060
Well, welcome back to another episode of Beyond the Bot AI and Mental Health.
00:01:20.219 --> 00:01:27.420
I'm Dan Florel over at Eastern Kentucky University Professor there, and I'm going to throw you over to my co-host, Adam.
00:01:28.540 --> 00:01:31.740
Hey Dan, Adam Lockwood, Kent State University.
00:01:32.540 --> 00:01:33.180
All right.
00:01:33.340 --> 00:01:41.180
So we spent the last few episodes talking a little bit about school psych and especially report writers and AI.
00:01:41.340 --> 00:02:17.740
And of course, it's a huge area to kind of discuss, but we're going to roll it back a little bit more and have kind of a broader kind of sense of just where is AI in the mental health workforce in regards to how are school psychologists using it, how are maybe clinical psychologists using it, and then getting and digging in a little bit into what some of the more current research is showing us and some of the issues that it maybe rears that we've discussed previously, or even new issues that come up as tends to do with almost any new technology product.
00:02:17.900 --> 00:02:20.860
So, Adam, what would you say as far as general use?
00:02:21.099 --> 00:02:26.379
Where are our practitioners using AI the most?
00:02:26.539 --> 00:02:31.580
And what is the cutting edge for using it within practice?
00:02:32.860 --> 00:02:36.939
Well, the answer is we actually don't know how much they're using it currently.
00:02:37.099 --> 00:02:41.020
What I can discuss is data from a year or more ago.
00:02:42.460 --> 00:02:52.860
Because I and some colleagues have done some surveys, and the last APA practitioner poll survey was published what December of this year.
00:02:53.099 --> 00:02:55.580
So quite a ways ago.
00:02:55.659 --> 00:02:58.939
But what we do know is that adoption's pretty widespread.
00:02:59.180 --> 00:03:08.460
So when Ryan Farmer and I did a survey at the end of the 2024 school year, adoption was like 50% essentially.
00:03:08.780 --> 00:03:13.180
And and power users, people who were using daily or weekly, was low.
00:03:13.340 --> 00:03:16.699
I think it was like a quarter of the people.
00:03:17.500 --> 00:03:25.259
Now, fast forward one year later, and we had adoption that was about 80% of school psychologists.
00:03:25.419 --> 00:03:27.419
That was at the end of 2025.
00:03:27.659 --> 00:03:37.180
Now, we don't have data for the end of 2026, but we did get approval from NASC to access members for a survey that's going out next week.
00:03:37.419 --> 00:03:43.340
We'll have updated numbers, and my guess is we're gonna see that number even higher.
00:03:43.579 --> 00:03:53.019
At some point, we're gonna hit a ceiling where there's there's a holdout, whether it's 10%, 15%, or as long as they can, I'm gonna just say nope, I'm not using AI.
00:03:53.819 --> 00:03:55.579
I don't know if we hit that at 80%.
00:03:55.900 --> 00:03:57.019
I'm guessing not.
00:03:57.659 --> 00:03:58.859
But we're gonna see.
00:03:59.340 --> 00:04:07.019
And I I've also done some surveys of of health service psychologists, so PhD level psychologists essentially.
00:04:07.259 --> 00:04:16.139
And when APA first did its survey 20 practitioner polls 2025, I think it was, or 24, I forget.
00:04:16.860 --> 00:04:23.259
Uh those adoption rates were um it wasn't December of this year, it was December of last year.
00:04:25.819 --> 00:04:27.180
It was 2024.
00:04:28.060 --> 00:04:30.460
And the use was very low.
00:04:30.860 --> 00:04:34.699
It was it was like less than 30 percent.
00:04:35.100 --> 00:04:49.180
It's jumped up, and in our survey and APA, they're finding that over 50 percent are using AI of PhD level psychologists are using AI, and we found that about a third were using weekly.
00:04:49.660 --> 00:04:52.540
And if my numbers here are right, is that right?
00:04:52.860 --> 00:04:53.660
That can't be right.
00:04:53.740 --> 00:04:55.900
I'm gonna have to go back and check my own statistics here.
00:04:56.379 --> 00:05:00.780
But people are using pretty frequently, and seeing a lot of people.
00:05:00.939 --> 00:05:09.660
I was gonna say my informal non-research, just using my gut, is I went to the APA conference, which we talked about last episode, and it was AI everything.
00:05:09.819 --> 00:05:11.579
There was like no escaping it.
00:05:11.740 --> 00:05:21.020
I would say 85% of the booths had some sort of AI integration within their product, and so the sense is it's getting widely adopted.
00:05:21.100 --> 00:05:37.020
And I would really be interested in knowing is this adoption curve the fastest type of adoption curve of new technology or other kind of product that you would see, you know.
00:05:37.100 --> 00:05:45.260
So they talked about how rapidly some of these other technologies have been adopted, like the cell phone, the laptop computer, other things like that.
00:05:45.420 --> 00:05:59.420
And to me, from the numbers you're talking about, you know, something from 15% in a single year to say 80%, that seems like pretty rapid adoption or penetration, anyway, into the field.
00:06:00.060 --> 00:06:16.220
So I don't know the numbers on Psychologist, but I know in general, the adoption rate for large language models blows the world wide web, the smartphone, the PC, all of those out of the water by like multiples of four or more.
00:06:16.379 --> 00:06:21.660
Um, you know, they're talking like ChatGPT is claiming a billion users at this point.
00:06:21.740 --> 00:06:23.820
I think Gemini is also claiming a billion.
00:06:23.900 --> 00:06:25.260
I think Meta's claiming a billion.
00:06:25.340 --> 00:06:27.420
There's a lot of overlap there, and it's hard to tell.
00:06:27.660 --> 00:06:29.100
Gemini might be higher than that.
00:06:29.180 --> 00:06:34.060
You know, if you think about if you just use Google at this point, you're using Gemini.
00:06:34.460 --> 00:06:43.420
Load full transcript
But the people who are intentionally adopting this going to chatbots kills any any technology by far.
00:06:43.500 --> 00:06:59.980
And when you when you look at how long it took people to adopt a PC, I mean it took a decade or more, you know, like during the 80s, we had the same Apple TC that sat there gathering dust for 10 years, and it was like that for most non like people who aren't in the computer science field.
00:07:00.220 --> 00:07:03.340
It's just it's rapid adoption.
00:07:03.900 --> 00:07:22.220
Yeah, and so I think you know, for a lot of our listeners, uh, if you get a sense of being overwhelmed by everybody using this, and maybe some of the subtle pressures that even if you're a little reluctant to use to feel like you have to use it, I I think that'd be a natural feeling.
00:07:22.379 --> 00:07:33.020
There's so much changing in our society generally, and then to have this kind of permeate not just professionally, but also from you know a more leisure standpoint.
00:07:33.180 --> 00:07:35.420
You have kids using chatbots now.
00:07:35.500 --> 00:07:43.900
Um, I'm hoping it doesn't go down the road of you know, like social media where everybody said hands off and we've been living with the implications of that.
00:07:44.140 --> 00:07:57.020
But, you know, for our profession, too, given that we are, I would argue, a fairly conservative profession for good reason because of the privacy issues and the sensitivity of the data that we're exposed to, to have that at 80%.
00:07:57.500 --> 00:08:05.020
Uh, there's a lot of trying to figure out what role AI should play in my professional life.
00:08:05.340 --> 00:08:09.260
And I don't know that a lot of people have figured out that balance yet.
00:08:09.500 --> 00:08:21.900
Have you gotten a sense, like in your presentations and stuff, what type of things people talk about with you, about their concerns, or how they integrate it within their workflow, things like that?
00:08:22.939 --> 00:08:25.260
Yeah, I also know from the data as well.
00:08:25.580 --> 00:08:29.980
I mean, the concerns are some a lot of them pretty legitimate.
00:08:30.060 --> 00:08:32.059
You know, people are concerned about the environment.
00:08:32.299 --> 00:08:34.460
And I'll say this about the environment.
00:08:34.779 --> 00:08:39.659
Environmental concerns for AI are very real, but not how people think.
00:08:40.059 --> 00:08:53.019
So if you're to run a report using an AI report writer or to have a single chatbot conversation, not a lot of water use, not a lot of electricity, it would be like running a microwave for 30 seconds.
00:08:53.100 --> 00:08:55.820
I mean, less than than nuking a burrito.
00:08:56.379 --> 00:08:57.820
But and people will point to that.
00:08:57.980 --> 00:09:00.059
See, AI is not going to harm the environment.
00:09:00.220 --> 00:09:09.580
But that's pretty disingenuous because people who are really using AI, such as myself, are running multiple agents at once that are running for hours
at a time.
00:09:09.820 --> 00:09:16.379
So, you know, if you're concerned about using AI because of environmental concerns, that's legitimate.
00:09:16.460 --> 00:09:21.740
But I think if you're just using it as a chatbot or to help write reports, you're probably okay.
00:09:21.899 --> 00:09:34.940
If you're someone like me who's doing, or you know, someone who's on a$200 a month plan or burning thousands of dollars in tokens of compute, which power users are.
00:09:35.100 --> 00:09:39.100
Those are the people who are really using electricity and water.
00:09:40.139 --> 00:09:52.540
Basically So basically, you're accusing yourself of killing the planet, where the rest of us are just, you know, using very responsibly, and you're here just racking it up, all the damage.
00:09:53.019 --> 00:10:03.100
Well, we're all consumers, and we're all consuming products that are being software that's being built with AI agents and AI coding tools.
00:10:03.340 --> 00:10:21.500
So, yes, I'm more actively engaged in this, but the truth is, even if you're like, I'm not gonna use AI, but you use pretty much any other product, uh, the truth is that AI is is being used heavily uh for marketing, for research and development, for coding.
00:10:21.899 --> 00:10:34.860
So yeah, not to be too gloomy, because I do believe that AI has the potential to solve a lot of the problems that it's actually creating or perpetuating.
00:10:36.139 --> 00:10:36.300
Yeah.
00:10:36.779 --> 00:10:39.580
Well, I feel like we've uh deviated a little bit though.
00:10:39.659 --> 00:10:51.899
You know, I mean, I think the environmental concerns we do see in a lot of the presentations, they I probably get that question at least once almost every presentation, and try my best to answer it succinctly.
00:10:52.060 --> 00:11:08.540
I mean, my basic belief is that when factories came online in the Industrial Revolution, they were terrible polluters, they were very inefficient, they had all sorts of negatives, but over time efficiencies developed and their footprint became less.
00:11:08.620 --> 00:11:21.980
And I feel like that's the you know, that's the curve of almost every technology that inevitably breakthroughs come through to be more efficient because the companies don't want to be burning through all that, you know, resources if they had a choice.
00:11:22.060 --> 00:11:23.259
It's cheaper not to.
00:11:23.420 --> 00:11:29.420
So I think there's well, you know, this is the goal, this is the gold rush.
00:11:29.500 --> 00:11:31.659
I think there will be a time where they don't want to.
00:11:31.820 --> 00:11:36.700
Otherwise, they have to jack up the price eventually and people won't stand for it.
00:11:36.860 --> 00:11:37.899
You don't think so?
00:11:38.379 --> 00:11:46.220
I think that companies started taking better care of the environment because we the public made them.
00:11:46.620 --> 00:11:57.100
We had a massive movement in the 1960s and 70s, and even prior to that, where people were like, you need to take care of the the planet.
00:11:57.259 --> 00:12:07.980
Now, some of that's been rolled back, but I want to encourage people to you know know that they have some say in how the how companies treat the environment.
00:12:08.139 --> 00:12:13.980
And but partly you vote through your dollar, and I've already admitted that I'm not I'm not perfect in that regard.
00:12:14.220 --> 00:12:18.540
But I think to see, like, oh, the capitalism will solve this, I don't believe that.
00:12:18.700 --> 00:12:26.060
I think that external pressure from the people from us, from regular citizens, is also necessary.
00:12:26.700 --> 00:12:29.980
Well, and I think the advantage too is we don't have a monopoly here.
00:12:30.060 --> 00:12:38.220
We have kind of the big three, and we do have a couple other competing firms, and we won't even address necessarily all the Chinese models that are out there.
00:12:38.379 --> 00:13:04.379
So, you know, uh that's that's kind of the root of competition that consumers, if they are, you know, one company is particularly egregious about not being more conscientious about their environmental footprint, and another one is, you know, people can vote with their feet too, which is nice, versus somebody who has some breakthrough technology and they're the only one in town, so there's no incentive for them to do have that type of change.
00:13:04.540 --> 00:13:05.899
So point well made.
00:13:06.139 --> 00:13:18.779
You know, uh I my own use of AI, I've had a summer to have some relative free time, at least it seems like it now, and now that I'm back at the university and we're in full, full mode.
00:13:19.019 --> 00:13:26.300
I've got to say I've I've begun to learn more about how to treat it as a partner, not just an answer bot.
00:13:26.540 --> 00:13:51.899
And what I really have liked about it is that it kind of tracks my thought process when I'm giving it feedback and pushback, and it's starting to offer to create like rules, uh like style books, rule books for the type of things I want it to do for certain projects, and that I can just upload that and it serves as a reminder of kind of how to structure it.
00:13:51.980 --> 00:14:03.180
And so I know you've done this for a long time, but I've not had that much embedded, extended use until the summer with the AI, and I'm really I'm really liking it.
00:14:03.259 --> 00:14:07.100
It's starting to really have my voice down pretty well.
00:14:07.259 --> 00:14:15.820
I like the back and forth, and then I did what you suggested a couple episodes ago about having it run a personality of my usage.
00:14:15.899 --> 00:14:21.659
Uh, and I liked what it said to me, which was basically you know, you don't want just the answers.
00:14:21.740 --> 00:14:35.500
You want to have an iterative process when you are writing things, you want feedback, you don't want to sound like a bot, you know, and that you want to make it a very practical hands-on from what research you found.
00:14:35.659 --> 00:14:45.580
And so uh, you know, I I don't know, I found that to be very helpful insights, and I believe I've gotten even better responses from it as a result.
00:14:45.659 --> 00:14:47.820
Uh I've even dabbled, Adam, you be proud.
00:14:48.060 --> 00:14:56.300
I dabbled into projects too, as I have to do some refiguring of some of my classes, and I had to figure out times and things like that.
00:14:56.540 --> 00:15:03.899
And uh I was able to develop the global kind of rules for the project, and it's really enhanced effectiveness.
00:15:04.060 --> 00:15:18.379
And so um I think one of the I think one of the big difficulties for professionals is do you get an out-of-the-box program that supposedly has already been curated for your profession, but it's not been curated to you.
00:15:18.540 --> 00:15:39.899
And I don't know, I haven't spent enough time with these report writers to know whether the more all AI ones kind of a hue to a generic this is you as a professional, or if it really starts to mold in the same way as the just general LMs do to your style and techniques and things like that.
00:15:40.060 --> 00:15:47.180
I know you've worked with and consulted with report writers a lot more, so uh what insights do you have kind of regarding that?
00:15:48.379 --> 00:15:50.540
So a couple of things.
00:15:51.019 --> 00:15:53.500
You may I said a lot.
00:15:53.740 --> 00:15:54.620
I said a lot there.
00:15:59.420 --> 00:16:15.180
Uh usage because you asked what school psychs are using AI for, and what last time we asked, they said for writing reports, rewriting sections, answer work questions, understanding writing explanations, creating presentations, things like that.
00:16:15.500 --> 00:16:19.340
So that's what what we the last time we asked, what people said they were using it for.
00:16:19.580 --> 00:16:24.220
But to your question of do these products add something different, I guess it probably depends on the product.
00:16:24.300 --> 00:16:25.340
I don't know all of them.
00:16:25.500 --> 00:16:34.940
And you know, yeah, I have to be completely up front that I do consult, have consulting relationships, most notably with with par on theirs.
00:16:35.100 --> 00:16:46.139
And I can tell you that one of the benefits from most of these is that they are, well, hopefully, you've got to read the fine print, but they're they're HIPAA and FERPA compliant, uh, which is nothing.
00:16:46.779 --> 00:16:48.620
Um you need to protect client data.
00:16:48.779 --> 00:17:09.579
And then I can say that I know PAR has a lot of development around their product to make sure that hallucinations are minimized, which hallucinations are just the generation of incorrect information that the way that it's organized is is logical.
00:17:09.819 --> 00:17:22.220
I mean, I started out with a notebook, Gemini Notebook, which used to be notebook LM, of all of the report writing articles and and books that I had and went from there.
00:17:22.299 --> 00:17:42.859
So so I know for at least the work that I did is we came from a standpoint of what does the literature suggest is best practice, how can we organize in that way, and then without getting too much into intellectual property, how do we constrain this model so that it um it doesn't hallucinate, that it doesn't use language that's harmful, things like that.
00:17:43.019 --> 00:18:17.629
So I would guess it really depends on the product you're referring to.
00:18:17.709 --> 00:18:40.029
And the only one that I really have in-depth um knowledge of at this point is pars, and I can say that they've definitely invested quite a bit in making sure that it's not just you know your frontier models, although most of them do run on frontier models, meaning OpenAI or Claude or Jim and I or the like.
00:18:41.229 --> 00:18:56.349
Do you find, though, with your experience with these report writers, the ability to mold uh over time with the constant interactions with them more to your personal style and preferences in, say, your report writing process?
00:18:56.429 --> 00:19:14.429
So let's say, would these report writers start picking up on, hey, this guy really leans hard on wanting applied interventions, but likes to kind of wordsmith and have a certain way of expressing the overall cognitive functioning of the individual.
00:19:15.789 --> 00:19:20.269
Well, there's functionality built in, so you can tell it what you want.
00:19:20.909 --> 00:19:27.149
You can you can choose tone, you can choose length, so so it's customizable in that way.
00:19:27.389 --> 00:19:55.149
As far as learning, I mean for any of these models, they're pre-trained, so they they all that so there's different phases of training that happens, and so they use like reinforcement learning, for example, where they do a bunch of training before this thing even comes out, and then they fine-tune it essentially by having humans give feedback, and now they're actually having agents give feedback to it.
00:19:55.309 --> 00:19:57.789
So, like, you know, which of these do you prefer?
00:19:58.029 --> 00:20:00.349
Which of these two outputs, or that was wrong?
00:20:00.669 --> 00:20:02.509
There's a lot of that that goes into it.
00:20:02.749 --> 00:20:14.669
So none of those models can really fully learn beyond what they were trained or pre-trained on, but they they can somewhat remember preferences.
00:20:15.869 --> 00:20:16.909
Now they get wrong a lot.
00:20:16.989 --> 00:20:26.349
Like I just today, you know, was using semi availability, you know, and usually it knows my the link to my my availability, and it was like, insert your link here.
00:20:26.429 --> 00:20:29.309
I'm like, come on, how we've been doing this for four forever.
00:20:29.469 --> 00:20:34.749
So there's there is some memory built into some of the frontier models.
00:20:34.909 --> 00:20:42.589
But my understanding of these psychology-specific ones that no, they they don't have as much of that built in.
00:20:42.669
--> 00:20:46.189
Because the same, this is a model, it's not everyone's not getting their own.
00:20:47.069 --> 00:20:54.269
Like when you have a ChatGPT account, for example, you can tell it to remember things about you and you can tell it things not to remember.
00:20:54.429 --> 00:20:55.709
It's not like that.
00:20:55.789 --> 00:21:08.749
It's it's just tapping into the APIs, which is an intermediary between you or the company and the larger frontier models like OpenAI, which is ChatGPT, or Anthropic, which is Cloud.
00:21:08.989 --> 00:21:12.509
So my understanding is at this time that that doesn't exist.
00:21:12.669 --> 00:21:14.589
Do I see it moving there in the next few years?
00:21:14.669 --> 00:21:16.589
Like people being able to use templates?
00:21:16.749 --> 00:21:17.309
I do.
00:21:17.469 --> 00:21:25.069
But it's my understanding, and I can't speak for all of the companies, that right now, no, it's not as customizable.
00:21:25.709 --> 00:21:26.189
Mm-hmm.
00:21:26.829 --> 00:21:27.389
Yeah.
00:21:27.709 --> 00:21:50.029
Well, and I mean, I think that's a reasonable approach for a specific, like, psychology-oriented product to not just give total free reign to the individual using it, but having best practice, for instance, embedded in some ways, it may be more of a uh kind of structured setting than what we had prior to AI, right?
00:21:50.189 --> 00:21:59.469
Because like if I, as a licensed psychologist, want to go rogue and write the things, you know, whatever the hell I want to, there's nothing really stopping me.
00:21:59.629 --> 00:22:02.189
I've got my license, I can throw it on out there.
00:22:02.509 --> 00:22:08.749
Even if I want to ascribe to a you know phrenology projective, I can go ahead and do that.
00:22:08.909 --> 00:22:13.309
And an AI with some of the structure may push back a little bit and say, well, Dr.
00:22:13.389 --> 00:22:21.229
Florel, the phrenology projective probably is not enough to diagnose paranoid schizophrenia or something like that, right?
00:22:21.469 --> 00:22:40.429
So maybe it's going to lend to um, well, if you are an excellent report writer, very knowledgeable, it may be restrictive, but on the opposite end, it's going to lift up the least competent people to at least have some of the best practices in how we go ahead and do things like assessment.
00:22:40.749 --> 00:22:41.069
Yeah.
00:22:41.309 --> 00:22:47.149
And I can tell you that there's been customer feedback that's, oh, we would like this integrated.
00:22:47.309 --> 00:22:53.549
And because part of you use that feedback to improve their product all the time.
00:22:53.709 --> 00:22:58.349
But if there's been times where it's like, people want this, and like, no, that's not best practice.
00:22:58.589 --> 00:22:59.709
We can't do that.
00:22:59.949 --> 00:23:05.069
And so there's times where you don't want it to be customizable to however a person wants it.
00:23:05.309 --> 00:23:11.309
That's nothing that happens frequently, but there has been instances that I can remember where, like, no, we're not gonna do that.
00:23:11.869 --> 00:23:14.109
That's not ethical or that's not best practice.
00:23:14.269 --> 00:23:26.589
So yeah, sometimes it's good to not have it just be how anybody wants it, because there are unfortunately people in every field, and ours included, that are doing things that they shouldn't be.
00:23:29.149 --> 00:23:31.469
Yeah, I know I know we've been talking.
00:23:31.869 --> 00:23:35.069
Talking a lot about report writing, and that isn't the only focus.
00:23:35.389 --> 00:23:43.469
What other beyond the assessment and report writing do you see the use of AI within, say, the clinical profession?
00:23:43.629 --> 00:23:52.589
I know one that comes to mind right away for me is the electronic medical records and creating like counseling notes, transcription services, uh, and things like that.
00:23:52.829 --> 00:23:57.309
As far as your research, are are you getting a sense like where do clinical?
00:23:57.629 --> 00:24:11.229
I know school folks are going to definitely heavily side on the report writer aspect, but in clinical world where you're doing a lot more counseling as part of your ordinary job for most, I imagine there's a tad different focus.
00:24:11.789 --> 00:24:12.909
Absolutely.
00:24:13.149 --> 00:24:24.189
And so, yeah, a lot of a lot of treatment plans, treatment goals, summaries, soapnotes, you get emails across fields.
00:24:24.589 --> 00:24:26.269
People are using AI for emails.
00:24:26.349 --> 00:24:42.429
And whenever I whenever I give a talk, I say this is the lowest hanging fruit, you know, I talk about you get those emails from someone who who is rude or impatient, and you just you're you're miffed and you just want to like come back to them and be like, leave me alone, you know.
00:24:42.669 --> 00:24:58.829
But you want to be professional and you want you know help helping professionals, so you want to be kind, and so you have to sit on that email and come back to it over and over again and soften it and soften it and soften it, so you're still getting across your meaning, but you're not telling them, hey, buzz off.
00:24:59.549 --> 00:25:06.589
AI is perfect for helping with that, assuming you have a HIPAA compliant tool that can help with that.
00:25:06.909 --> 00:25:14.749
But we see that so that's when I see across both that that's the biggest one where there's a lot of overlap is people like yes, emails.
00:25:16.029 --> 00:25:28.989
Well, it's a it's a client-oriented or student-oriented type of profession, so of course you've got to interact with a wide range of humanity, and like you said, some are a little more difficult to deal with than others.
00:25:29.309 --> 00:25:40.189
Has there been talk in the clinical world about concern about treatment plans getting overly generic because of these EMR systems that are recommending?
00:25:40.509 --> 00:25:52.269
Or is it the opposite where they're actually way more
personalized because it's able to say access the transcripts of those notes and being able over time?
00:25:52.589 --> 00:26:00.429
And I'm not sure if these systems really do look over time or if they are just transcribed, put it into the note, delete the transcription.
00:26:00.589 --> 00:26:02.109
What type of model is used?
00:26:02.269 --> 00:26:04.269
Have you noticed as far as with the MRs?
00:26:04.429 --> 00:26:23.389
Do they have more entire access to the records to make the most up-to-date recommendation, or does it more self-contained where it's just for that session, takes those notes and uses just that in like its own little compartment and then purges it from their system?
00:26:24.429 --> 00:26:37.229
I'm not as familiar with electronic medical records like your Epic, but I do know research looking at, let's say, IEP gold generation and recommendations for psychology reports.
00:26:37.549 --> 00:26:39.069
Some of the research that I've done.
00:26:39.229 --> 00:26:46.509
And that is, yes, I get a lot of people saying, oh, these are going to be generalized, they're not going to be individualized, they're going to be, you know, road.
00:26:46.749 --> 00:26:54.829
But the truth is, if you ever had an IEP, you're you're probably going to find, if you use the same teacher, the same goals over and over again.
00:26:54.909 --> 00:27:02.749
And probably they're going to be not very understandable, you know, four to five trials, 80%, you're like, you're like, whoa, I I can't follow these numbers.
00:27:02.989 --> 00:27:10.829
Or if you read a psychology treatment plan or recommendations, it's the same, the same recommendations.
00:27:11.069 --> 00:27:16.109
You know, for a school psychologist, it's going to be time and a half, preferential treatment, and things like that.
00:27:16.349 --> 00:27:27.229
So the truth is when you look at the research, they're actually more individualized because yeah, AI can analyze a ton of data at once.
00:27:27.309 --> 00:27:36.429
So they can look at the student or patient profile, their strengths in areas of needed support, they can look at their interests, they can look at their location.
00:27:36.589 --> 00:27:46.989
So here's here are some you know things that might be of interest to them or supports to them in their location, things that you just probably don't have time to research.
00:27:47.069 --> 00:27:50.509
Now you have to double check, make sure that this is all accurate.
00:27:50.989 --> 00:28:05.789
But what the research seems to be suggesting is that AI actually has more individualized recommendations that are actually as at least as good or better than psychologists and special education teachers.
00:28:07.229 --> 00:28:09.069
Yeah, that's that's really encouraging.
00:28:09.229 --> 00:28:27.549
I've just started to really dip my toes into just even using a general AI for some recommendations and just giving like a broad, like, you know, severe intellectual disability in like their 40s, you
know, and extremely limited adaptive, obviously.
00:28:28.349 --> 00:28:30.909
What sort of general things would you suggest?
00:28:31.149 --> 00:28:54.429
And I that's not as specific as all of it, but you know, I had generated a few recommendations with a case that I had not long ago, and I really did like the extra few options and things that I hadn't thought about, like you know, an occupational therapist coming into the home and you know making the environment better suited for an individual at that particular operating level.
00:28:54.589 --> 00:29:11.869
And so um I see some real promise, and of course, it you know, Likety Split was able to spit them out, so it's not like a huge time drag at typing for me a couple of sentences or for you, I'm sure, chatting with it a little bit, and you're not disclosing anything with that level of detail.
00:29:12.029 --> 00:29:18.909
And then, of course, within these more closed systems that are HIPAA compliant, you can obviously get a lot more detail on that.
00:29:18.989 --> 00:29:20.349
And I've been impressed.
00:29:20.509 --> 00:29:23.869
I always kind of use the smell test on the recommendations.
00:29:23.949 --> 00:29:25.149
I've been around a while.
00:29:25.309 --> 00:29:28.269
You know, if something seems a little wacky, I am gonna look up that.
00:29:28.349 --> 00:29:37.389
But if it seems like, yeah, that's nice, logical, makes sense, and is good for this, you know, seems consistent with what all the other things are, you know, I may go with it.
00:29:37.469 --> 00:29:41.149
So you're not spending hours having to research every single recommendation.
00:29:41.229 --> 00:29:48.589
I mean, there's certain things that are not going to uh you can kind of trust just because you know have your own knowledge base.
00:29:48.749 --> 00:30:02.109
And the moment you don't have a knowledge base, I would certainly say you need to go follow up and try to make sure there's not some hallucination so that you're what's the famous old example, you know, gluing pepperoni on your pizza kind of thing.
00:30:03.709 --> 00:30:04.349
Yep.
00:30:05.309 --> 00:30:11.709
The old Gemini, I think that was Gemini that suggested as a recipe glue in pizza or something.
00:30:12.189 --> 00:30:12.669
Right.
00:30:12.909 --> 00:30:14.269
Right, I remember that.
00:30:14.509 --> 00:30:19.629
So believe it or not, though, Adam, we are almost out of our time for this episode.
00:30:19.709 --> 00:30:38.909
So any uh wrapping up kind of thoughts we've covered a lot, uh, from environmental concerns to uh report writers and a little bit into the EMR and writing counseling notes and kind of how psychologists and mental health workers are beginning to incorporate AI.
00:30:40.109 --> 00:30:44.749
Yeah, you know, there's a lot of efficiency to be gained.
00:30:44.829 --> 00:30:53.389
I think there's a lot of improvement in services that can occur, but I also want to validate people who have concerns.
00:30:53.709 --> 00:31:02.509
Because, you know, I talked about I kind of went on a tangent there was environmental concerns, but anything related to confidentiality bias, all those things are legitimate.
00:31:02.829 --> 00:31:12.429
And so, you know, I don't want to first of all I don't want to come off as too bollyannish about this, and which maybe I did, but you asked about people have concerns.
00:31:12.589 --> 00:31:27.629
Yes, there's a lot of concerns, but there's whether you like it or not, this is this is happening, and so I don't think you have the choice to to sit this out on a long enough time scale, you know.
00:31:27.869 --> 00:31:31.789
So well, and I think some of the issues that you just brought up there, right?
00:31:31.949 --> 00:31:33.309
There are ways of mitigating it.
00:31:33.469 --> 00:31:38.189
We may even devote some future episodes of ways to mitigate prompt.
00:31:38.589 --> 00:31:46.989
You know, AIs are infinitely malleable and more responsive to feedback than probably any other technology up to this point.
00:31:47.309 --> 00:31:59.869
So with some guidance, it can create whatever world you want in regards to feedback and considerations when it's providing recommendations or you know, how to write reports.
00:32:00.029 --> 00:32:11.309
So the fact that if you do have concerns, it just means maybe I need to spend more time on figuring out how to get these AIs to do it what I want to do at the level that I want it done.
00:32:11.469 --> 00:32:23.709
And I think that's encouraging too, because that means it's simply a case of us being able to inform that AI and train it to have that particular perspective or write in a particular way.
00:32:23.949 --> 00:32:26.429
But it doesn't mean like the technology can't.
00:32:26.669 --> 00:32:28.269
I think it's certainly capable.
00:32:28.589 --> 00:32:34.989
You just maybe take a little longer than just uh typing in something and having it automatically spit stuff out.
00:32:36.509 --> 00:32:41.149
Just be super, you know, super critical and super careful.
00:32:41.229 --> 00:32:42.909
And I think that's important too.
00:32:42.989 --> 00:32:52.429
You know, just look over everything you're signing off on is your an AI cannot have a therapeutic relationship, an AI does not have liability.
00:32:52.749 --> 00:32:53.229
You do.
00:32:54.589 --> 00:32:58.909
I think with that cautious note, we'll go ahead and wrap up for today.
00:32:59.309 --> 00:33:02.029
Really appreciate you all tuning in.
00:33:02.109 --> 00:33:19.789
Uh Adam and I always have a range of things, and we're constantly flooded with the new AI advances, and so uh even though this is a more general episode, I think as you see, there's just so many ways that the field is being influenced by AI and will continue to be.
00:33:20.029 --> 00:33:23.789
So stay tuned for more in a couple weeks in our next episode.
00:33:23.869 --> 00:33:26.829
And I hope your AI adventures are productive.
00:33:26.989 --> 00:33:28.109
We'll put it that way.
00:33:28.349 --> 00:33:29.549
Thanks, Adam.
00:33:29.869 --> 00:33:39.149
Thanks for