With all the discussion about the advantages and disadvantages of artificial intelligence, behavioral health providers are beginning to assess how AI may impact their clinical and administrative responsibilities. In Illinois, Governor JB Pritzker recently signed a bill that limits the use of AI in mental health services. Basically, AI is not allowed to be used to make clinical decisions or guide treatment. AI is allowed to be used for administrative tasks, however.
Many of us remember “Eliza,” the computer therapist [program. Written in BASIC on mainframe computers in the 1960s, Eliza gave rudimentary answers in a style reminiscent of Carl Roger’s Person-Centered Therapy, usually just repeating the user’s statements back and making occasional comments based on a primitive word-parsing scheme. Science Fiction writers also came up with various fictional AI therapists, like Lorry Niven’s “Sigfrid Von Shrink,” who appeared as a hologram resembling Sigmund Freud.
Over time, “Chatbots” have been developed as an interface between users and professional mental health services. These operated as somewhat glorified phone-tree systems, substituting text responses for prompts. If the user needed anything beyond psycho-educational materials, they would be connected to an actual live clinician. However, skilled and experienced practitioners still need to be the ones making clinical judgments.
For many years, publishers of psychological assessment instruments like the MMPI have offered computer scoring and interpretive report services. These reports are given in plain language that can be incorporated into clinician’s reports using the assessor’s clinical skills and judgment. These interpretations are based on research and limit themselves to wording like “it is likely that this person…” and similar phrases. Since they don’t know the respondent and are just giving statistically guided suggestions, they must be used with caution and not communicated directly to patients or other agencies (such as probation officers). Although AI-like algorithms are used in these reports, they are not as sophisticated as today’s large language models.
Current AI models have developed the capability to respond in a more sophisticated way, potentially giving the impression that they are using clinical judgments in actual practice. However, human skill and judgment is still needed in making decisions and recommendations about human behavior. This caution is likely the basis for current legislation limiting the use of AI in behavioral health practice.
In keeping with the new legislation, many therapists (myself included) have begun using AI-enabled documentation and reporting tools. My electronic health record platform (Simple Practice) now has a “note taker” function, which can transcribe session content and organize it into a standard progress note. I have made sure, before using these tools, that they will not compromise client privacy. The session transcripts are erased as soon as the note is created and edited / approved by the therapist, and notes are always reviewed and any clarification or corrections that need to be made have been added before signing the note. Basically, the note is (like a manually created note) simply a summary of what was discussed in the session, including therapist recommendations and any “homework” assignments given.
Simple Practice can also add the results of any assessments that the client has completed to the initial progress note and draft treatment plan note. This is for documentation purposes, and not for actual treatment planning purposes.
I have added a consent for AI-enabled progress notes and treatment planning documents to my consent forms. If you don’t want me to use them, you can decline.
AI and Behavioral Health by Fitzgerald Counseling is licensed under a Creative Commons Attribution-NoDerivatives 4.0 International License.