Name: Navigating AI in a Clinician Centered Practice
Uploaded: 2026-04-24T19:30:05.481Z
Duration: 54 min 21 s
Description: Navigating AI in a Clinician Centered Practice

Transcript for "Navigating AI in a Clinician Centered Practice": And with that, let's really get some of our material going and started. So I will start with again. Today, we're talking about navigating the now, and we picked that name, as you'll see in some of our early slides. And what we're talking about is things are moving quickly in AI. And at first, when we were playing around with titles, it was like navigating the future, and it was like, no. The future is actually now. We're having to face these things and face these decisions as clinicians right now in the space, in the world that we're sitting in. I wanna do some brief introductions, and then we'll go through the agenda, and, and share some of the information with you. So, first, let me introduce myself. I am, doctor Lindsey Oberleitner. I am a clinical psychologist myself, and I'm the head of clinical strategy here at SimplePractice. I specialize clinically in chronic health conditions and addiction, and also hold appointments at the Yale University School of Medicine and the Oakland University William Beaumont School of Medicine and University of Bridgeport. Some of those schools are a mouthful to to get out. With that, I'm gonna pass it over to Allie. Thanks, Lindsay. Hi, everyone. I'm Allie Hartley. I'm the chief legal officer here at SimplePractice. I'm a licensed attorney, and I've been practicing in the health care field, mostly representing providers and working with provider organizations, for the last twenty years of my career, including, you know, applicable today. You know, once sent, I was working for a large hospital system where I also served on an ethics committee for one of the hospitals there, but have a lot of experience in health care regulations that impact providers. Awesome. Thank you, Allie. And now, to give you a little sense of the things that we're going to talk about today, we're gonna start with a little bit of evidence based horizons. Where are we with some of the clinical AI data that we know as a field? And then Ali is gonna share with you some of the guardrails. What are some of the ethical and legal considerations you should have when thinking about integrating AI in your practice? And, finally, we will, think about how do you build your own future vision. Whether you're someone who's like, not at all right now, but I wanna think about how I would consider tools in the future. Or if you're someone who's like, I wanna integrate in all the ways I can do so safely and ethically. Wherever you fall in that spectrum, we wanna talk about a little bit of framing around how you think about making those decisions and how you can take in different areas of information to come to those conclusions. And finally, as mentioned, we'll make sure to save some q and a time at the end and look forward to your questions. So please don't please don't hesitate to, add any and all things you might be thinking of. So I mentioned when I said the title that, you know, part of our decision, even at the point of the title was, like, why now? Why are we using the phrasing now and not the future? And I think one of the things that I think about often in my own life and in my own kind of clinical mindset of how I'm thinking about the world is when we are navigating this now, it came quickly. Right? There we have a long history of things that, you know, integrate data and algorithms, but the use of AI in our daily lives really seem to hit and come on quickly. And although we might think of it as a future concept sometimes in our clinical practice across some areas, it really is expanding in our day to day, both for us and our patients or clients. So we really see it as a present reality. Again, not in how you make the choice necessarily, but how you think about that it's here. It's here and it's impacting you, it's impacting your clients, and it's impacting how you talk about this in practice right now. So wanted to share a couple interesting facts along the way about how, again, we came to this concept of, like, this is the now and not the future is when looking at some data. So a Pew Research study suggests that fifty five percent of adults in The US acknowledged regular use of AI in their personal and professional lives. That's a whole lot, except for what I find really interesting and why I put that fact right alongside this next fact is that's how many of us acknowledge that we're using AI. We are purposely going in and saying, I'm gonna, you know, prompt AI for this question. I'm gonna engage with this AI tool. Those are people who are acknowledging use. But the truth of the matter is AI is integrated in our daily lives in so many ways that we may not even realize right now. Even as simple as things like weather forecasting, personal assistant streaming services that try to tell you your next best thing to watch or online shopping, It's here, and it's here for us. And, again, it's here for our patients too. So it's one of those things that I think, you know, coming from the addiction world of treatment, you know, one of the things that I often said when I was was providing trainings was ignoring and not naming the thing, not asking about use was always one of the things that led to doesn't lead to better outcomes. Right? Acknowledging it actually helps. So I'm you know, part of it is just recognizing it's here. It's here for all of us. And, again, as that continues to rise, so can concerns about how AI impacts our lives, our daily lives, and our patients' lives. So we need to think about it. Right? We need to think about these impacts, and we need to think about where it sits with us. So the next layer, I'm kinda thinking of this as like a funnel. So, like, AI is here in our lives, day to day lives. It's different when it starts touching health care. So I'm going to narrow down a little bit from broader life into what's AI use look like in health care right now. And these numbers continue to grow. So when we're thinking about the big official surveys, what we can say is the growth is exponential within health care. From the practitioner practitioner perspective, again, in health care, not specific to mental health yet, I will get there. But there has been year over year almost an 80% increase, in the use of AI in the health care system by practitioners. From a patient perspective, approximately forty percent of adults say they're okay with AI use in their health care. And a similar amount even say that they really believe that AI being integrated into health care could improve their outcomes. So we don't have this in as a poll. And as I mentioned, we can't see, everyone's responses, but I'll ask you to reflect for a moment. If you're someone who has ever touched AI, you've gone to OpenAI, Claude, ChatGPT, however you've accessed. What are some of the questions you've gone and asked? I can sure acknowledge that I've gone and been like, explain this result to me, right, on my own or my children's health care questions. People are going, and they're seeing benefit of being able to get more information. Now this is different than mental health care. Right? In physical health care, we've probably all had those experiences where you go and you walk out of a doctor's office. You're like, I have no idea what that meant. They used a term. I walked away, and now I'm left with Google. However, I think this is one of the reasons why people are seeing growth in that health care practitioner use so much is we know that there's not time to provide all of that information. So growth has been far more rapid in broader health care than we've seen in mental health care as well. In addition, health care related AI devices have continued to grow rapidly over the years also in the most recent years. But let's think a little bit now about that use within mental health professionals. So that rate continues similarly to increase, but at a lower, percentage within mental health professionals. So upwards of twenty percent of clinicians report using AI in some way, in their practice in mental health practice. This does not mean always this isn't to say one single approach to it. This doesn't mean documentation support, doesn't mean, other client facing AI. It can mean things like, I'm using it to help me build my marketing tools. Right? There's a whole host of ways in which people are using AI in their mental health practice, but it is rapidly growing. And just over half, according to a APA survey, report having some knowledge of AI. But, again, this is a real opportunity for us if we think about how deeply embedded AI is in our lives, and we know that it's coming into the mental health field, as clinicians and as providers, we we have a really important role of having a voice in how those things are developed and where it's working, where it isn't, and where those guardrails need to be as well. So being engaged in that process, whether we use it or not, is really important to stay informed and on top of this. Also, there are really clear concerns that get raised, like breaches, social harms, biased inputs, lack of testing, transparency, inaccuracies. These are all very real concerns that mental health professional professionals hold about why they may be hesitant to use AI. As we'll go through today, I think part of it is being able to tier and think like, what are you using it for and what level of those concerns are more priority based on the way in which you're using it. And finally, in thinking about usage, like I said, I'm gonna funnel down even tighter and think about our patients and clients' use of AI. There's a lot of conflicting rates on this, but I'm not gonna read through all the numbers on the slide except to say that a lot of people feel like, going to an open chat is beneficial to them. They seek it out. We may hold concerns about that use. We may know the harms that can occur when that's not done in a controlled way, but what we know is from surveys that it's actually a higher rate of individuals report using an AI based chat for mental health if they're already in mental health treatment or seeking mental health treatment. So, we aren't gonna talk in detail about this today. It's just to give an overall frame, but I'd say this is, again, one of those things coming from my own history of working in addiction. I'd say, if we don't ask our clients about their use, we're we're really missing out on a chance to hold a conversation about where safety is, how what they're getting and not getting, and how we think about keeping them safe in that process. So it's an opportunity for discussion, but what we do know is many people are continuing to use and seek new ways to use AI, for their own mental health. So before I pass to, Ali to talk a little bit more about legal and ethical guardrails, within the use of AI in in practice, I do wanna make sure we go in with this in a frame that not all AI is the same. And we'll come back to this when we wrap up at the end. But being able to really distinctly think about the buckets of categories of things that you might use AI for in practice. And what I will say is they hold very different levels of what we need in kind of clinical support to know that they work, what we need to think about in safeties, how deeply involved we need to be as clinicians in that process. So we bucketed these into you can cut these a lot of different ways, but I think you can see the flow from kind of the least concern. Notice I don't say no concern, but the least concern down through some of the more, more intensive needs for clinical evidence and support. So things like business management. Help me write I sure use this. Help me write my description. Here's my outline of, you know, who I am as a practitioner and what I do. But, man, those final, like, glossing of the words can sometimes really get me hung up. So being able to use AI in those ways, people have found very useful ways to find supporting writing documentation like that that's not client interacting in some way. Access. So thinking about how AI can be involved in managing inquiries, how it can be used in things like screenings, up to clinical management, things like documentation that gets really important that you know a tool does it well, up through clinical monitoring and decision support. Once a tool is doing things like helping you determine homework, that's a very different type of tool than a tool that helped me think through the best way to word what my type of practice is. Right? Those are two very different levels. All the way to the most intensive probable need for research and support, which is client facing clinical tools. So anything that involves client reporting, anything that comes back, we're not talking in this category only about kind of chat based tools, but we're talking broadly about anything that interacts with your clients. That that hits a different level of oversight and, need for evidence within that use. And, sorry, one one little bit piece more before I pass over, which is a little bit about that evidence based review of data. So what I wanna mention in this, I as I said, when you think down that line of categories, I would say you can almost think about a very clear arrow that goes into the level of oversight that we need. The level of oversight that we need to know that a tool really does what it says it's doing increases in complexity and need down that list. So the closer you get to a client, the the closer the the more you need of all of those pieces, oversight, research, support. In the current state of science, what we know is there is a lot of research showing the role of AI in assisting, you know, determining risk level. There's been some really great research there. There's been some really great research in determining ways to reduce burdens on clinicians. We're still in limited science, in anything that touches that concept of this can do those things clinically that we do as clinicians, right, for many reasons. I won't go in-depth right now except to say that science is far earlier. And I think we all hold concerns that there are some things that it just can't do. What we do know is being able to look at things that even if we're using it for documentation support, thinking through what's the impact on alliance and on patient outcomes. So some early again, I'm using a little bit of those buckets that I had already shared with you. When we think about administrative support, this is where our strongest evidence is, that there are useful ways that clinicians find reducing burdens through administrative support, whether that's documentation, whether that's, you know, logging calls, keeping track of those things, but it reduces burdens that make it easier to do, the remainder of our work. Now what I will say is when I'm saying there's more evidence, there's more studies. We've been using this more in the field. So we can say that it reduces burdens. We can know that it reduces burnout. We can know that it reduces time. These aren't randomized controlled trials where we're saying this impacts the client's experience, but what we do know is that it impacts the clinician's experience at this level. This has happened a lot longer. This is used more consistently, and there's a whole lot of caveats to how we think about using it. When we're thinking about now I'm lumping a few categories across access and decision support and everything to think about these kind of broader tools that support things you do, but you still own as a clinician, that clinician centric space, which is that, there's a lot of research growing that shows there's some real hope in AI and machine learning in outperforming risk discrimination from more static scales that we've used. I described in my prior world in addiction, I'll say I also did a lot of forensic assessments. We use a whole lot of Right? So we use a whole lot of scales in trying to predict risk and predict, you know, whether it's violence, whether it's risk to self, whether it's addiction risk. But a lot of those measures are not able to individualize to the level at which they can predict, like AI can, when tested and formed correctly, the models, in that way. We also know that it's shown some really hopeful outcomes in things like crisis text lines and showing promise for not just flagging high risk, high risk conversations, but really flagging in ways that are like this is the clinically meaningful content. We can think about older ways of doing so where we can do word analysis and, like, someone mentions risk and we jump to a conclusion, those aren't very helpful as a clinician. Right? We all have clients who report some level of risk kind of at their baseline all the time. AI has gotten much better, and there's been a lot of studies to support that. So as we continue to see growth in this area, this is one in which it really aligns with prior research we have in the field to say, like, we have data models that can do better when they're combined with our clinical judgment. These are clinical. These are processes that are meant to be clinically integrated in with clinicians to support decisions, not take away decisions. And for client facing tools, there is less support, as I mentioned, but growing support. Some of these models are meant to be used alongside a clinician. But these results are early, and and and I'm not gonna go into detail in any of those specific studies except to say we have so much more to learn over what pieces can be done and when what can't be done. What I like to think of a parallel as is I used to, as a clinician, say, like, go buy this workbook we're gonna work through together. It may be a good parallel to something like that that we can test, But it is not in the same way of what psychotherapy sits as. So, again, spending limited time there. So there's a lot of support to say, you know, we need less clinical evidence for administrative based AI, and there's a lot of early findings saying, like, this works. There are less the farther that we go down the line despite the need for more intensive research. So we need to continue to grow here. We need to understand things like the black box and decisioning of how AI is getting to conclusions. We need to better understand biases, and understand where these tools work and don't work. And I mean that across the spectrum, even including administrative AI. But with that, now I really will pass over to Ali for ethical and legal considerations. Thanks so much, Lindsay. Appreciate it. So I guess I'll just start by by saying that, like, when we're thinking about ethical and legal guardrails, you know, the principles, with the exception of a few very specific laws, we'll talk about it at the end. But, like, the the principles that you see on a lot of these slides are ones that everyone in this virtual room is already familiar with and live by every day. You know, informed consent, HIPAA, beneficence, malfeasance. These aren't new concepts to any of you. It's just that, you know, AI doesn't AI gives us a new context to apply them in. Right? So instead of thinking about them as a very specific AI framework, you know, we're thinking about the the principles we got we already know within the context of AI. Right? So let's start with informed consent. You know, when when we think about this with with in general, right, but also in connection with AI, it's really more than a checkbox. Right? This is a conversation. It's not a form. Many industry associations including the APA have weighed in on this saying that, you know, mental health providers should inform clients about the benefits and risks associated with the use of AI. We'll talk about the laws around that in a little bit but, you know, this so this it allows your clients to make informed decision about, you know, working with you working with AI in your practice. So, you know, recommendation here is to update your consent forms, And simple practice, I'll just put a plug, we we have we have a, a template we've provided that includes, you know, the use of AI, and we'll continue to update that as things change. But also talk with your client about, you know, your use of AI in your practice in very plain language so that they can understand and and make an informed decision about around that. When we think about beneficence versus non malfeasance, you know, there there's a core tension there. AI can reduce your documentation burden by hours per week. Right? That's the benefits the benefits part of it. But, also, we know things like, you know, errors in the AI, risks around privacy, consent, bias, etcetera, are all real, and that's the malfeasance risk. Right? So the the study that Lindsay talked about and walked you through showing dropout, you know, the dropout of of the burn rate, you know, burnout rate going from something like I think it was, like, 52 to 39 or something like that with AI Scribes, that is time back, for you, which is generally good for you and your clients. Right? It's important that you also think about that in the context as long as you're using the right tool and feel good about the use of that tool. You and your client feel good about the use of that tool. So these principles, again, haven't changed. What has changed is the practical landscape that we're navigating. So let's get a little bit more specific about how to apply each of these. When we think about applying ethics to AI, confidentiality and privacy are things that you should and rightfully do have questions about. So before any AI tool touches your client data, it needs a permission slip. And I know for many of you, when you hear client data, one of the very first things that comes to mind is HIPAA. Right? So think about HIPAA like a bouncer at the door of your practice. Someone needs to show ID to to get access to that client data or that PHI, and that ID is a business associate agreement. So where that vendor commits to being subject to the same rules and restraints that you're subject to with respect to HIPAA as a covered entity. So I would say the practical application here is, you know, treat any tool that offers to provide you with AI efficiencies without a BAA with skepticism. And that's even if even if the vendor says they're not gonna touch PHI. It would just be very skeptical about, you know, not having a b business associate agreement or a BAA in place. The next concept here when we think about competence, it really means, you know, using the when we look back at what Lindsay was talking about with the various sort of, like, levels of tools that are out there, you know, it means sort of, like, staying, within the areas of your appropriate training and what you feel comfortable with. So it's a principle that really maps directly to that tiered model. So level one administrative tools, I'll call them level one, like the administrative tools, scheduling, note prep, etcetera. You know, maybe have a lower, what we might call a confidence bar in terms of how you feel comfortable using it and and what conversations you need to have with your client, etcetera, versus, you know, a higher level tool that is making maybe making, decision support for you, those types of things. So, really, the the important thing here is know which tool you're using and where it falls before you adopt it and what, you know, maybe additional education you need or your client needs or something like that before you adopt it. With respect to integrity and honesty, you know, I I would say think about here, like, don't oversell what AI can do, You know? Any tool that provide you know, any tool that you're using does not alleviate your responsibility to your clients. Right? So, you're gonna learn very quickly. I love analogies. So the AI tool here is your is your sous chef. You know? It preps the ingredients. It keeps the station organized, you know, but it is not designing the menu and is definitely not deciding what your client needs to or should order, right, when you're thinking about that sort of sous chef analogy. And then I would say finally, you know, justice and fair access deserve a real pause when we're thinking about ethical principles here. You should be thinking about things like whether the use of AI impacts your clients in any way, whether that's because they don't have the same access to technology, whether it appropriately serves a diverse client population that you may have. Things like that are are really important to be thinking about as you're adopting AI. So now let's, zoom out from ethics a little bit and talk about the legal rules that actually put teeth behind these ethical principles. So legal picture, I would say, breaks down into sort of three different areas. Privacy and data safety, liability and malpractice, and informed consent, which we've already talked a little bit about. You don't need a law degree to navigate these. I'll just say that. What you need is a good checklist and and good habits. So privacy and data safety, let's start there. I would say HIPAA is your baseline, not your ceiling. So going back to my bouncer analogy, you know, the bouncer is not just there to check IDs. Right? You know, the bouncer is also thinking about enforcing safe behavior, within your practice. So HIPAA HIPAA is the same. You know? It requires much more than a BAA. It, you know, can require things like data encryption, access controls, audit logs for electronic PHI. And on top of that, the health and human services, just last year proposed the biggest security rule update since 2013, which could also make changes going forward require there might be required things like encryption and multi factor authentication. Those might be mandatory across the board going forward. So, the direction of travel when we're thinking of privacy and data safety it is it's pretty clear, honestly. This the bar is going up, not down. And so what I would say to that is as you're thinking about implementing AI within your practice, pick vendors that are already there, and don't sort of just dismiss you with, like, just a BAA. Like, we've got a BAA. You're good. You know, you should be asking additional questions around their, you know, data practices and things like that. With respect to liability, I I'm sure you all have heard about hallucinations and AI, and this is where the AI output either, makes something up that is not there or doesn't get something exactly right. And, even if those hallucinations really only affect, say, like, 1% or less of the notes that are out there or the outputs that are out there, the way that, you know, putting on my lawyer hat, the way that I think about it is, you know, one bad note in a subpoena, if you happen to be involved in a case, can totally define your case. So the the phrase I want you guys or I hope that you will walk away with here is is you sign it, you own it. Right? So treat every, AI note generated less of or we're focused on notes right now, but cheap every AI output that's generated, like a letter that you asked your 13 year old middle schooler to write for you. Or you dictated it and they wrote it. Right? And I say that I can say that because I'm a 13 year old middle schooler, and I can only imagine, what what a letter would look like if I tried to dictate to him. But I guess I would say in other words, like, put on a a big, huge magnifying glass and read every word of that AI output because you sign it, you own it. Right? I'll just repeat that. So with that being said, sort of, like, where does the clinician's responsibility end and, like, you know, AI tool developers begin? Right now, the honest answer is that the clinician bears the burden. Courts haven't really sorted out AI developer liability. Of course, we are seeing, some things out there with respect to chat gbt and OpenAI around sort of some responsibility there, but it hasn't been fully flushed out yet in the therapy setting. And and that combine that with sort of the professional standard of care hasn't changed. Right? So your signature on a note makes it your clinical testimony. So, that's an important thing to to walk away from this with. With that said, let's take a look next at some of the, you know, federal and state rules that are shaping this right now. And, again, with the with the framework that, like, there are we are some seeing some emerging laws, but really a lot of the stuff is is things you know about already. And what that is are things like HIPAA, FDA, FTC, which is the Federal Trade Commission. And so those frameworks are already there but what we're seeing is is really like a lot of people believe that additional AI regulation is needed in healthcare. And and a federal framework is a difficult one There's a I don't mean to nerd out on sort of the the laws and the the constitutional law here. But, when you think about a federal framework versus state frameworks, you know, states are responsible for the safety of their, of their residents, and the federal government is responsible for, you know, overarching frameworks and things like that. However, I would say right now, you know, we we have HIPAA in place, we have FTC, Federal Trade Commission rules in place, which basically say that, like, AI claims can't be misleading or unsubstantiated. We've got, Lindsay mentioned, you know, AI powered devices and that's overseen by the FDA. And I will also say Health and Human Services has, published a request for information specifically on accelerating AI adoption in clinical care. So so we think more federal guidance is coming. But that being said, you know, the states aren't waiting. And and what we're seeing emerge here is sort of what I'll call this 50 state patchwork. Right? It's out there and it's moving fast and not every state is coming up with laws that are exactly the same. And so that's really critical to keep in mind as you're thinking about, if you have a multi state practice, you know, what might be allowed in one state might not be allowed in another with respect to the use of AI. And again general privacy rules, and general you know, delivery of care rules, licensing rules, those are not changing. Again, sort of back to that overall theme of, your use of AI is, you know, plugged into that pre existing legal and ethical framework. That being said, we are seeing a few states coming out with very specific laws related to the use of AI and and mental health care. For example, Illinois, prohibits AI from making independent therapeutic decisions. It is one of the few states that requires informed consent for the use of administrative AI. Nevada prohibits AI from providing services that constitute mental health care. Tennessee prohibits AI from representing itself as a qualified mental health care provision or professional. Texas requires disclosure and practitioner review of AI created records. And then California and New York, Florida, South Carolina all have, you know, other laws sort of per, like, over overarching, what you would call, like, a AI companion or chatbots and deliver delivery of mental health care directly from a chatbot to a client. But I would say, like, the common thread when you're thinking of thinking about how these laws are popping up among the states is that, some of the things we we see is, you know, disclose AI use, get consent, review everything, and don't let AI operate without your professional oversight. And I think these principles have very much aligned also with, you know, the principles that we at SimplePractice believe to be true is like human is therapy is human centric. Right? And so you need to be in the middle of that. So all of these rules have, you know, live within what I would call a broader context that we can't ignore, and that's equity. So when we think about equity and cultural considerations, there are a few things to think about. The first is, the digital divide, is is a real clinical reality. You know, not every client has reliable broadband, a smartphone, or comfort with technology. So the AI tools that you use in your practice, you know, you wanna make sure that they're not creating barriers, for some clients but not others because that can create an equity issue, wrapped up in a tech issue. So ask yourself, you know, does this tool work for my entire caseload or only the digitally comfortable person, or only the person with access to to, you know, reliable broadband or whatever it is. AI validation gaps, can map to mental health disparities and so many AI models we're seeing out there are trained on datasets that may underrepresent communities, specific communities, communities of color, non English speakers, LGBTQ plus populations. So when we say, you know, ensure AI tools are validated for diverse, populations, you know, that that's I I would say that's sort of a very practical thing you wanna be thinking about as you're as you're adopting tools. We've talked about an informed consent a lot, but this also has a cultural dimension. Right? So, you know, from an ethical standpoint, informed consent specifically calls for, you know, disclosing AI in the use of culturally and linguistic appropriate manner. And so consent form in an English only or loaded with technical jargon, or other things like that may not meet this bar. And so think about how you'd explain AI to your most cautious client in whatever language they're the most comfortable with and build your disclosure from there. Right? So just to wrap up, I would say, you know, you guys are already doing the hard part, which is showing up for your clients, navigating super complex clinical situations, and holding yourself to a really high professional standard. And the legal and ethical framework around, AI isn't here to stop you from from doing that or using these tools. It's here to make sure you use them in ways that protect the people who trust you. Right? And that's something, you know, everyone in this virtual room already knows how to do. So with that, I will pass it back to Lindsay. Awesome. Thank you, Ali. And just a couple more things we wanna mention to kinda bring together these pieces we've talked about, where the clinical sciences and where we still need to grow, where the ethical legal guardrails are to using AI in, again, clinician centric way and how you go about taking all of that information and thinking about how you apply this in your own practice. Whether you're already using AI in some ways or you're making plans to use AI in your practice in the future, a lot of it starts with being able to make the decisions that you believe will work for your practice, for your clients, for the way in which you believe, the future of your practice will be. And in fact, we did I I did not share to be able to have it uploaded, but can share a document in, like, what are some steps you can take as you work through these decisions? How do you think about what your goal actually is and how you align your AI decisions in matching to those goals? Why do you wanna use AI? Right? Like, right now in the world, there's so much pressure to integrate, but the truth of the matter is AI isn't great just for AI for AI's sake unless it actually matches the goal that you're trying to accomplish. So if you are looking for ways to reduce burdens in some of your documentation, that is a very valid way to explore what tools can do. If you are looking for ways to better track symptoms of your clients from session to session, you know, I I always it's like one of the things I'm sure I feel like a broken record with clients on is that I get to see you for one hour, but there's a 160 of the week that you're sitting and doing other things. So that's been a gap sometimes. I know that that I've felt. Is AI the answer? It depends on the tool at this point. Right? It depends on understanding how you're actually using these things, but you start with that overall goal. So how do we tie these pieces together? I would say, in this brief summary, two ways I wanna share in thinking this through. One is using from again, I'm I'm coming from a psychology background. And one of the things we talk a lot about in testing to understand clients' problems is how we think about the level of data and detail we're going into. So I'm gonna use a really, simple example to talk this through and then tell you how it applies to AI. So we talk about level one to level three types of data and when we're doing assessments sometimes. And level one is I'm not making any interpretation at all. So if a patient walks in and they look sad and down well, down would be going too far. If they are appearing tearful and they appear sluggish, and I denote in my record that this patient looks sad because I see x, y, and z. They're sluggish. That is, the least level of interpretation I can make. Cheerful, least amount of interpretation I can make. To go up to level two, I'm categorizing it. This person appears to be struggling with mood currently because I've witnessed these things. I'm pulling together a picture. I'm putting a label on what I'm seeing. And level three is I'm interpreting and making, guesses about where this person's trajectory is gonna go. So I'm taking that this person came into an appointment looking to your full level one interpretation. Level two, I interpret this as kind of ongoing mood and sadness concerns. Level three is saying, I have concerns that it may be difficult to engage this patient in behavioral activation, getting them to engage in sessions. I'm making interpretations from my clinical standpoint. What I would say is you can use the same frame for AI and how much oversight you need in the way that decisions are being made. So not everything is the same level. As Allie said, if it's under your signature and under your practice, it is yours. But it's still important to consider, like, what is that level of detail I need to go into before I even decide to use a tool in understanding what oversight I need? I would say the closer it is to something like taking your session data, we'll use notes are an easy example. Typically, that is, when you are taking, notes, it is taking what it sees and what it experiences and just putting it into a document. Unless that note is allowed to make interpretations, that's a pretty easy what you need oversight on is, like, is this what I would document? That is something that you can test along the way. You could review your notes. You can and should always review every note. But in determining whether it's a tool you should use or something that you, need more evidence for, it's kind of the lower level. When you think about all the way up to this is gonna make predictions or make guesses on a treatment trajectory, you need a lot of oversight, not just in the process, not just in your clinical practice, but before you even decide to use a tool. How do you understand how it's making those interpretations? This is where bias can come in heavily. Is it does the tool have an answer for how that product for how they're determining how they ensure bias does not get inserted into the process? So the higher you go into interpretation, the more you need to know before you even choose to use a tool, let alone how much oversight you give in the session. And finally, to kind of put a picture into how I think about some of these decisions about where you will and won't use AI in your practice, it is taking together each of these pieces. Right? There's regulations, laws, ethics that Allie described in detail. There's the clinical impact and evidence and what degree of evidence you need to support a tool based on what it's doing. But then there's your view and your patient's view. And somewhere in the middle is where things work, and it may not work. You may have tools that work perfectly for three of these bubbles, but a patient absolutely is not interested. And that is going to happen as we continue in the field. Right? There may be things that a patient comes in and says, I actually really want that support tool. And you might feel like the evidence is there, and you might feel like they do it in a safe and ethical way, but it doesn't work for you and how you see psychotherapy in the process. That is okay. It is always about a balance of where do tools meet across all of these needs. And I'd say on the outer edge of that is AI is developing quickly. That means even for things that you decide to use, they are probably changing over time as well. And having a process in your practice where you allow that kind of continual feedback loop to examine each of these pieces. Maybe a patient was really okay and excited about a tool that you intended to use, but something changes along their way because of a personal experience or something that happened with the tool, be being able to continuously revisit, I would say, part of what when I think about consent and the process with patients in our practices is consent's not a one time thing. I think we all know that. Right? We think about that in many ways. Consent isn't static, though. Just because someone put a signature on a paper or agreed and conversed with you and said, yes. Let's do it today, doesn't mean that three days from now, they feel the same. Things could have changed. And this is a continual process to ensure that you have a system to check back in, that patients don't have to feel that uncomfortable of, I know I said yes, but I don't feel yes now, just like you would on any other part of your practice, having that continual check-in. And one other piece is just ease of integration. That can't be understated. If we're working in 10 different systems, we can't be on top of what the science behind the tool is, what the, ethical guidelines they have in place, the agreements they have in place. The more that we can find tools that integrate into less systems, the more oversight we can have, which also means the better informed consent we can have with our patients because we understand how to answer the questions that they may have. With that, two things we are gonna switch to now for the ten remaining minutes we have, which is, one, I know we are gonna share a brief poll just to get some information back from you, about your experience, and that should show up for you. Also, we wanna switch over to q and a. So I saw some questions coming in, and we're gonna go ahead and share those on screen and try to answer as many of them as we can. Please don't feel like this is static. You could be in there right now, upvoting some of these questions, adding your questions. So, we will see them as they continue to come in and do our best to answer. But with that, I am gonna share our first question, which is from Paula saying, would you advise getting a cyber rider on professional liability insurance? So, Ali, do you wanna. I can take this one. That's a great question, and I would say the short answer is yes. But I'd also maybe go a little bit further and recommend getting a stand alone cyber liability possible a cyber liability policy, not just a rider. And the reason for that is, you know, your professional liability insurance, which is, you know, covers malpractice and it covers a claim that you made an error, you know, claims that you made an error in the delivery of your clinical services. But if your laptop gets stolen or you use a vendor, that gets hit with a breach or something like that, you accidentally, you know, disclose information, you know, it's unlikely that your malpractice policy is going to cover it even if you have a cyber liability, rider. So, the cyber liability insurance as a standalone policy is gonna cover all those types of things. So, without, you know, having the specific of each policy or what the rider says, I guess, like, my overall recommendation is to make sure that things like what I just listed as an example of, you know, stolen laptop, breaches are covered under what policy you get, whether it's a rider or a stand alone policy. Awesome. Thanks, Sally. So the next question is from Judith. Do we need to put a disclaimer for AI disclaimer for AI use at the end of every note? This was suggested on the recent NASW conference. So I'll take a start, and then, Ali, if there's more you wanna add also. What I would say is I see this. I'm gonna draw a parallel to our worlds. Again, Ali mentioned this in thinking about ethics and practice that there's a lot of pieces of the world that we can talk from how we practiced in the past and what we can draw parallels to. And what I would say is, you know, I myself have been in hospital and forensic settings where there were maybe there was an interpreter there with me. Maybe there was someone who assisted with documentation and took notes. And those were always things that I felt like were an important part of the care process to acknowledge. So though it is not, in most places explicitly required that you must state that, I would say that if we follow our best practice standards in that way of what is the safest of what tools did I use, how did I use them, it makes sense to be able to say, yes, AI was used, and noting your involvement in reviewing that documentation. Right? So as Allie already mentioned once, but I'll I'll also reiterate from the the clinician side as well. This this statement that you this agreement to yourself and your use of AI tools that you are reviewing, you are the owner still of whatever goes in that record. Again, to draw another parallel, when a supervisee writes a note and we sign it, right, we're taking some responsibility responsibility over those notes, especially when they're earlier in training. I see AI as very much the same. It is something that I am overseeing. It is not able to act without my review, and I would want to make sure that that is stated as well. Ellie, I don't know if there's more that you would add. And, no. You you said it so well, and I I totally agree. I I would say, you know, from a legal standpoint, this isn't legally required that you do this, nor necessarily. I mean, I don't wanna I'm not a clinician, so I don't wanna speak fully on ethics. But the the do I need to? I think the answer to that straight is no. Is it a best practice? Yes. Absolutely. And I think you could even add something, that says, you know, this note was prepared with the assistance of an AI note, AI scribe or an AI note taking summary tool. I have reviewed, the note and I sign off. You know, something as simple as that can be good can be really helpful. Awesome. Thank you. Thank you for the question, Judith. I think that's a great one. And one that we don't always think about as we're in that rush. Right? We're trying to documentation is a huge burden. But having a standard boilerplate of that piece, right, that that is one of those statements that we can say again and again, without concern, if you have it prepped in advance. So next question from Jennifer. I was using another platform and their AI reached out pretending to be me. I have left that platform, but what is the legal responsibility on us as a provider? And even on reflecting on this question, I just wanna say, I mean, that sounds incredibly stressful. Thank you for sharing your experience, with us and everyone here. I know we can't ask you more questions to understand this in in that way, but what I would say is, I would have a lot of con concerns about that price. Sounds like you you left that that platform pretty quickly. But, Ali, I don't know if you, from a legal side, have, perspectives on Yeah. this. I would say, again, without having, like, more details and understanding the platform and, like, the terms of service of the platform and what specifically happened, I mean, I think there are a couple things at play here. You know, one, you know, as technology evolves, you know, mistakes happen, and we we realize that. And I would say the law realizes that also. And so oftentimes, if something like this happened and you were fate, you know, someone let's just say someone sued you, and you're like, what's my liability? A whole investigation would go into, like, what led to this? You know? Was it a technology error? Was it did someone hack the platform and send out emails as you or whatever? And oftentimes, you know, you you through that investigation process, you're gonna find that, like, you're the liability where that falls. Right? And so it's hard to say exactly without the details there. But what I would also add is, you know, in situations like this as Lindsay mentioned where it's an incredibly stressful and concerning as a provider, you know, it's not necessarily a bad idea to reach out to anyone who may have been the recipient of those emails and or or I don't assuming it's emails. I I I shouldn't assume that. I just said reach out. Whoever was the recipient of the of those reach outs, you know, you you can also follow-up with them and say, like, this was sent an error, apologies. And oftentimes, what I found in my practice, is that, you know, when you can help, sort of calm down concerns and help people feel okay about the situation, what happened, you're less likely to end up in a lawsuit at the end of the day. Right? Thank you. And I know we are almost up on time, and I know I said I wouldn't pull up necessarily simple practice specific ones. But I do see some upvotes, and I think this is an easy answer. So I wanna make sure to to, address this question as well, which is, can notetaker see the video of a telehealth call? At times, notetaker post session will say client is dressed casually and comfortably curious about this. So notetaker can it is not a video assessment that is based on on transcript where notetaker is building the note. So if there are things like that noted, it is likely from something within a transcript that is being built or something being pulled, you know, as as as explained, and we can't go into enough detail right now, but happy to follow-up, but uses your prior notes, in kind of predicting some of those pieces. I would guess that it it comes from those terms, but it cannot see or visually capture pieces of your session. Ali, pieces I left off in that response. No. That's great. Thanks. Awesome. Well, time goes so very quick, and we so appreciate all of you being with us here today. We do wanna make sure, you know, again, you can download the slides. If you have questions, the recording will be available, and shared with you as well. And, also, I will make sure to share like I said, we have a decision making, kind of guidance of, like, how do you plan for and develop your own personal plans for implementing AI in your practice and evaluating ethics and research and all of those pieces so we can also make sure that that is shared with you. But, really, thank you for spending some of your Friday here, with us and really enjoyed getting to see your questions come up. And, I love seeing the chat when I can keep up in how, wide and far people are coming to join us today. So thank you. Thank you all so much. Appreciate your time.