ETA: I'm not sure how the AI therapist would work for more detailed traumatic material
So I've been specifically testing it for this and I do have some advice if you'd like! The most helpful, the longest, and most detailed responses I've gotten have been to queries where I take my specific experiences, generalize them, and then ask for ChatGPT's advice and interpretation of the event
before getting into personal details. For example, I created a thread on indoctrination and brainwashing where I discussed my specific experiences - this was tailored for human beings, was full of prosaic and figurative language, used lots of idioms and metaphors, and had several different questions, topics, interpretations etc (as well as simply being a lot of words/text).
ChatGPT did its best with this but you could tell that it genuinely didn't understand what I was saying. Its answer was less than a paragraph long and basically boiled down to "you used the word brainwashing!" However, in the threads where I modified my experiences to tailor to AI, I instead started by establishing a baseline understanding of brainwashing/indoctrination, simply by asking ChatGPT "what is your opinion on the concept of indoctrination and brainwashing? I want to discuss this in relation to organized groups such as cults, the military, gangs, etc."
It immediately produced a ton of valuable information (including information I hadn't known about) and clarified that "brainwashing" was not a scientific term - so I asked it about how my therapist refers to it, indoctrination, which again produced valuable results. Once we had that baseline token I was able to clarify about my specific experiences by asking it how it interprets the concepts of personal responsibility and brainwashing, if this coercion occurred to me as a child, etc.
The goal will vary depending on different users. Some will just want to practice talking about trauma, some will want specific information related to their trauma (I have tested it both ways, as long as you keep your experiences concise and factual, it will understand what you're saying and provide feedback - I used some examples such as COCSA which are more complicated, and it grasped the nuances easily - it very clearly understands the difference between coercion, assault/rape, and the understanding that children aren't as legally responsible for their behavior as adults, and that the facts of the situation are separate from my feelings).
In a lot of ways, this is actually reflective of the type of therapy I'm doing with my human therapist! In FORNET, as I mentioned to the AI, the goal is to create a trauma narrative that is devoid of thoughts, feelings, prose and opinions and instead strip all experiences down to their factual elements. So this is actually a type of therapy that humans do with one another, and being able to practice getting into this mindset through speaking to ChatGPT is beneficial, because ChatGPT will call you out for opinions (like, "I'm a monster-" it told me "you are not a monster," and then we discussed self-judgment).
And provide multiple theories of interpretation (when asked to discuss moral injury, it directly references the kinds of questions therapists are most likely to ask, and asked me some of those questions - like, "what's your evidence that you're a bad person because you killed someone?" etc.) The data that it appears to be trained on is very up-to-date with trauma processing, it will all come down to being able to make yourself understood. And if it gets that wrong, you can actually correct it - quote the part they didn't understand and say "this is wrong, I meant XYZ."
This is how generative AI actually learns, so you will get a ton more out of your threads if you take the time to clarify the concepts you want to discuss, get it to define those concepts, refine the definition, and then use personal examples.
