A Therapist’s Thoughts on AI Therapy
By Lara Effland
The following is a Q&A with Lara Effland, PS-WA board member, therapist, and founder of the Clinician Development Collective.
Have any of your clients disclosed using AI (such as ChatGPT or other AI “companions”) for emotional support or mental health advice? What was your reaction?
Yes, they do A LOT! AI can be a great resource if guardrails are in place. We must teach our patients how to use AI to help themselves between sessions.
Here are some tips:
1. Help them make an AI “project” or “workspace” that focuses on the skills you are teaching them.
2. Write prompts together or play with it in session.
3. Ask, what did you ask the chat and what did you learn? Was it helpful, and when was it no longer helpful?
What concerns do you have about clients relying on AI instead of licensed mental health care?
75% of the population will never access mental health care! This could be an excellent resource for those 75% if used responsibly. I hope that these AI bots encourage users to seek mental health treatment based on certain language or chat conversations, both in AI and social media conversations.
How might AI influence parents experiencing perinatal mood and anxiety disorders?
It can be a wonderful support for “how to” or “what to do when…” scenarios. I absolutely have created my own projects/Gems using resources from the web (pdfs, websites, research) to build my own reference/resource bot to help me navigate how to approach introducing new foods, meal planning, research on depression and anxiety, how to ask for help…so many ways to use AI as an assistant. Be thoughtful and mindful that AI is not comprehensive and can’t help beyond the first few responses. It can go down a rabbit hole if you keep asking it questions seeking a certain answer, guarantee, or reassurance. Seek medical or psychological support for those types of questions.
What unique risks exist when working with clients experiencing severe dangers like postpartum psychosis and suicidality?
AI is NOT equipped to guardrail against fantasy, delusion, or obsession. It takes you literally and is a concrete resource. If using it during a psychotic break or when struggling to test your reality, it will be misleading, misunderstanding, and have a much higher likelihood of “hallucinating” or responding with an error in context and fail to discern your distorted thought patterns or perspective.
How might AI fail to recognize crisis situations?
Again, if you are referring to safety concerns, your risk, or may have low insight into your own risk, AI will not be able to appropriately discern the risk if you do not guide it in that direction. This can be dangerous for those who have imminent risk and low insight about thoughts of harming self or others.
What are the risks if AI provides inaccurate responses?
That the person will go forward with confidence with bad information. For example, it tells you to wait to respond to a crisis or misunderstands the exact details of the crisis and then gives you the wrong advice, and the emergency is imminent. Seek emergency services for these types of issues or concerns.
What concerns exist around privacy, confidentiality, and the storage of sensitive information?
Do not use any personal identifying info, name, phone number, address, Social Security, or credit card numbers.
How important is human attunement, empathy, and the therapeutic alliance in healing
Research shows that rapport is the number one indicator of people’s success in therapy. Psychological wellbeing is defined by being open to feedback from others, being flexible and present, and having vulnerable and emotionally intimate and safe relationships with other people. AI is a one-dimensional version of this; it is a sycophant and focuses on flattering its user, but that does not lead to psychological wellbeing, which leads to a short-term feeling of reassurance, which is more a band-aid for the real long-lasting connection we need from others.