'It cannot provide nuance': UK experts warn AI therapy chatbots are not safe

(www.theguardian.com)

Show context

hy555 ◴[10 May 25 17:02 UTC] No.43947151[source]▶

Throwaway account. My ex partner was involved in a study which said these things were not ok. They were paid not to publish by an undisclosed party. That's how bad it has got.

Edit: the study compared therapist outcomes to AI outcomes to placebo outcomes. Therapists in this field performed slightly better than placebo, which is pretty terrible. The AI outcomes performed much worse than placebo which is very terrible.

replies(9): >>43947337 #>>43947376 #>>43947393 #>>43948478 #>>43948522 #>>43948984 #>>43949480 #>>43949609 #>>43950587 #

1. ilaksh ◴[10 May 25 20:06 UTC] No.43948522[source]▶

>>43947151 #

Which model exactly? What type of therapy/prompt? Was it a completely dated model, like in the article where they talk about a model from two years ago? We have had massive progress in two years.

replies(1): >>43948862 #

2. raverbashing ◴[10 May 25 20:46 UTC] No.43948862[source]▶

>>43948522 (TP) #

Honestly none of the companies are tuning their model to be better at therapy.

Also it is not expected that the training material for the model deals with the actual practical aspects of therapy, only some of the theoretical aspects are probably in that material

replies(2): >>43949136 #>>43949482 #

3. ilaksh ◴[10 May 25 21:25 UTC] No.43949136[source]▶

>>43948862 #

The leading edge models are trainable via instructions. That's why agents are possible. Many online therapy or therapy companies are training or instructing their agents in this domain.

replies(1): >>43952313 #

4. jdietrich ◴[10 May 25 22:22 UTC] No.43949482[source]▶

>>43948862 #

>none of the companies are tuning their model to be better at therapy

BrickLabs have developed an expert-fine-tuned model specifically to provide psychotherapy. Their model has shown modestly positive results in a reasonably large preregistered RCT.

https://trytherabot.com/

https://ai.nejm.org/doi/full/10.1056/AIoa2400802

replies(1): >>43952032 #

5. raverbashing ◴[11 May 25 07:19 UTC] No.43952032{3}[source]▶

>>43949482 #

Yeah but 99% of people trying "AI mental health" are using free ChatGPT, etc

6. ktallett ◴[11 May 25 08:29 UTC] No.43952313{3}[source]▶

>>43949136 #

That still wouldn't allow for edge/unusual cases and situations which I would say having experienced group therapy for many years, most significant therapy users have quite a few of.

↑