Comment by wisty
8 days ago
> Therapy isn't supposed to scale. It's the relationship that heals.
My understanding is that modern evidence-based therapy is basically a checklist of "common sense" advice, a few filters to check if it's the right advice ("stop being lazy" vs "stop working yourself to death" are both good advice depending on context) and some tricks to get the patient to actually listen to the advice that everyone already gives them (e.g. making the patient think they thought of it). You can lead a horse to water, but a skilled therapist's job is to get it to actually drink.
As far as I can see, the main issue I see with a lot of LMMs would be that they're fine tuned to agree with people and most people who benefit from therapy are there because they have some terrible ideas that they want to double down on.
Yes, the human connection is one of the "tricks". And while a LLM could be useful for someone who actually wants to change, I suspect a lot of people will just find it too easy to "doctor shop" until they find a LLM that tells them their bad habits and lifestyle are totally valid. I think there's probably some good in LLMs but in general they'll probably just be like using TikTok or Twitter for therapy - the danger won't be the lack of human touch but that there's too much choice for people who make bad choices.
Respectfully, that view completely trivialises a clinical profession.
Calling evidence based therapy a "checklist of advice" is like calling software engineering a "checklist for typing". A therapist's job isn't to give advice. Their skill is using clinical training to diagnose the deep cognitive and behavioural issues, then applying a structured framework to help a person work on those issues themselves.
The human connection is the most important clinical tool. The trust it builds is the foundation needed to even start that difficult work.
Source: a lifelong recipient of talk therapy.
>Source: a lifelong recipient of talk therapy.
All the data we have shows that psychotherapy outcomes follow a predictable dose-response curve. The benefits of long-term psychotherapy are statistically indistinguishable from a short course of treatment, because the marginal utility of each additional session of treatment rapidly approaches zero. Lots of people believe that the purpose of psychotherapy is to uncover deep issues and that this process takes years, but the evidence overwhelmingly contradicts this - nearly all of the benefits of psychotherapy occur early in treatment.
https://pubmed.ncbi.nlm.nih.gov/30661486/
The study you're using to argue for diminishing returns explicitly concludes there is "scarce and inconclusive evidence" for that model when it comes to people with chronic or severe disorders.
Who do you think a "lifelong recipient" of therapy is, if not someone managing exactly those kinds of issues?
Your understanding is wrong. What you’re describing is executive coaching — useful advice for already high-functioning people.
Ask a real practitioner and they’ll tell you most real therapy is exactly the thing you dismiss as a trick: human connection.
No, what they're describing is manualized CBT. We have abundant evidence that there is little or no difference in outcomes between therapy delivered by a "real practitioner" and basic CBT delivered by a nurse or social worker with very basic training, or even an app.
https://pubmed.ncbi.nlm.nih.gov/23252357/
Yeah, my issue is that I suspect an LLM based app may be easily "jailbroken" (since they tend to be highly agreeable due to their training) and turned into an enabler rather than a helper.
Even if some LLM therapists are good, with zero friction to go "doctor shopping" will result in a great many patients picking the bad ones that make them feel better, rather than the good ones that make them do better.
Where by “real practitioner” you mean “human being” which is my point entirely, thus we agree.
They’ve done studies that show the quality of the relationship between the therapist and the client has a stronger predictor of successful outcomes than the type of modality used.
Sure, they may be talking about common sense advice, but there is something else going on that affects the person on a different subconscious level.
How do you measure the "quality of the relationship"? It seems like whatever metric is used, it is likely to correlate with whatever is used to measure "successful outcomes".