[westcort]

Very cool! I tried to find transcripts of therapy to develop a system. This is the result: https://locserendipity.com/Therapy.html

It doesn’t always respond sensibly, but repeating the query will usually hit on a sensible answer after a few tries. It is best used with an intermediary who can enter the queries and respond in a way that is consistent with how a therapist might respond in a real session.

[ianbicking]

Hm... it feels about on par with Eliza. Or at least I haven't gotten it to provide any constructive insight or meaningful questions. At its best what kind of interactions can it produce?

[westcort]

Sure, here is an example of a productive exchange:

Bot: Hello, I am a chat bot to talk with about problems. What can I talk with you about?

You: Well I have a bit of a problem with having trouble getting myself to exercise

Bot: And so it seems like there's a way to find out what the right opportunities for you are.

You: well I enjoy rowing on my machine

Bot: It sounds like a challenge to me, isn't it? Is there something in you says, I don't want to something unless it's really right?

You: Sure, but I enjoy rowing on the machine, it's just hard to set up a consistent time for those things

Bot: You said you want to to deal with this problem and get rid of it.

You: I am not sure how to set up a system so that I will exercise

Bot: And again, tell me some times where you felt like you were doing the right thing. What were some other times where you had those kind of feelings?

You: Well, I had that feeling when I was out running

Bot: You're saying too that you know perfectly well the feeling within yourself that occurs when you are really doing something that's right for you.

You: Yes, so you are saying maybe I need to vary my exercise routine to some degree?

Bot: And that's something of interest to you?

You: Maybe, I could try it. Maybe I could do an hour of a different exercise on some days

Bot: That's true. That's very true.

You: Okay, so that's what I will try. Do you think there might be something else I can do to make myself more consistent?

Bot: Is it fair to say that you feel, really, that at times you're acting in ways that are not in accord with your own inner standards?

You: well yes

Bot: And you want to learn new things. Is that fair to say?

You: Yes I do want to learn new things

Bot: What might you say?

You: I might say that I could learn some strategies to be more consistent about my exercise, or try something new and see what works for me

Bot: What are you doing about that?

You: I could set up an exercise log in a paper book

Bot: All right. So you mentioned feeling guilty about making a choice. How do you feel about that right now?

You: Maybe a little better about it because I know there is something I can try to improve my exercise routine or improve my consistency

Bot: The point is, you haven't forgotten.

You: Right, I am motivated now

Bot: It sounds like a tough assignment.

You: Yes, but I am up for it

Bot: One thing I might ask, what is it you wish I would say to you?

You: I guess you could say you think I can make it work

Bot: Well, yeah. It sounds like one of the hard things is that you really value the time late at night.

You: Yes, maybe I could shift more to the morning

Bot: Is that the main idea?

You: Well maybe ultimately it is because if I got up early I could make the time to exercise with more consistency

Bot: You mean you want to get over this for yourself?

You: Yes, it is my responsibility

Bot: That sounds right.

[ianbicking]

It feels pretty hollow to me. If the person wants to guide themselves through some issue it's possible for them to do it in the format of the bot, but the bot isn't really assisting, and it's actually harder to follow through the thought process when responding to the bot than it would be without it.

Well... that's kind of critical, but I'm specifically interested in this stuff so I'm apt use to a critical eye.

I'm still trying to understand Rogerian therapy... I do think it has particular application because it's centered on the idea that the healing or changes the patient seeks ultimately comes from within that person. The humility of that stance is one where there might be a place for a computer, as it has an opportunity not to try to tell the person who they should be or how they should act, but only to act as a kind of emotional mirror, reflecting the patient's own statements and intentions back to them (hopefully constructively!) Most computer-assisted mental health products right now are, in comparison, very behavioralist.

There's also more to Rogerian therapy, especially empathy which is perhaps forever out of reach of a computer. PRETENDING to feel empathy is possible, of course... but I believe another principle is honesty, and pretending to feel empathy is not honest.

Anyway, it feels like this could benefit from being more reflective: have the bot repeat back to the user more often. It's okay if it can't come up with new ideas or tell the person what to do, but being able to rephrase statements or pull up meaningful past statements could be genuinely insightful without attempting to generate insight.

[makan_amini]

This is really cool. Would love to chat and exchange ideas if you are up for it! /Makan Amini (Co-founder at Meomind). My email is [e-mail]. Thanks!