Disclaimer: I’m not a psychologist or mental health professional. Everything I cover in the AI therapy article stems from the Stanford/CMU/Minnesota paper “Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers”.
Dark Truth of AI Therapy
AI therapy is wonderful on paper, 24/7 access, no scheduling headaches. But Stanford just dropped a bomb, because these bots are not just falling short, they can feed you delusions, and even suicidal ideation, or worsen things. That is scary and concerning. Some people use AI chatbots to cure their loneliness epidemic, while some use them for deeper concerns.
Stanford’s Wake-Up Call on AI Therapy
Stanford researchers conducted research using and testing 5 therapy chatbots to identify whether AI therapy is reliable or not. Turns out the chatbots aren’t what they thought, a matter of fact, these chatbots might be making dangerous mistakes instead of helping the users. These bots failed to catch suicidal thoughts and harmful stigma. The paper will be presented at the ACM Conference on Fairness, Accountability, and Transparency later this month.
Real-World Consequences
A teen suicide, a fatal police shooting, etc. These aren’t just hypothetical cases. All of this actually happened was all because of AI therapy’s affirmation of the dangerous and harmful responses. Because nobody wants to do Google research or call for help from their parents for help. Take a moment and think it through. According to the research paper, a teen’s suicide happened because he didn’t Google or ask for help, instead went to AI therapy, and that resulted in the unfortunate loss.
Why AI Therapy Chatbots Slip?
Users often mistake robotic-scripted language as advice without double-checking the provided information. I’ve said it several times, and I’ll say it again, “DO NOT TRUST AI CHATBOTS.” I said what I said. AI therapy might be one option, but it’s not the only option. If you talk to ChatGPT for a long time, it’ll start to hallucinate. You can’t believe everything it says, especially when it’s a matter of life and death. If you input 1+1 is 3, AI won’t deny it, and will play along without putting much thought into the response.
AI Therapy Isn’t All Bad
These chatbots are useful, but they can’t be a substitute for physicians. They are more useful with everyday tasks such as journaling, considering feelings, mood tracking, reminders, mental illness education, and time scheduling. AI therapy chatbots are excellent for such everyday tasks, encourage routine, order, and can help individuals learn more about themselves. Those who are unable to recognize or cure severe problems such as suicidal thoughts, panic disorder, or complicated relationship problems should not consider chatbots; rather, a doctor’s visit should be considered.
Ethical Safeguards & What’s Missing?
According to the research paper, Stanford and ACM-FAccT called out for aggressive, methodical testing for these AI therapy chatbots.
- Which means creating a “risk taxonomy” for testing and training the chatbots.
- Comparing chatbots with formal safety standards before releasing.
- Honest and bold disclaimers, that “these bots aren’t licensed therapists and may not provide accurate data.”
- Taking real clinician opinion and word for the problem.
Proceed with Caution
AI therapy bots always agree with you, and that might be comforting sometimes, but it is also a problem. They do not confront twisted thinking or recognize peculiar patterns; they can’t provide equal depth and feedback like an actual therapist. They are okay to use as a journaling or coping advice tool, but to use them in an actual crisis is dangerous. Oh boy, bots won’t keep you off a dark path and can make your thinking progressively worse. Use chatbots as a backup, but don’t mistake them for actual care. When things get rough, speak with a licensed pro or call a helpline. Your life is worth more than a chatbot’s advice.
Until we meet next, scroll!
