VRTAC Manager Minute: Stay Curious, Stay in Charge: AI and Ethics in VR

Artificial Intelligence is everywhere—but where does it actually fit in vocational rehabilitation?

In this episode, Carol Pankow sits down with Dr. Robert Froehlich to cut through the noise and get real about AI in VR practice.

This isn’t hype—and it’s not fear. It’s practical, grounded guidance.

🔹 What AI should be doing (and what it absolutely shouldn’t)
🔹 The ethical guardrails every VR professional needs to know
🔹 Why AI can boost efficiency—but can’t replace judgment
🔹 Real risks: bias, hallucinations, and “too helpful” answers
🔹 How to stay compliant, confident, and in control

The takeaway?
AI can be a powerful assistant—but it should never be the decision-maker.

Because when it comes to careers, independence, and life direction…
that’s human territory.

Listen Here

Full Transcript:

{Music}

Rob:

Ai, Artificial Intelligence, is not and should not replace the professional.

 

She's listening to this podcast, and the professional said, don't be afraid that AI is going to take your job. However, do be concerned that your job could be taken by someone who knows how to use AI better than you.

CRCC indicates that some of the potential benefits for AI could be increased counselor capacity, enhanced job exploration, skills credential matching, things like that.

Intro voice: Manager minute, brought to you by the Vocational Rehabilitation Technical Assistance Center. Conversations powered by VR. One manager at a time, one minute at a time. Here is your host, Carol Pankow.

Carol: Welcome to the Manager Minute. Joining me in the studio today is Dr. Rob Froehlich from George Washington University, who's been thinking deeply about the ethical dimensions of AI and rehabilitation counseling. Rob’s a colleague of mine and also serves on the VRTAC grant. So welcome Rob, how are you?

Rob: I am doing well and thank you for having me today. And I have been thinking deeply. Let's hope that we get something worthwhile from all that deep thought.

Carol: Oh, I'm sure we will. So we've talked on previous episodes about how vocational rehabilitation agencies are beginning to experiment with artificial intelligence. You know, we've got everything from transcription tools to drafting reports to analyzing labor market information. But today we want to take the conversation in a slightly different direction. Instead of focusing on what AI can do, we want to talk about what it should do and what it should not do. When you're working in a profession built on ethics, informed choice, and professional judgment. So let's dig in. So, Rob, before we jump into AI, could you share a little about your path into vocational rehabilitation and what led you to focus on ethics and technology in this field.

Rob: Sure. Absolutely. Thanks, Carol. So I've been providing technical assistance and continuing education at GW in collaboration with state VR agencies for more than 20 years. I'm an associate professor of counseling, a certified rehabilitation counselor, and a licensed professional counselor. Most of my research has been focused around ethics, because I've always found that to be a really fascinating sector. There's some topics in our profession that are right or wrong and are dichotomous in nature, but really the vast majority of topics we deal with fall somewhere in between. In the past, I've been the chair of the CRCC Ethics Committee, and in that capacity, I had the ability to sharpen my ethical decision making skills. But to your question, why ethics in AI. I feel like we're at an inflection point, and many people are trying to figure out where AI fits, what it is, what it can be used for. And rehabilitation counselors are but one population asking the question of how will AI impact my work? To put this inflection point in this moment in time that we're at into context, think about some other big inflection points the industrial revolution, the movement from horses to cars, or more recently, things like the introduction of social media or even the pandemic. Absolutely. Every one of them changed the way that we do things in society and in our world. So they're all big changes. But even this AI moment that we're in, we can navigate through this.

Carol: I think so too. I like that you brought up all those other big inflection points because I thought about back when I was at State Services for the Blind and we were emailing parents, and parents were going, hey, could you just text us like, we don't, we don't really check our email anymore. And it was this huge to do. Yeah. Can we let counselors text people instead of emailing? And oh my gosh, we agonized for months and wrote up policy and all the pieces. But you know what? It all ended up working out just fine.

Rob: Yeah. So Carol, I'm reminded when I talked about social media, Oh, forever and a day, I was doing ethics and social media training. And when we started out, the whole message and it kind of relates to what you're saying, the whole message was, and I'm going to date myself here, but like a kind of like a Nancy Reagan just say no. Which ,back in the 80s, there was this big program of that's how to deal with addictions. Just say no. Now, we all know it's much more complicated than just say no. But as that message evolved, so did our message with social media and really like the Just say no was like, oh, no, never. Like don't friend, don't do, don't, don't use anything. Yeah. We found effective uses for it that still are in compliance with policy that still relate to legislation. So I think we're going to probably start like probably a little tighter with AI and then things that are going to naturally evolve over time as they tend to with big revolutions like this.

Carol: Yeah, that's a really good point, Rob, because it does. Everybody starts out having, you're kind of in a freak out mode. You're like, no, nope, we're not doing that. Oh, we're going to just a little bit in there. And now you're like, I can't even believe we had a fit about that. Like, it just seems such a part of every day that everybody texts, you know, like stuff like that. You just...

Rob: Exactly.

Carol: It becomes normal. So a lot of professions are talking about AI right now, but vocational rehabilitation involves helping people make major life decisions about careers, education and independence. So what makes the ethical conversation around AI particularly important in rehabilitation counseling?

Rob: Yeah. So the first thing that comes to mind for me that I want to just stress for people is that AI, artificial intelligence, is not and should not replace the professional. It's really important for us to remember that AI, artificial intelligence, is a set of tools and things that are supposed to be helpful to us, but they're not supposed to replace the functions that we play. One of the essential functions that I'm speaking about, the essential functions of being a counselor or a rehabilitation counselor, is the use of critical thinking as it applies to decision making for people that we collaborate with and provide services to. One of the biggest reasons that we're paid for professional function is to use our critical judgment. We don't want to abdicate that to another entity. We want to make sure that we're using AI tools as just that tools to make things more efficient where possible. And we want to make sure to protect confidentiality, like almost above all, to be able to have oversight regarding decisions that are made, but at the same time. So we want to get that efficiency. We want to kind of have oversight, but we want to make sure that we don't delegate tasks or decisions to AI that go back to those life decisions about career, education and independence. Information is one thing, but decisions is an entirely different set of principles.

Carol: That is such a good point because when you look at when you put anything into, whether you're using something like ChatGPT or Notebook LM or Google Gemini, you know, or even Microsoft Copilot, you've got all these different types of AI tools that can sit alongside and help you do something, but it's not like you can just throw in a question and it's going to perfectly answer it. This is it. Here's the beautiful plan for the customer. And somehow they know them and all the nuances about that individual, that just isn't the case because you would have to feed it so much stuff. And it's not the counselor. It doesn't have, you know, it does have, I don't know if you call it thinking ability or how it does its reasoning or brings up, It's, you know...

Rob: Well it's math. It's just basic math. It's yes, no, yes. No.

Carol: Yeah. It's not the human element and all the other parts of it. So you can't just like abdicate. I still remember last year we were somewhere in one of the states, and the hotel was trying out robots that were bringing things to your room, like, say you forgot your razor and you called down to the front desk instead of a person running up, this robot showed up. But I don't think we're going to have robots taking over the hotel and making the beds and doing all the things I think people envision. This is like Terminator, you know, when you had all the little whatever they were looked like people and then they're back into their little metal bodies. I don't see that. I mean, just not yet.

Rob: Yeah.

Caro: So rehab counselors always emphasize things like informed choice, confidentiality and professional judgment. How do these values shape how we should think about AI in practice?

Rob: Yeah. So those are really high level and important topics for us to discuss. And as a part of informed choice and everything that we do in vocational rehabilitation is about informed choice and is centered around informed choice, but as a part of informed choice, we want to make sure that we have the capacity to understand and to be able to explain what tools we're using, why we're using them, and what the intended outcome of that tool is. So as counselors or professionals who are using AI, we want to be able to answer when people have questions, why did you do that and how does that work? We want to have that capacity to be able to explain that. We want to make sure that clients, first and foremost, consent to us using AI in our work with them. It's their choice to be able to decide, do they want that to be a part of the services that they're receiving? We want to know where any information that we enter into a tool, for lack of a better term, go. Where does it go? Does it go to a cloud? Does it go to training AI? Does it go to other people? We want to know where that information we put it goes and what might be threats relative to personally identifiable information or protected health information.

Anytime there's a threat, we want to avoid using the tool. So a big distinction, there's a difference between a closed system of AI and an open source tool or a public facing tool. You just mentioned a number of publicly facing tools a moment ago. Closed systems keep information within an organization's own sphere, if you will, and calculations and decisions don't go beyond the walls, if you will, of that sphere of that organization. Whereas an open source system like ChatGPT, for instance, will include the data entered into the systems for their own training purposes of the model. And they don't protect confidentiality. And they also have the potential to introduce bias based upon what has been entered into the system. They're really it's more like a quantifiable thing. The more information that goes in, the more it's going to likely lean to that. But what if that information is wrong? What if it's inaccurate? It's still going to rely on that because it has a depth of information to draw from, if that makes sense.

Carol: Absolutely. I mean, you have to trust but verify. There's stuff that goes in. You can't just go, oh, this crazy thing came out. It's all good to go like, uh, uh, because the information is only as good as what it has been kind of fed before. And if it starts putting all this other, your math reasoning, you know, on to it, it can come up with some really weird stuff.

Rob: And I want to make sure to say I'm not the AI dream killer. I think it's awesome. I think it's wonderful. I think it can do really great things. But as of all tools, like it's good for some things and we're going to talk about that and it's not intended for some other purposes. And we don't, we want to stay out of that neighborhood. You know what I mean?

Carol: Yep. I'm really curious. You know, like I think it was notebook LM, so George Washington University has a listing of different AI tools that the, you know, the campus has said, yeah, we're good to go with these. We have a closed system with some of those. So they've been able to somehow, I guess, work with the company so that our application doesn't go anywhere else. It doesn't feed the model to like train it. The information stays there. So I thought that was super interesting that they did that.

Rob: Yeah. You know, and I do think that other organizations and agencies and entities are going to be learning about that and how that fits for them and are going to make some decisions. But it's important, as the user of the system, to understand that distinction between just all out there, something that's more of a closed system.

Carol: Yeah, really good point, Rob. So the Commission on Rehabilitation Counselor Certification has started issuing guidance around AI and ethics. What are some of the key ethical considerations counselors should be aware of right now?

Rob: I'm glad you brought that up. The document that you're referring to is a document entitled Frequently Asked Questions and Guiding Statements to Support Certified Rehabilitation Counselors using Artificial Intelligence. You can find that on the CRCC website CRCCertification.com. But let me kind of give you a little high level overview. The first thing that is recommended is that professionals using AI learn about what AI is AI literacy, right? So distinctions between things like large language models, generative AI, machine learning, deep learning, natural language processing, and agentic AI. It's not the topic of our conversation today to go down a rabbit hole on all of those topics, but know if you go to this document, they're going to give you a really nice description of what are all of those things. And it's important because, you know, we hear AI and we're like, oh my God, like it only came out when ChatGPT. That's really not true. Ai has been around for ages. Like a simple example is spell check, right? Spell check really has an artificial intelligence piece of it. And how many people, let's be honest, how many people can really spell anymore now that we've become reliant on that little squiggly line or whatever underneath. So like I said before, I'm not kind of trying to be the dream killer, but like, let's put these things in context and let's use things that make sense and things that don't. We want to avoid that.

So other recommendations protect client confidentiality. Don't put PII or PHI into a public facing tool. Only use a HIPAA compliant system that your agency has thoroughly vetted and has a BAA. What's a BAA? Well, I'll tell you, it's a Business Associate Agreement. Right? And so let me give you an example. So I also provide counseling in a private practice, right? We do a lot of virtual telehealth counseling. We use a Zoom based product, but it's not Zoom. Just regular zoom would not be protected and would not be encrypted. We use something called Secure Video, which uses Zoom as the platform, but it's an encrypted system, so it's safe. So it's the same deal with AI. You want to use the encrypted, more secure version that your entity will have an agreement with. Couple of other things. You want to uphold the principle of informed consent. Clients get to choose if they're okay with you using AI. You want to think about how you can use AI tools with clients or consumers or participants in ways that make sense. It lends itself to some of the things you said before, Carol, right? Job market analysis, document drafting, skill credential matching, reviewing large groupings of information from things like policies and procedures or legislation.

Rob: And those large groupings of information can then be broken down into manageable chunks of information. CRCC indicates that some of the potential benefits for AI could be increased counselor capacity, enhanced job exploration, objective analysis of situations with that skills, credential matching, things like that. Clinical documentation, enclosed AI platforms. There are entities that will allow for note taking of meetings with clients or sessions. Now first off, you're going to need to have consent on that. And second, you're going to need to learn how the system works, right? We use something called therapy notes and practice. And it gives you multiple levels. Like it'll literally be there and listen to a session and then help with a note afterwards. Now I have to say personally, super uncomfortable with that. Even if I have consent from clients, I'm not likely to use that. But is that generational? It could very well be. I've heard good arguments either way, but the very first starting point is make sure anytime anything is recorded that your client gives consent to that. It's also going to have the potential to impact what clients will share with you if they know it's being recorded. So just a thought there are a couple of other things. You want to help clients and counselors, yourselves and your clients. You want to help them to understand things like foundational knowledge, understanding AI, critical evaluation.

Okay, so here's a really important one looking for based upon your professional background stuff that doesn't make sense. It might generate a response. And based upon your education and your training, you might be like, okay, that makes sense. And that makes sense. But that thing in the middle, oh, that doesn't make sense at all. That's really important in terms of responsible AI use, because you can receive things referred to as AI hallucinations, right? If it doesn't know an answer, instead of saying, I don't know. It's going to make something up. So it's really important to think about that and to respond accordingly.

Carol, I feel like you had an example of something like that that happened recently that you and I chatted about. So you were like, okay, I put this policies and procedures manual or some sort of legislation in, and you were like, I need you to help kind of summarize real quickly. It came back and like a good portion of it was really helpful to you. But in the middle, you were like, that is just absolutely wrong. And so you put back in, you were like, blah, blah, blah is wrong. And it actually almost did sort of like a tongue in cheek like, oh, you got me.

Carol: Yes, yes.

Rob: Sorry about that. Right? Like trying to respond. Right.

Carol: It does that a lot. It's really funny because we work a lot with policies and procedures. And so when we're trying to like synthesize or make something plain language, it can be super helpful for that because sometimes people write really verbosely and you're just going, oh my gosh, nobody's going to read, t's too dense. So you get it to kind of sort it out and then it'll just take some random wacko piece, throw it in the middle, and you're like, this has nothing to do with what we have at hand. And I always do that, I'll type back in, where'd you get that? That wasn't even that didn't make any sense. That was dumb. Sometimes I say like, I'll be like, what are you doing that's bad? And it'll be like, oh yeah, you feel like you're talking to a person then because it is like, you caught me. Yeah, you caught me. Good catch. Oh, good grief, ya think?

Carol: Yeah, thanks.

Rob: But there also is like the environmental impact of AI that is a consideration. And  so there is this consideration that kind of conversing with AI is impacting resources and what have you. So I just put that out there just for a consideration for everyone.

One of the other really important things about learning literacy is about prompt engineering. Learning how to give clear ethical instructions to AI, tools to elicit useful, unbiased information, and to be compliant with policies. So can I start at like a really basic, like personal example that's super simple about AI prompting?

Carol: Sure.

Rob: Okay. Prepare yourselves. This is not clinical in any way, shape or form, but in terms of prompt engineering, a recent example of that occurred for me. So, I'm a runner. I like to run. I like to listen to music while running. It takes a lot of energy to kind of find the right playlist. And I have like most people, I have very diverse., I'm as likely to listen to Taylor Swift and Lady Gaga and Beyonce as I am to the hair bands of the 80s, as I am to disco, as I am to country music, whatever, right? So there's something on a streaming platform that I like. It's called the AI DJ. Okay, all the AI DJ is, it's like a first level. It goes into what you've listened to. It generates four songs you've already listened to, and then one that maybe is like them. It's a real basic level. I got super excited a couple of weeks ago. There's a whole new line on this streaming platform that says, for instance, one of the prompts kind of has some pre-loaded prompts. I want songs that are similar to what I listened to, but with a demographic ten years younger than me. Who doesn't want to be ten years younger, right? So I'm like, cool, okay, awesome. So I generate songs that I haven't heard before, artists that are related to the ones that I like. And I think that was like kind of the majority of them, right? So, all right, awesome. Generates information. And so I start running and it is like all of this like slow country music. And I'm like, all right, like on another day. Maybe the prompt that was missing there was the beats per minute. Like it, it wasn't there, right? So you're going to get something and it's going to be all the things that you think, but you might have forgotten to ask for one important question that is going to drive. The whole process has nothing to do with rehabilitation counseling, but I hope that that kind of demonstrates it's really important how you ask.

Carol: Yeah, that's a great point. Very true. Really good example. So when people are talking about AI, Rob, it often swings between two extremes. Kind of like in the beginning, you were like, just say no, Nancy Reagan, we should never use it at all. And others think AI should do most of the work for us. So where do you see the responsible middle ground for the VR professionals?

Rob: Let me just start with like a real quick, there's a book on change management. Our Iceberg is Melting by Rathgeber and Kotter. Okay, so the real premise is simple. There's this group of penguins that are living on an iceberg, and one of them realizes there's a crack and the iceberg is melting. The whole story is about how they manage and navigate during this time of change. But one character has relevance to a story I want to tell you. One character is this gruff and grumpy older fella, and his name is NoNo, and he's named NoNo because his response to any kind of change or any kind of anything is like, no, absolutely not, not now, not ever. So I had sort of like a NoNo moment last year. So I was on this trip was this active hiking, biking, kayaking thing. My wife and I were on this trip, small group of people, six people in total. Everybody was a medical professional, physical therapist, a doctor, a disability management exec, me a counselor. Anyway, everybody was somewhere in there, right? Everyone other than me came down on the AI is going to make work so much more productive. It's going to transform things and sectors of the employment will be changed, evolved or eliminated forever. And I was like, no, not now, not ever. I feel like I've kind of evolved a little bit since reflected and begun thinking, hey, what can we take that's positive and how can we safeguard against the stuff that is not really good? So back to your question,

like all the time, never, I say at this point, don't be frightened of it. Be curious, but understand that guardrails have to be in place to secure data confidentiality and informed consent, and realize it's not for everything. We have all seen the extreme examples of AI giving absolutely awful suggestions for people having mental health crises. So don't use it as a therapist. And we're going to talk in just a bit about some limitations inherent to AI that would relate to it, giving ultimately awful information in a mental health crisis. With that said, do realize it can give credible information about role playing job interviews. It can talk about labor market segments. It can synthesize vast amounts of information. As long as you superimpose your expertise on it and make sure it's not making stuff up, right? So be curious, be cautious. One other piece of information that I find personally helpful is I'm navigating this change is something one of our associate deans at GW shared in a meeting. She said she was listening to this podcast and the professional said, don't be afraid that AI is going to take your job. However, do be concerned that your job could be taken by someone who knows how to use AI better than you.

Carol: You know, that is funny you say that because I've been seeing more and more. There's just lots of articles, especially on LinkedIn. Like you need to understand how AI can help you in the workplace. Like you can't just let it go. And so there's lots of little ten minute videos or little clips and things to help get you a little bit more acclimated instead of being like, oh, AI is out there and it's this thing and you're in the ether and I'm scared of it. And I don't know really what it means. And you think of the Terminator, you know, instead. So that's a fair point. So counselors are trained professionals. They bring judgment, experience, human understanding to their work. How should counselors think about the boundary between AI assistants and their professional judgment?

Rob: Yeah. So again, AI should be considered as a set of tools, not some vast entity that can play all the roles and functions that counselors play. So to your Terminator reference to, I always want to be considering myself and I want AI to know, and I know that that's a silly term because it doesn't really know. Like it's an, it's not an entity, but I want AI, the tools that I'm using to know I'm in control. And I always want to have the opportunity to check behind AI for the little fibs that might make up. So the first deals with sycophancy, right? And so what is sycophancy? It is the tendency that AI may want to please the user by reinforcing existing biases rather than offering clinical challenges. So it's going to tell you what you want to hear sometimes. So you've got to be aware and factor in, maybe it's trying to tell me what I want to hear. When you look at people getting information about things like couples counseling or relationships, a lot of times it'll come down on the giving of information that is like, oh, you're right, you're totally right. That's how it is. Something that you really want to be thinking about in terms of balancing from a multicultural perspective. AI can suffer from algorithmic bias, right? And what does that mean? AI systems are trained on historical data, which often reflects and perpetuates social and structural biases. So therefore, what it generates may lead to unfair or inequitable recommendations that limit opportunity for marginalized groups. And that's not a really small consideration to think about. Also, remember, I was just saying before about mental health crises and why this would be a challenge. AI is based on math. AI doesn’t have feelings. In short, AI can lack empathy and clinical context. So as I mentioned before, these shortcomings from above, it's trying to please you may be problematic in terms of suggestions from a multicultural perspective, and the fact that it doesn't have empathy. All present significant challenges regarding clinical recommendations from AI. And that's not me hating on it. It's just being objective.

Carol: And now I learned a new word of the day that I'm not going to be able to say, but I can spell it. I think it's s y c o p h a n c y. Everybody can go look that up. Your special word of the day.

Rob: And Google's wonderful. It'll give you also the opportunity. It'll pronounce it for you. And I've heard it like five times and it's gone in and out.

Carol: That's cool. So are there decisions in VR practice that should never be delegated to AI tools? Like what would be the hard stop for you?

Rob: Yeah. So these are my opinions on the hard stuff. Eligibility determination and the absence of any other data eligibility determination in the absence of me going behind and making sure that what is suggested makes sense. Fiscal spending or denial in the absence of any other data. Again, you want to be able to use your critical thinking. Counseling and guidance without any oversight. For all of the above mentioned reasons, we just talked about documentation without any check behind having it just do your notes without ever looking at them. You don't know what is in there or not in there. That should be in there. So those would be kind of the hard stop areas off the top of my head for me.

Carol: Because it can hallucinate. So you have those notes.

Rob: Or it can transfer who said what.

Carol: Yes, absolutely. Oh, that's really good advice. So if a VR counselor or agency leader is listening right now and wondering how to approach AI responsibly, what's one piece of advice you'd give them right now?

Rob: I would say learn what you can regarding AI basics. Think about the legal implications in your state and how implementing AI usage is impacted by those implications and that context of your state. In the words of the American Counseling Association. AI guidance stay open, informed and cautious. Also, like I mentioned before, realize more rudimentary forms of AI have been in place for years. They just haven't been so publicly known or prominent, right? So we've been kind of living with it for a while and it hasn't bothered us. But now that it's taken on a broader perspective, now is that inflection point. So it's not like this is all brand new. So just think about your own reactions. Carol, that was such a great example before about like texting, like, oh my God, can we text people? Oh my gosh, can we text people? We're sort of at that point, like, what can we use this for? What makes sense? Technology can be super helpful, but let's just be careful out there in terms of how we use it.

Carol: Very good. I kind of remember that you did a webinar I believe recorded and probably for CRC credits on AI and ethics. Did you not?

Rob: Yeah, I'm sure that CITVR, I'm sure there's multiple.

Carol: Yeah.

Rob: And actually rehab symposium that Dr. Nicole Tichy, who had a prominent role in this document from CRCC. She directed the subcommittee that generated the document. But Dr. Tichy and I just a couple of weeks ago, and it should be on CITVR pretty soon. Did a presentation on ethics in AI.

Carol: Oh, excellent. So, Rob, before we wrap up, what are like three quick questions VR professionals can ask themselves when thinking about AI in their work?

Rob: Yeah. First one might be does the tool that I'm using protect client confidentiality? Another might be sort of a self-check on myself. Am I relying on AI instead of my professional judgment? And then maybe a last one might be based upon what information is generated from AI. Have I verified the information before actually putting it in place or using it? Those are three quick things that kind of come to mind for me.

Carol: Excellent. AI is likely to become, you know, it's going to be the future part of our work and invoke rehab as well. So the key is going to be making sure technology supports ethical, thoughtful counseling practice rather than replacing it.

Thanks, Rob. Glad you were here today. Appreciate you.

Rob: Absolutely glad to be here. Take care.

{Music}

Outro Voice: Conversations powered by VR, one manager at a time. One minute at a time. Brought to you by the VRTAC. Catch all of our podcast episodes by subscribing on Apple Podcasts, Google Podcasts, or wherever you listen to podcasts. Thanks for listening.