In this episode, we delve into the crucial topic of ethics and confidentiality when integrating AI into therapy practices. Many therapists are hesitant to adopt AI due to concerns about privacy and data security. I share insights from real-world applications, like how medical offices use AI to record and transcribe sessions while maintaining HIPAA compliance. We explore the importance of using HIPAA-compliant AI tools and obtaining informed consent from clients to ensure their privacy and trust.
We also discuss the potential biases in AI outputs and the importance of customizing AI tools to align with our therapeutic approaches. Additionally, I address the issue of AI hallucinations, where AI might fill in blanks with inaccurate information if not given complete data, and the need for therapists to review AI-generated notes thoroughly. This episode is packed with practical advice and real-life examples to help you navigate the ethical landscape of using AI in your therapy practice.
https://www.berries.icu/therapists?utm_campaign=podcast
Transcript:
00:00
This is Ask Me Anything AI sponsored by berries, dot ICU slash therapist, what’s going on in the AI world? It’s changing every day. Yesterday in the AI club, I had a intro to AI people that are brand new and just have a lot of questions about how AI works and all of that. So if you’re here watching this and you have questions, let me know, trying to get into see we had any questions come through. I know someone messaged me about ethics, and I’d love to hear your thoughts and what you guys think about ethics. So confidentiality and privacy was one thing that people always ask me about with using AI. I think it’s the biggest reason a lot of therapists don’t want to use AI. Every time I post something about writing progresses today or something like that, I always get like, pushback, really. People say it’s not confidential. I would never do that. I’d never use AI with my clients. But it’s, it really is the wave of the future. And I don’t know if you guys go to doctor’s offices at all, but I was in a neurologist’s office, I guess, probably, probably a year ago now, and she was talking to a recording in the ceiling. She was saying, okay, record, and she was having the session with me, and it was recording the whole thing, and then it dictated it. And before I was even out of the office. They had sent me a summary of what had happened in there. Sure, I signed something in the, in the, you know, entry paperwork or something like that. But a lot of these medical offices are using AI to record our sessions already, and almost all of these platforms are HIPAA compliant. So, you know, I mentioned in the beginning, this is sponsored by berries. They’re an AI note taking HIPAA compliant program that records your sessions, and then after the sessions over, it even deletes the recording instantly. It doesn’t even store it. That’s how, like, you know, confidential and private it wants to be. It really wants to follow those HIPAA standards. It deletes it. It turns it into a node immediately. It does give you a transcript, and then it has tons of capabilities where you can go in and edit it and do amazing things for the but the important thing is, it is HIPAA compliant. So a lot of us, we were talking about this in the AI club beating yesterday. A lot of us have zoom I have something called Fathom AI that’s on in my Zoom Room, and that’s it, to my knowledge, is not HIPAA compliant, even though some of us have the zoom that has a BAA and it is HIPAA compliant, overseeing our clients, but these little plugins that we’re using even zoom AI, I need to go check on that to see if it’s HIPAA compliant. But anyway, using AI, that’s the biggest concern, is like, is this confidential? Is this private? And then the other part of it is like, what are my clients going to say if, if I’m using AI, or if I want to record a session, or, you know, I’m putting their information into chat GPT, trying to make some homework assignments, or something like that. And really, it’s our job to explain and give them informed consent that we’re using any type of AI that we’re using within our private practices with our clients. So if we want to record the session, we have to make sure, one, the way we’re recording it is HIPAA compliant, and two, that they have our consent to be recorded. That’s a huge one, because you could easily record the stuff in the background and our clients wouldn’t know, but that would not be ethical for us to do that, right? It would not be ethical just to put on a recording and not even tell them that with being recorded and then used to turn into a progress note, or however we’re going to use the recording now, berries is like, super innovative, and within the next couple weeks, I think they’re going to be able to take, you know, the recording, they get rid of the recording. We’re going to have the note, but they’re note, but they’re going to be able to take that note and have a button basically, that turns that note into a treatment plan, or turns it into a homework assignment for your clients, but all of it in a HIPAA compliant environment. So that’s really what you want to be looking for. Now, a lot of us, especially in the clinical AI club, we talk about using chat to PT to write our progress notes or treatment plans or that sort of thing. But that is a place where, of course, we have to adhere to not entering any phi at all. And you guys probably see this in the Facebook groups too. Like, if a therapist helping therapist Facebook group, you know, they’ll say, I got this really tough case, and this is what’s going on. And even then, you have to be super careful to make sure that that client is not identifiable in any way at all. So you have to use those same standards with chatgpt or Claude or Gemini, any of these free chatbots that we can just, like, hop on real quick and just put information in. So I’m in the clinical AI club. I do have a list of, like, the 20 things that constitute phi. So it’s, you know, any identifying numbers, dates of birth, actual names, addresses, information about relatives, anything like that. There’s a list of 20 you can Google, like, what, what constitutes phi? Google it and make. Sure if you’re putting anything into chatgpt, none of that’s included, and it’s also, look at what you’re putting in there. Read back to yourself. Is this something, if I were to read this as a client, what I knew who you were talking about, or if I was this person’s neighbor and I read this, would I be able to figure out who they are? So really, following standards and guidelines like that, if you’re using the Free Church BT, another sort of ethical thing that comes up with using AI is like bias within the input that it gives us back. So, you know, there’s a lot of research around this, and I don’t know ton of like statistical information about it, but I have read quite a few articles. It just says that that AI is biased based on the way it’s been programmed. So it’s programmed from information that it’s gathered from the internet, and it’s basically based on certain sources that they’ve used to fill this database. So it’s not all. You know, it’s not like the entire Internet. It’s taking information from different places. So because of that, it’s a conglomerate of information that it’s giving you back, like output is giving you back, but it is biased information based, obviously just on the training that it’s been given. So that’s another ethical thing that people have concerns about when they’re talking about using AI in general. I’ve heard people bring that up, but yeah, but it’s trained on the certain bit of information. And there are ways, and we talk about this in the clinical AI club to get super, super specific with the way that you are training your chat GPT, or even if you want to build your own GPT, we have a training that teaches you how to do that. That’s for, like, the pay version of chatgpt, where you can go in and train it to not be biased, to be trained basically on our own biases, on, you know, we have a certain theoretical perspective, or clinical perspective, the way we see things, that there’s a certain way we want our progress. And it’s to be written. There is a way to go in there and train it to, you know, produce what you wanted to produce. So there are ways to get around that and make sure that it’s giving you the information that you want to have back if you’re worried about, like, social justice issues and that sort of thing. You could train it to be more knowledgeable around that type of information when you’re working with this. But still no phi when you’re doing that. So those are just some things, some of the ethical things. I hope that answers your question. I don’t think there’s any other things doesn’t really fall in the ethical category, but this idea of chatgpt or AI hallucinating information. So sometimes, if you put in a client blurb and you tell it to turn it into your progress note, it’ll, like, fill in the blanks for you with information. So a lot of times that information isn’t accurate, like if you have programmed chatgpt To give you a soap note, and part of that soap note is the objective where we’re looking at, like, tangential thinking, or, you know, disheveled in session, like those kinds of notes that we make in our progress notes. If you don’t fill that stuff in and you’re asking for it to be a soap note, chat, GPT is going to hallucinate answers. Basically, they’re going to come in and make up answers, basically. So what a lot of us do, you know, we kind of get lazy with this stuff. It makes it super easy to write progress notes. You copy paste, you’ll put it over into your EHR, and you’re done with your note. And might bypass that step of let me just read through this to make sure it’s actually accurate with everything that it did as an output. So you have to double check then too. That’s not really any ethics, but that is sort of in that realm of we have to be careful what we’re taking from here and putting into our EHR, because this is our client record. So, you know, we might not realize that it said disheveled or no suicidal ideation, where maybe there was suicidal ideation, because we didn’t put it, you know, in the information we gave it, we had to be real careful about that. So I will see you guys next week for Ask me anything. AI bots are by berries, and let me know if you have questions about berries too. I’ll put a link in the bottom in case you want to go check the mouse. All right, bye, bye.