This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
I'm curious how the Motte sees using AI for therapy / life advice? Online I'm seeing a ton of people recommend Claude especially, but others are skeptical.
On the one hand I could see it being useful because of the fact that you have nigh-unfettered access to it, and can really dig into deep problems. Also, it's trained on all the therapy texts of course.
The other, more culture war issue, is that due to the way RLHF works, they will likely be pushing one ideological lens over another. Especially about deep topics like morality, relationships, casual sex, etc.
Overall I think it's a fascinating area of development, and I'm still optimistic that LLMs could help people much more than the average therapist. Mainly because I'm pretty bearish on the help people get from the average therapist.
Anyway, what do people think about therapy becoming AI?
Tremendously poor idea, general purpose chatbots have already led to suicides (example- https://amp.theguardian.com/technology/2024/oct/23/character-ai-chatbot-sewell-setzer-death).
Purpose built ones will have more safeguards but the problem remains that they are hard to control and can easily go off book.
Even if they work perfectly some of the incentives are poor - people may overuse the product and avoid actual socialization, leaning on fake people instead.
And that even if is doing a ton of work, good therapy is rare and extremely challenging, most people get bad therapy and assume that's all that is available.
Services like this can also be infinitely cheaper than real therapists which may cause a supply crisis.
I'm afraid at least this particular example is wrong, and popular media grossly misrepresented what happened:
https://www.lesswrong.com/posts/3AcK7Pcp9D2LPoyR2/ai-87-staying-in-character
(Note that one of links has rotted, but I recall viewing it myself and it supported Zvi's claims)
Anyway, I have a more cynical view of the benefits of therapy than you, seeing it rather well described as a Dodo Bird Verdict. Even relatively empirical/non-woo frameworks like CBT/DBT do rough as well as the insanity underpinning Internal Family Systems:
https://www.astralcodexten.com/p/book-review-the-others-within-us
Even I have to admit that Freudian nonsense grudgingly beats placebo.
You seem to agree that good therapists are few and far between, but I'd go as far as to say that I'm agnostic between therapy as practiced by a good LLM and the modal human therapist.
Oh hey!
When you get a chance I would love to hear how things are going for you!
On to the matter at hand -
Please update my understanding of that particular suicide if it's incorrect, but what I'd heard is that the person was substituting human contact with the chatbot and his parents didn't catch the worsening social withdrawal because he was telling them he was talking to someone. My fear is not that chatbots will encourage people to do things, but that they won't catch and report warning signs, and serve as an inferior substitute for actual social contact. Not sure what the media presentation is since I'm relying on professional translation.
Moving beyond that however, I think you underweight the value of therapy. DBT and CBT have excellent quality evidence at this point. The reason for those two specifically is likely two fold - they are "simpler" to perform, and because they are more standardized they are easier to research.
Also, good psychodynamics is not Freudian nonsense, it's mostly CBT with different language and some extra underlying terminology that is very helpful for managing less severe pathology. Again I tell you to read Nancy McWilliams haha.
At its absolute worse therapy is stuff like forcing social interaction, forcing introspection and so on. Some people can function well off of a manual, and some people can study medicine on their own. But nearly everyone does better with a tutor, and that's what therapy is.
A tutor is also more likely to catch warning signs because of (at this time) superior human heuristic generation and the ability to perform a physical and mental status exam.
I've been rather miserable since I've gotten here, for a multitude of reasons, which had notably dampened my appetite for chatting up my day job online. I'm slightly less miserable right now, which is why I'm back at it! I can elaborate in DMs if you'd like.
I raised objections against claims made exceedingly uncritically in the Guardian post you linked to (having assumed you endorsed it). For example-
I can cut a grieving mother some slack, but the facts don't bear out her beliefs, and the Guardian doesn't really do much journalism here, since it would otherwise suggest her suit is unfounded.
Your personal claims seem more subtle, but even then, I find it very hard to blame the chatbot for social withdrawal here. I'd point out you can make the same argument for anything from reading books to watching anime (a bullet that some may bite, of course). In other words, a potential refuge for the maladjusted, but also something that the majority of people would be loathe to ask others to consume less of or ban altogether, on the grounds that it's a net negative.
(I think the case for social media being far worse for teenage mental health is significantly more robust, and I still wouldn't advocate for it to be banned. In the case of chatbots, I haven't been nudged out of the null hypothesis.)
Imagine the chatbot was replaced by, idk, a Runescape girlfriend (do kids these days have those? Potentially substitute for someone grooming them on Discord), would you expect said person to be significantly more helpful, or at least worthy of blame? I wouldn't.
I'll have to see if it's relevant to the MRCPsych syllabus, God knows that having an unpleasant time with the subject makes most reading on it feel unpleasant :(
A fair point. But I contend that an AI therapist is capable of doing those things, in a limited but steadily improving fashion. You can have a natural language spoken conversation with ChatGPT, and it's very capable of picking up minor linguistic nuance and audio cues. Soon enough, there'll be plug and play digital avatars for it. But I think that therapy through the medium of text works better than doing nothing, and that's the standard I'm judging chatbots by. Not to mention that they're ~free for the end user
God knows what the standards for AGI are these days, with the goalpost having moved to being somewhere near a Lagrange point, but I would sincerely advocate the hot take that an LLM like Claude 3.5 Sonnet is smarter, more emotionally intelligent and a better conversationalist than the average human, and maybe the average licensed therapist.
It is, of course, hobbled by severe retrograde amnesia, and being stuck to text behind a screen, but those are solvable problems.
To run with your analogy, an AI therapist/teacher is far closer to a human therapist/teacher than they are to a manual or textbook! You can actually talk to them, and with Hlynka not being around, the accusations of stochastic parrotry in these parts has dropped precipitously.
What I'm really advocating for is not letting the perfect become the enemy of the good, though I'd certainly deny that human therapists are perfect. I still think that access to AI therapists is better than not, and I'm ambivalent when putting them up against the average human one.
Though I'd also caveat that Character AI probably cheaps out, using significantly dumber models than SOTA. But it's not the only option.
I find it deeply frustrating to see orherwise intelligent people (who by all rights ought to know better) anthropomorphizing algorithms in this way.
In order to "be hobbled" by retrograde amnesia it have to be capable of forming memories in the first place.
An LLM is literally just a set of static instructions being run against your prompt. Those instructions don't change from prompt to prompt or instance to instance.
I genuinely don't understand the objection here?
Drawing an analogy isn't the same thing as excessive anthromorphization. The closest analogue to working human memory is the context window of an LLM, with more general knowledge being close to whatever information from the training set is retained in the weights.
This isn't an objectionable isomorphism, or would you to object to calling RAM computer memory and reject it as excessive anthromorphization? In all these cases, it's a store of information.
An otherwise healthy child born with blindness can be said to be hobbled by it even if they never developed functioning eyes. I'm sorely confused by the nitpicking here!
The utility of LLMs would be massively improved if they had context windows more representative of what humans can hold in their heads, in gestalt. In some aspects, they're superhuman, good luck to a human being trying to solve a needle in a haystack test over the equivalent of a million tokens in a few seconds. In other regards, they're worse off than you would be trying to recall a conversation you had last night.
You can also compare an existing system to a superior one that doesn't yet exist.
I never claimed otherwise? But if you're using an API, you can alter system instructions and not just user prompts. But I fail to see the use of this objection in the first place.
More options
Context Copy link
More options
Context Copy link
Anterograde, not retrograde. It didn't forget something it knew from its life before; it's unable to permanently remember new things. LLMs are like Clive Wearing or Hermione Granger.
Isnt anteretrograde just prograde
More options
Context Copy link
An LLM can be loosely said to have both kinds of amnesia. It has retrograde amnesia in the sense that any information it had in its context window becomes "forgotten" when too much new information is accepted and overrides it. Or simply a conversation it had in a previous instance, treating different copies as the same entity.
More options
Context Copy link
More options
Context Copy link
Will message you.
And yeah no doubt the media fucking sucks.
My fear is that people will engage in HER style stuff and this example is a bleeding edge version of that.
McWilliams is useful even if you are just skimming the personality disorder chapters because you will have colleagues with those. It's also interesting enough to make you go through it at pace haha.
I think things like your therapist looking at you like you are an idiot and you going "yeah I know" are underrated parts of therapy and the chatbot isn't going to do those things for now.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link