AI and Clinical Social Work

Artificial intelligence offers new and expanding tools, but also risks

By Sue Coyle, MSW

Editor's Note: This is part one of a three-part series on AI and Social Work. Look for part two in the Fall issue, out in August.

Over the past several years, artificial intelligence (AI), specifically generative and conversational AI, has become increasingly available to and known by the public. ChatGPT, for example, opened to the public in November 2022, adding a subscription model in February 2023. Today, ChatGPT estimates (it reported being unable to give an exact number) that more than 100 million unique users access it daily—and it is just one of the many platforms available.

It is no surprise that with so many users, some concerning trends have started to appear. One thing social workers and other mental health providers have taken particular notice of is the use of generative AI as a replacement to a therapist. NPR and other reputable news agencies have highlighted individuals who are turning to ChatGPT for therapy—subscribing so as to avoid usage limitations. And a quick search on Reddit shows multiple threads created by people using or interested in using AI this way.

In response, mental health professionals and lawmakers are working to put up guardrails around what AI can do for/offer to users, ensuring that those in need of therapeutic support receive human-first care. Such legislation is important and needed in an ever-evolving and expanding technological world.

However, that is not to say that AI does not have a place in clinical social work. It is not a therapist, but it can be a tool to help therapists, including social workers, as they provide services to those in need. For that to effectively happen, social workers must understand what AI tools are available, what they can do and what they should not do, as well as the risks and challenges that accompany integrating AI—in whatever form—into their practice.

AI Tools and Uses

Clinical social workers have access to many types of artificial intelligence, with programs and systems that run the gamut from virtual assistants to documentation tools and more. It is up to the social worker to determine what tool best suits them and their needs.

For example, a social worker may employ a form of AI that allows them to dictate notes, saving them time from inputting it manually, particularly if they have difficulty typing or other accessibility concerns.

In fact, increased efficiency is, quite likely, the most appealing part of AI. “One of the biggest benefits is saving time,” says Marina Badillo-Diaz, DSW, LCSW, founder of The AI Social Worker and adjunct professor at New York University, Columbia University and Louisiana State University. “But I put a caveat with that, because it can’t save you all the time,” she cautions. “You still have to read over the responses; you still have to edit; you still have to re-prompt it if you’re not getting the responses you’re necessarily looking for.”

Even some time saved is better than no time saved, and social workers can turn to AI to help with dictating or transcribing notes, writing assessments and writing letters or other communications, provided they are responsibly reviewing whatever is created.

“(AI) can be helpful in creating a treatment plan or finding resources for someone that you can print out at the end of the meeting,” adds Amanda Baranski, DSW, LCSW, LMNIT, lecturer at the Department of Social Work in the College of Health and Human Sciences at Metropolitan State University of Denver.

For social workers who choose to use AI to dictate or transcribe notes or even record sessions, they may also access programs that suggest diagnoses and treatment plans. Many of these tools are built into available HIPAA-compliant electronic health record platforms.

AI also can be useful in clinical learning. Baranski has used conversational AI to create interactive lessons for her students, as has Joan Groessl, PhD, professor and director of Social Work Professional Programs at the University of Wisconsin Green Bay. “In two of my classes now, I require students to interact with AI. So one of my assignments is I have those in my psychopathology class —so they’re trying to learn to diagnose—they interview AI. I had AI create a case study,” she explains, adding that the students are completing the assignment on the university’s secure platform and the case study is fictional. “They are the social worker interviewing the client to [determine] a diagnosis.”

The second assignment also involves an interview. However, in this case, Groessl is looking to see how her students draw out emotions in a client rather than a specific diagnosis. While all of these possibilities seem beneficial for a clinical social worker, none should be adopted—particularly when those tools might have access to personal client information—without a full understanding of the risks and ethics surrounding their use.

Confidentiality and HIPAA Compliance

One of the primary concerns when implementing an AI tool is confidentiality. How will the data that is entered or recorded be stored and used in the future? Can the platform guarantee secure privacy, in compliance with HIPAA, FERPA (Family Educational Rights and Privacy Act) and other state and federal regulations? A number of AI products market themselves as HIPAA-compliant, for example. Social workers should make sure they do their own research, confirming the veracity of such statements.

It is better, cautions Baranski, to be more careful than not.

“We have to assume that nothing is confidential,” she says. “We cannot trust any kind of technology. We shouldn’t. We need to be really careful.”

She adds that thinking this way is even more important when a social worker is inputting information into something aside from an electronic health record or other HIPAA-compliant tool. “If you’re putting direct information into ChatGPT about someone and it has their name and their diagnosis and then they ask ChatGPT about themselves, the information you put in might come up.”

Even with privacy and other settings enacted, inputting identifying information into a public AI tool is a risk not worth taking.

Badillo-Diaz agrees. “If you’re going to choose to use a free tool, you have to recognize that it doesn’t come with any HIPAA compliance. If you’re working in schools, it doesn’t come with any FERPA compliance. All of your data that goes into these free tools are being tested by the company and potentially shared with third parties.

“Our profession cares deeply about confidentiality, [and we are] certainly bound by HIPAA if you’re in clinical practice. So if you’re going to choose to use any free tool, you cannot put in any identifiable information.”

Even when using a purchased, HIPAA-compliant tool, Badillo-Diaz advises, “let’s just err on the side of caution.”

Roberta Bronecki, LCSW, founder of Serendipity Studio in Lyndon Station, Wis., says social workers also should be mindful of AI with features such as automatic deletion. “They (the software manufacturers) need to assure me that it’s going to be completely confidential. Like, where does this (recording) go? Supposedly, it records it and then it goes away? For years and years and years, decades, we’ve been told everything you put out there stays out there. Nothing actually goes away. Which is it?”

The lesson from all is ultimately to do the homework, which, says Bronecki, shouldn’t be new to social workers. Email, text messaging, electronic health records, she says, have all been introduced over time. “We’ve questioned all of them all along the way, and this is another one where I’m open because I’ve been open to all of these things. Most of them have ended up being good and helpful to the profession, but I don’t think I would be comfortable introducing it to patients until I had definitive information about keeping patient information confidential.”

Additional Concerns

Beyond confidentiality, social workers should ensure that the tool is not a stand-in for skill, on the part of the professional. With or without AI assistance, social workers need to retain their ability to write notes, create treatment plans, make diagnoses and more. AI is a tool, not a replacement.

“I’m a clinical supervisor and have been for a couple of decades,” says Bronecki. “I see the importance of people having to learn this on their own. Your ability to formulate an assessment, to formulate a diagnosis, to assess a patient (is important). On the other hand,I also see the benefit in terms of how efficient you can become.”

The balance can lie in turning to AI for recommendations or an alternate perspective but not relying on it to generate official diagnoses or plans. Notes that are transcribed or generated should be thoroughly reviewed and edited for accuracy, as well. In thinking of AI in this way, social workers ensure that their education and experience is the driving force in a client’s therapeutic services.

Additionally, social workers need to be aware of AI’s limitations. It will not always be correct or use up-to-date information, no matter the tool. For example, Groessl, with Brittany Maas, MSW, LCSW, assistant teaching professor at the University of Wisconsin Green Bay, conducted a study in which they asked two AI platforms to diagnose individuals from textbook case studies.

They used case studies that had individuals presenting with anxiety, depression, bipolar disorder or schizophrenia. While the chatbot was typically able to identify anxiety and depression, it most often missed the mark with schizophrenia, says Groessl, mislabeling it or not including subtypes for example.

“[The platform] did in most cases say ‘have a psychiatric examination to determine the diagnosis,” says Groessl, “which is frankly better than being definitive.”

In one instance, adds Maas, “One of the diagnoses that was given was actually one that was no longer in the DSM (Diagnostic and Statistical Manual of Mental Disorders). Other concerns that were noted was there wasn’t the inclusion of different aspects that we look at in social work. [It] wasn’t looking at somebody’s past trauma or their race or ethnicity.”

Even if it had considered those factors, “AI is biased,” says Badillo-Diaz. “A real concern is that our world has racism, sexism, classism. All of those isms are now also embedded into AI.”

Social workers must also be conscious of how any AI tool aligns with the social work Code of Ethics. Taking into account the aforementioned risks and concerns is a significant step in doing so.

Discussing With Patients First

Once a social worker has confirmed that an AI tool is safe, ethical and beneficial to their practice, they must also be prepared to discuss the tool with their client.

“I think AI needs to be treated like another person sitting in the room. If you had another person sitting in the room, you’d ask for [your client’s] permission. And you’re going to have to provide some information around what [the AI tool] is, such as ‘this helps me write my notes so I can see more people and it will give you resources right away and it can be helpful to you and to me,” says Baranski.

“I would just be really mindful about how we have those conversations with people,” she continues. “Because if you say, do you mind if AI records our session, I would say no. Or I would be like ‘Say more; what do you mean?’”

“Even if you’re not using AI to dictate notes or anything like that, it’s important to be open and up front about how you might be using it,” says Maas. And, she emphasizes, if a client says no, it means no.

Badillo-Diaz recommends making a conversation about AI a part of onboarding. In doing so, the clinician has the opportunity to clearly explain the technology being used and talk about the client’s relationship with AI in their everyday life.

Finding Resources

For a social worker interested in any of the available AI, they should do the research, find resources to help them troubleshoot and talk to other people about it. NASW and other member organizations in the U.S. and abroad (Badillo-Diaz notes that the British Association of Social Workers has recently released guidelines) are working to keep clinicians up to date. There are also sites, such as The AI Social Worker, that offer resources and additional information.

Baranski suggests asking yourself the following questions—questions she created as a general framework when thinking about AI:

• Are you using AI to better the profession or your ability to grow as a social worker?
• Did you acknowledge AI use to ensure your own ability and professional integrity?
• Did you verify that references provided are current, relevant or accurate?
• Are you ensuring that the information you use or upload is free from bias?
• Are you consistently working to make AI accessible to your clients if they choose to engage with it?

“If you can say yes to all of them,” she says, “very confidently, I would say go for it and also seek consultation.”

Sue Coyle, MSW, is a freelance writer and social worker in the Philadelphia suburbs.

NASW’s Social Work Online CE Institute (Search: Artificial Intelligence)

NASW’s Social Work Talks Podcast

The Social Work Podcast

Facebook Live: Chase Chats

Social Work Advocates (June-July 2023)

Social Work Advocates (Spring 2024)

NASW Ethics: AI and Social Work

Where Social Workers Seek Support

When asked what clinical social workers are turning to ChatGPT for in regard to their profession, the AI chatbot offered a list of seven things that mental health providers—it expanded beyond social workers—are asking. Number three on that list is how to work through burnout. ChatGPT reported that the burnout questions asked include:

• How do I stop taking this home with me?
• Is it normal to feel numb/angry/exhausted?
• How do I stay compassionate without breaking?
• How do I set boundaries when the system is failing clients?

While social workers largely agree that AI is not a replacement for a human therapist, they—like so many others—are also turning to this advancing technology for emotional support when they are not sure where else to go. It is a telling reflection on the weight that social workers carry and the structure of the profession, systems and organizations within which they work. It is also not where social workers should find support.

“The system that we work in pushes us toward productivity, leaving us putting clinical supervision and consultation [in the] backseat, because it’s not a billable hour,” says Roberta Bronecki, LCSW, founder of Serendipity Studio in Lyndon Station, Wis. “People are getting burned out. The importance of taking that time, even if it’s in your own time, to have and participate in consultation or clinical supervision and use it in a meaningful way [cannot be understated].”

“There are a lot of different reasons why someone might use ChatGPT for support in general, whether it’s finances or insurance issues. At the end of the day, we are not addressing self-care as much as we need to or as much as we think we are,” adds Brittany Maas, MSW, LCSW, assistant teaching professor at the University of Wisconsin Green Bay.

Social workers struggling with burnout should look for support within their organization, their local NASW chapters and other networks, relying, perhaps, on AI to help locate those resources but not be them. After all, says Amanda Baranski, DSW, LCSW, LMNIT, lecturer at the Department of Social Work in the College of Health and Human Sciences at Metropolitan State University of Denver, “Healing comes from human connection.”