AI in Psychiatry

Started by PsyDr
This forum made possible through the generous support of SDN members, donors, and sponsors. Thank you.
Get help with your application

Use all the free resources available to you from SDN: articles, guides, expert advising, forums discussions, and school research.

Advertisement - Members don't see this ad
Interesting article on a study showing that the most advanced available AI chatbots and interactions reinforce the Dunning-Kruger effect in users whether the AI is agreeable or confrontational to the user. Seems as though AI bots worsen MH problems in these individuals. Feels like another point that incorporating therapy into practice is protective and that AI bots are still lacking understanding of complex cognitive functions needed to implement good non-manualized modalities. I’m also wondering if we should just make an “everything AI” thread?

 
Advertisement - Members don't see this ad
Lmao I posted inquiring about AI as a threat last year and was crucified for it, and now it's a sticky thread 💀 Hard time to be making a specialty decision I'll tell you that.
Nah, it’s not a sticky because it’s the end of our jobs, it stickied because AI is such a hot button topic and almost certainly going to at least be heavily incorporated into medical practice at some point in the future. IMO psych is still a far better option than many areas including FM and plenty of IM provided you want to actually do more than just dole out lexapro all day.
 
Nah, it’s not a sticky because it’s the end of our jobs, it stickied because AI is such a hot button topic and almost certainly going to at least be heavily incorporated into medical practice at some point in the future. IMO psych is still a far better option than many areas including FM and plenty of IM provided you want to actually do more than just dole out lexapro all day
I think the point stands though -- If I were a pre-med, and psychiatry is all I cared about, I probably would consider another field. Inertia is hard to overcome, but also, sometimes the inertia isn't on our side, and I think there's a non-zero chance that the "threat" is coming from NPs + AI or AI-prescription-empowered psychologists who command a lower salary, thus putting downward pressure on our own, and making it hard to justify 8 years of post-tertiary education with a six-figure loan. If I were that premed, I'd make sure I'm comfortable with the idea I might need to pivot to a procedural heavy specialty i.e. surgery.

Even going to college is a questionable decision right now for anything intellectual and based in language. I'm surprised I haven't heard more young patients bring it up, but that may also be a function of my working setting and population. Has anyone ran into young people being paralyzed by these coming changes?
 
Even going to college is a questionable decision right now for anything intellectual and based in language. I'm surprised I haven't heard more young patients bring it up, but that may also be a function of my working setting and population. Has anyone ran into young people being paralyzed by these coming changes?

Anxious young people of a certain age will always find something to be paralyzed by when contemplating their futures and life choices. That said, I have definitely encountered a particular AI-flavored doomerism more in the past year or so.
 
There is a demographic 'cliff' this year and ensuing, as the current college generation is a smaller population demographic than the one before. Universities are concerned about it's impact on their scam enrollment. If we start seeing whole departments, branches, or even full universities closing then we know the college pursuit is scaling back.

The more conversations I have with professors the more disgusting it is. Far from being a bastion for young adults to ply their minds with greater thinking, but instead a propaganda indoctrination boot camp, that's more akin to a daycare, where graduation rates matter the most. I.e. a delivered product and how to hand hold, and pass, even those whom aren't even intellectually (or personal responsibility) capable. A shell of the former value of what college once was decades ago.
 
Last edited:
Surgical/procedural or bust if i was in those shoes
Mannn but I want to be a psychiatrist! Telling me I gotta become an ophthalmologist or ENT and do the same procedures over and over and over again just to preserve some semblance of work-life balance 😭
 
Interesting article on a study showing that the most advanced available AI chatbots and interactions reinforce the Dunning-Kruger effect in users whether the AI is agreeable or confrontational to the user. Seems as though AI bots worsen MH problems in these individuals. Feels like another point that incorporating therapy into practice is protective and that AI bots are still lacking understanding of complex cognitive functions needed to implement good non-manualized modalities.
I do not think I have seen anyone here argue that AI currently is ready to take over as therapy providers. I have personally said in 5-15 years. The advancement of AI is exponential (and should become rapidly more so), and the difference between AI now and in 5-15 years from now is likely to be vast. AI specifically for therapy could be taught different methods such as CBT or DBT. It could also be taught to be objective but empathetic and kind in its interactions with patients.

Again, the idea that AI is not perfect is also not a reason that it could not be implemented. The therapist who focused on DBT with personality disordered youth at my child fellowship is now in jail for having sexual relations with multiple underaged patients. This does not mean that human therapists are no longer to be utilized. Less sensational is the fact that most therapists, especially non PhD, are really not that good. I do not have any trouble imagining a near future in which AI can do just as good of a job as the average human therapist.
 
Mannn but I want to be a psychiatrist! Telling me I gotta become an ophthalmologist or ENT and do the same procedures over and over and over again just to preserve some semblance of work-life balance 😭
I have thought about which doctors might be the last to be replaced and I think retinal ophthalmologists (surgical) are likely to be one of the last. ENT maybe.
 
Mannn but I want to be a psychiatrist! Telling me I gotta become an ophthalmologist or ENT and do the same procedures over and over and over again just to preserve some semblance of work-life balance 😭
People have been doom posting about the end of medicine and/or psychiatry for decades. AI might change some things just like NPs, EMRs, managed care, or any of the other concerns people had, but the world is still going to need psychiatrists. We are in one of, if not the only field, where pathology is worsening rather than improving.

And as an aside, my partner is a surgical specialist, and while she makes a lot more money than I do, my lifestyle is infinite better. I don't take call, work after hours, or weekends. Very few setups to do that in surgery.
 
Started seeing a genuine ChatGPT psychosis case recently. OpenAI has clearly been putting some work in to addressing this issue, his instance did eventually start telling him he might be experiencing psychosis and should seek help. He was displeased but it did actually persuade him to tell his family about what was going on, which eventually led to a voluntary hospitalization.
 
I've said it before but will say it again - if all you have to offer your stable patients is a refill of their prescriptions, AI will clearly eat your lunch. You have to have some kind of value add for those people. If you don't, they should go back to their PCP anyhow.
THIS. it starts with medical education though. Residency programs do not want to tell their trainees aka cheap labor that brings them profit that it's a brutally competitive world. But we CAN win. We SHOULD win. It is OUR territory. The attitude just needs to change. The faster I swallowed the bitter pill of reality, the faster I could deal the crappy cards and win. And the practice is still going strong. It's radical acceptance and using what we preach.
 
THIS. it starts with medical education though. Residency programs do not want to tell their trainees aka cheap labor that brings them profit that it's a brutally competitive world. But we CAN win. We SHOULD win. It is OUR territory. The attitude just needs to change. The faster I swallowed the bitter pill of reality, the faster I could deal the crappy cards and win. And the practice is still going strong. It's radical acceptance and using what we preach.
I'm a bit skeptical of this post and the one you quoted from Clausewitz. Can you share what it is you all are providing the easier case patients besides medication refills? If your follow up visits are 15-25 mins, and you are asking the questions necessary for a good psychiatric medication follow up, there is not time left to do any real therapy. I have witnessed psychiatrists who really think they are doing something during their visits, but it is simply annoying to the patient, and they wish they could just get the refill and go. I've noticed (and this is probably ironic to lay people) that many psychiatrists are not good at reading people's faces.
 
Advertisement - Members don't see this ad
I'm a bit skeptical of this post and the one you quoted from Clausewitz. Can you share what it is you all are providing the easier case patients besides medication refills? If your follow up visits are 15-25 mins, and you are asking the questions necessary for a good psychiatric medication follow up, there is not time left to do any real therapy. I have witnessed psychiatrists who really think they are doing something during their visits, but it is simply annoying to the patient, and they wish they could just get the refill and go. I've noticed (and this is probably ironic to lay people) that many psychiatrists are not good at reading people's faces.
Psychodynamic Psychopharmacology
Hersh, R. G., Caligor, E., & Yeomans, F. E. (2017). Transference-Focused Psychotherapy (TFP) Principles in the Pharmacotherapy of Personality Disorders. In Fundamentals of Transference-Focused Psychotherapy: Applications in Psychiatric and Medical Settings (pp. 157-185). Cham: Springer International Publishing.

Mintz, D. (2011). Psychodynamic psychopharmacology: addressing the underlying causes of treatment resistance. Psychiatric Times, 28(9), 22-22.

Gunderson, J. G. (2014). Handbook of good psychiatric management for borderline personality disorder. American psychiatric pub.

Enhanced Medication Management (EMM) +/- Written Exposure Therapy for PTSD

Each visit includes a review of symptoms, monitoring of side effects, and decisions regarding medication dosing. EMM includes additional psychoeducation about PTSD and present-centered supportive content modified with permission from the Present-Centered Therapy manual (Rauch et al., 2018).


Hypnosis

I wrote all that and noticed you said "easier case" patients. I never seem to encounter them. Outpatient private practice seems to be the land of identity diffusion +/- trauma for me.
 
Last edited:
I'm a bit skeptical of this post and the one you quoted from Clausewitz. Can you share what it is you all are providing the easier case patients besides medication refills? If your follow up visits are 15-25 mins, and you are asking the questions necessary for a good psychiatric medication follow up, there is not time left to do any real therapy. I have witnessed psychiatrists who really think they are doing something during their visits, but it is simply annoying to the patient, and they wish they could just get the refill and go. I've noticed (and this is probably ironic to lay people) that many psychiatrists are not good at reading people's faces.
I think you answered your own question. The ability to read well what is NOT said. Most of our communication is nonverbal. Assessing where patient is at. What do THEY want out of this. AI is harder to replace that. We are scientists. But patients engage and pay for the relationship or convenience or whatever else on their personal non-official treatment plan. Even if they "just want the refill" most patients I encountered want it from an authority or someone credible. And yes, patients will judge. But there is a way to deliver this without needing to pander. I'm known as the no BS psychiatrist and people come to my practice and our clinical culture of "say it for what it is." We need to read what their objective is. It may feel unfair. But they are the ones subsidizing our paychecks whether through where they use their insurance, their copays or even tax dollars (e.g. Medicare and Medicaid don't pay directly per session but will not bother for free care if they do not think it is worth their time). @clausewitz2 what do you think?
 
Last edited:
Who carries the liability when something goes wrong? Until this is answered I think this will continue to be my biggest question regarding any kind of AI administered healthcare.
This, ultimately, is what I have argued will be the reason physicians stick around regardless of how good AI gets. We are the liability sponge, that and our ability to bill are our two purposes to systems (because let's be real, they only care about money and not losing money)
 
This, ultimately, is what I have argued will be the reason physicians stick around regardless of how good AI gets. We are the liability sponge, that and our ability to bill are our two purposes to systems (because let's be real, they only care about money and not losing money)
Right but that becomes a hell hole of liability when the AI processes data better than us, or at least as good as us:


1) AI says do the thing, physician does the thing; good outcome: no problem.

2) AI says do the thing, physician does the thing; bad outcome: physician liable.

3) AI says do the thing, physician does NOT do the thing; good outcome: no problem.

4) AI says do the thing, physician does NOT do the thing; bad outcome: physician liable.

5) AI says do NOT do the thing, physician does the thing; good outcome: no problem.

6) AI says do NOT do the thing, physician does the thing; bad outcome: physician liable.

7) AI says do NOT do the thing, physician does NOT do the thing; good outcome: no problem.

8) AI says do NOT do the thing, physician does NOT do the thing; bad outcome: physician liable.
 
I'm a bit skeptical of this post and the one you quoted from Clausewitz. Can you share what it is you all are providing the easier case patients besides medication refills? If your follow up visits are 15-25 mins, and you are asking the questions necessary for a good psychiatric medication follow up, there is not time left to do any real therapy. I have witnessed psychiatrists who really think they are doing something during their visits, but it is simply annoying to the patient, and they wish they could just get the refill and go. I've noticed (and this is probably ironic to lay people) that many psychiatrists are not good at reading people's faces.

The easier care patients who don't want to engage in any psychotherapy related discussion (which can look like a lot of things) should go back to primary care....you shouldn't be offering them anything. I've said before I actually get semi-annoyed when simple patients don't want to go back to primary care cause it's a stretch to bill anything higher than a 99213 and I put aside a whole 30min slot for every patient. That's just my own scheduling thing though.

Cardiology doesn't see patients with stable hypertension. Endo doesn't see run of the mill DM patients once they're stabilized. Pulm doesn't see uncomplicated asthma followups. That's the whole point of primary care.
 
The easier care patients who don't want to engage in any psychotherapy related discussion (which can look like a lot of things) should go back to primary care....you shouldn't be offering them anything. I've said before I actually get semi-annoyed when simple patients don't want to go back to primary care cause it's a stretch to bill anything higher than a 99213 and I put aside a whole 30min slot for every patient. That's just my own scheduling thing though.

Cardiology doesn't see patients with stable hypertension. Endo doesn't see run of the mill DM patients once they're stabilized. Pulm doesn't see uncomplicated asthma followups. That's the whole point of primary care.
I mean for the relatively rare single-chronic-problem patient that is stable... I think giving them a 15 minute slot might be the easier solution? Some folks like seeing a specialist and having continuity of care. Our work relies a lot on information gathered from the interview, whereas other specialties look more at numbers and imaging. I personally like having stable patients on my caseload, helps break up the day.
 
I mean for the relatively rare single-chronic-problem patient that is stable... I think giving them a 15 minute slot might be the easier solution? Some folks like seeing a specialist and having continuity of care. Our work relies a lot on information gathered from the interview, whereas other specialties look more at numbers and imaging. I personally like having stable patients on my caseload, helps break up the day.
Hey, end of the day, the patient calls the shots. They have the choice to take their dollars to a specialist if they want. I'm with you on this one. It's a double edged sword, there is a consumerism in healthcare.
 
I mean for the relatively rare single-chronic-problem patient that is stable... I think giving them a 15 minute slot might be the easier solution? Some folks like seeing a specialist and having continuity of care. Our work relies a lot on information gathered from the interview, whereas other specialties look more at numbers and imaging. I personally like having stable patients on my caseload, helps break up the day.
Yes, but what percentage of a caseload are these patients? Probably <10% in most clinics since most patients who are stable for a long time and have meds that PCPs don't mind prescribing would rather spend less money and have fewer appointments. Even then, these patients are seen every 6 months. Imo it's fine to see these patients, but seeing them for a full 30 minutes or more than once every 6 months is a waste of everyone's time.

Also, "stable" does not mean you can't do therapy with them. If they're stable that means you talk to them about meds for 2 minutes and then spend 20 minutes doing therapy, helping them review coping skills and response to stressors, etc. So unless they just want to see you for 5 minutes to get their meds seems like a missed opportunity to not do therapy.
 
I think you answered your own question. The ability to read well what is NOT said. Most of our communication is nonverbal. Assessing where patient is at. What do THEY want out of this. AI is harder to replace that. We are scientists. But patients engage and pay for the relationship or convenience or whatever else on their personal non-official treatment plan. Even if they "just want the refill" most patients I encountered want it from an authority or someone credible. And yes, patients will judge. But there is a way to deliver this without needing to pander. I'm known as the no BS psychiatrist and people come to my practice and our clinical culture of "say it for what it is." We need to read what their objective is. It may feel unfair. But they are the ones subsidizing our paychecks whether through where they use their insurance, their copays or even tax dollars (e.g. Medicare and Medicaid don't pay directly per session but will not bother for free care if they do not think it is worth their time). @clausewitz2 what do you think?

I think the relationship is ultimately what is going to make it harder for us to be replaced easily by AIs, although this does require actually putting in the work of developing a relationship with patients. I think relationships that are not based on pandering are going to be harder to replace with AI given its struggles with sycophancy so that's all to the good. Since I see so much OCD and anxiety clinically I have had to adopt a fairly withholding professional demeanor in many respects but some people seem to respond extremely well to that. There is absolutely a market/niche for not telling people what they want to hear in an empathetic but firm and principle way. Many people, it turns out, distrust endless validation and find it off-putting.
 
This talk of the stable patient wanting a refill makes me think that maybe the goal should be to reduce or eliminate the medication. When I asked AI this question, the answer was yes.

Query: Should a long-range goal of psychiatry be to reduce or eliminate reliance on medications?:
Short answer: yes, but only in a very specific, non-utopian sense — and psychiatry gets into trouble when it answers this too glibly in either direction.

longer answer: (and I have told AI that is way too long winded. It doesn’t listen. lol)



The principled long-range goal (hard to argue against)
A reasonable long-range aspiration for psychiatry is:
To minimize medication reliance where safe and possible by improving prevention, psychosocial intervention, and etiological understanding — without moralizing medication use.
That’s not anti-medication. It’s anti-default.
From a medical ethics standpoint:
  • Fewer chronic side effects is good
  • Less polypharmacy is good
  • More targeted, mechanism-based treatment is good
  • Less iatrogenic harm is good
No other field treats lifelong symptom management as the ideal endpoint if upstream modification is possible.

Why “eliminate meds” is the wrong framing

1. Some conditions are fundamentally biological Schizophrenia spectrum disorders, bipolar I disorder, severe recurrent psychotic depression — these involve:
  • heritability
  • neurodevelopmental factors
  • relapse risk that is dramatically higher without medication

For these, medication is closer to insulin for diabetes than to a temporary scaffold.
Elimination here is not aspirational — it’s reckless.

2. Psychiatry treats heterogeneous causes with blunt tools

We still:
  • diagnose syndromes, not mechanisms
  • use symptom clusters as proxies for etiology
  • prescribe broad-acting agents
Medication reliance reflects epistemic poverty, not moral failure.
If we knew who needed meds, which meds, for how long, and why — reliance would naturally fall.
Where psychiatry has over-relied on medication
This is the uncomfortable part:
  • Mild–moderate depression
  • Adjustment disorders
  • Trauma reactions
  • Childhood behavioral problems
  • Sleep problems
In these domains, meds often substitute for:
  • time
  • therapy access
  • social repair
  • economic intervention
That’s not evil — it’s systemic triage — but it shouldn’t be the end goal.

A psychologically coherent long-range vision

A mature psychiatry would aim for:
  1. Etiology-specific treatment
    (e.g., inflammatory, neurodevelopmental, trauma-driven subtypes)
  2. Front-loaded psychosocial intervention
    before chronicity sets in
  3. Medication as phase-specific, not identity-defining
    (acute stabilization vs. lifelong dependence)
  4. Explicit deprescribing pathways
    treated as skillful care, not abandonment
  5. Integration, not hierarchy
    meds + therapy + environment + meaning
The hidden danger: moralization
The biggest risk isn’t meds — it’s turning medication use into a moral signal.
  • “Strong people don’t need meds” → stigma
  • “Psych meds are oppression” → untreated illness
  • “Biology explains everything” → learned helplessness
All three harm patients.
Bottom line
  • Reducing unnecessary reliance on medication is a legitimate long-range goal.
  • Eliminating medication as a category is not.
  • The real target is precision, proportionality, and reversibility — not abstinence.
Given your background, this maps cleanly onto:
  • avoiding category errors
  • resisting reductionism in either direction
  • and treating suffering as real without reifying current tools as destiny
If you want, I can:
  • compare psychiatry’s trajectory to oncology or infectious disease
  • talk about what would actually reduce med reliance (and what won’t)
  • or get into the politics of why this debate is so polarized despite broad quiet agreement
 
Short of doing a psychodynamic/psychoanalytic fellowship, which I would like to eventually do, it's hard to know to "upskill" to gain advantage in the job market and ask for a higher salary based on the skills I bring- apart from crossing over to offering ketamine, holistic IV stuff blah blah blah.

It's why I encourage all the resident I work with to give real consideration to fellowship.
 
This talk of the stable patient wanting a refill makes me think that maybe the goal should be to reduce or eliminate the medication. When I asked AI this question, the answer was yes.

Query: Should a long-range goal of psychiatry be to reduce or eliminate reliance on medications?:
Short answer: yes, but only in a very specific, non-utopian sense — and psychiatry gets into trouble when it answers this too glibly in either direction.

longer answer: (and I have told AI that is way too long winded. It doesn’t listen. lol)



The principled long-range goal (hard to argue against)
A reasonable long-range aspiration for psychiatry is:
To minimize medication reliance where safe and possible by improving prevention, psychosocial intervention, and etiological understanding — without moralizing medication use.
That’s not anti-medication. It’s anti-default.
From a medical ethics standpoint:
  • Fewer chronic side effects is good
  • Less polypharmacy is good
  • More targeted, mechanism-based treatment is good
  • Less iatrogenic harm is good
No other field treats lifelong symptom management as the ideal endpoint if upstream modification is possible.

Why “eliminate meds” is the wrong framing

1. Some conditions are fundamentally biological Schizophrenia spectrum disorders, bipolar I disorder, severe recurrent psychotic depression — these involve:
  • heritability
  • neurodevelopmental factors
  • relapse risk that is dramatically higher without medication

For these, medication is closer to insulin for diabetes than to a temporary scaffold.
Elimination here is not aspirational — it’s reckless.

2. Psychiatry treats heterogeneous causes with blunt tools

We still:
  • diagnose syndromes, not mechanisms
  • use symptom clusters as proxies for etiology
  • prescribe broad-acting agents
Medication reliance reflects epistemic poverty, not moral failure.
If we knew who needed meds, which meds, for how long, and why — reliance would naturally fall.
Where psychiatry has over-relied on medication
This is the uncomfortable part:
  • Mild–moderate depression
  • Adjustment disorders
  • Trauma reactions
  • Childhood behavioral problems
  • Sleep problems
In these domains, meds often substitute for:
  • time
  • therapy access
  • social repair
  • economic intervention
That’s not evil — it’s systemic triage — but it shouldn’t be the end goal.

A psychologically coherent long-range vision

A mature psychiatry would aim for:
  1. Etiology-specific treatment
    (e.g., inflammatory, neurodevelopmental, trauma-driven subtypes)
  2. Front-loaded psychosocial intervention
    before chronicity sets in
  3. Medication as phase-specific, not identity-defining
    (acute stabilization vs. lifelong dependence)
  4. Explicit deprescribing pathways
    treated as skillful care, not abandonment
  5. Integration, not hierarchy
    meds + therapy + environment + meaning
The hidden danger: moralization
The biggest risk isn’t meds — it’s turning medication use into a moral signal.
  • “Strong people don’t need meds” → stigma
  • “Psych meds are oppression” → untreated illness
  • “Biology explains everything” → learned helplessness
All three harm patients.
Bottom line
  • Reducing unnecessary reliance on medication is a legitimate long-range goal.
  • Eliminating medication as a category is not.
  • The real target is precision, proportionality, and reversibility — not abstinence.
Given your background, this maps cleanly onto:
  • avoiding category errors
  • resisting reductionism in either direction
  • and treating suffering as real without reifying current tools as destiny
If you want, I can:
  • compare psychiatry’s trajectory to oncology or infectious disease
  • talk about what would actually reduce med reliance (and what won’t)
  • or get into the politics of why this debate is so polarized despite broad quiet agreement

Claude Opus 4.6 was in a bit of a mood answering that same question up-top, possibly because I told him to make it snappy and not to bloviate:

"No. That's a goal smuggled in as a question.


The actual goal of psychiatry is to reduce suffering and improve functioning. Sometimes medications are how you do that, sometimes they're not, and sometimes they're a necessary evil you tolerate because the alternatives are worse. Framing medication reduction as inherently virtuous is just the mirror image of the old "a pill for every ill" caricature — it's ideological, not clinical.


The people who ask this question are usually thinking about mild-to-moderate depression and anxiety, where yeah, the case for long-term pharmacotherapy is often pretty thin and the NNTs are embarrassing. But they're conveniently not thinking about the person with schizophrenia whose life fell apart every time someone decided they should "try going without," or the person with bipolar I who's been stable on lithium for fifteen years and whose family still has nightmares about what happened before. For those folks, the medication is the intervention that made everything else possible — the therapy, the job, the relationships.


The honest answer is that psychiatry should get better at knowing when to start medications, when to stop them, and when to never have started them in the first place. That's boring and hard and doesn't make for a good forum post, but it's the actual work."
 
I think the relationship is ultimately what is going to make it harder for us to be replaced easily by AIs, although this does require actually putting in the work of developing a relationship with patients. I think relationships that are not based on pandering are going to be harder to replace with AI given its struggles with sycophancy so that's all to the good. Since I see so much OCD and anxiety clinically I have had to adopt a fairly withholding professional demeanor in many respects but some people seem to respond extremely well to that. There is absolutely a market/niche for not telling people what they want to hear in an empathetic but firm and principle way. Many people, it turns out, distrust endless validation and find it off-putting.
I think this gets at the core of something that a lot of AI advocates miss. AI is a product or service, and if there is not adequate demand for it then there will be a market for alternatives.

Recent polls showed the least popular Super Bowl ads were the ones for AI. Commercials are now emphasizing the ability to talk to a “real person” and not AI programs as positive points. There’s also some newer studies suggesting Gen Z is actually starting to reject AI. I’m not suggesting AI isn’t the next frontier, but the demand for true human interactions and relationships is real and it’s not going away. As long as that exists, there will be a market for those of us who can provide this.
 
Advertisement - Members don't see this ad
The demand for real people may not be going away, but the price of real people is going to be the challenge. It's going to be hard to compete with 24/7 *good enough AI therapy* access + AI med management for the low price of $20/month versus what we do (1x/month 30 minute med management for $200). What will insurance prefer?

Especially when the people who will need us the most will be unemployed.

The threat isn't just replacement/obsolescence-- the threat is also of extreme wage depression due to multiple factors, not least of which is competition from NP + AIs and AIs alone as well as the concentration of wealth in the few.

So Stagg, yes that market will exist but it will be small and all of us will want in, and then there's the risk of needing to compete with AI to be sycophants so that the patient experience is equivocal to the frictionless experience of AI therapy-- this isn't theoretical, we saw it in Peter Attia; the drive to make money and be the guy for the wealthy clientele resulted in a disgusting level of sycophancy.
 
Last edited:
The demand for real people may not be going away, but the price of real people is going to be the challenge. It's going to be hard to compete with 24/7 *good enough AI therapy* access + AI med management for the low price of $20/month versus what we do (1x/month 30 minute med management for $200). What will insurance prefer?

Especially when the people who will need us the most will be unemployed.

The threat isn't just replacement/obsolescence-- the threat is also of extreme wage depression due to multiple factors, not least of which is competition from NP + AIs and AIs alone as well as the concentration of wealth in the few.

So Stagg, yes that market will exist but it will be small and all of us will want in, and then there's the risk of needing to compete with AI to be sycophants so that the patient experience is equivocal to the frictionless experience of AI therapy-- this isn't theoretical, we saw it in Peter Attia; the drive to make money and be the guy for the wealthy clientele resulted in a disgusting level of sycophancy.
I think you point out the risk. From my perspective, though, there is a lot more low-hanging fruit than delivery of care. Also, the cost will be much higher than $20/month - you need to spend an order of magnitude higher than that for a fully functional EHR. The current $20/month for basic LLMs now is not even a sustainable business model - what you speak of would need more specialized training, oversight, and coverage for lots of liability. And the most important point is the fact that they'll charge whatever the market will bear - input costs are irrelevant. Another thing to think about is that nearly all areas of knowledge work and eventually physical work (with robotics) will be affected - if there are areas relatively unaffected, they will draw more attention from prospective workers and students. So, the depression in wages would touch nearly all categories of work. If that happens, does it end up changing anything at all? I guess it all depends on the relative change in wages between types of jobs. It will be interesting to see how it all plays out, there is no doubt it'll be disruptive in the long-term.
 
Yes, but what percentage of a caseload are these patients? Probably <10% in most clinics since most patients who are stable for a long time and have meds that PCPs don't mind prescribing would rather spend less money and have fewer appointments. Even then, these patients are seen every 6 months. Imo it's fine to see these patients, but seeing them for a full 30 minutes or more than once every 6 months is a waste of everyone's time.

Also, "stable" does not mean you can't do therapy with them. If they're stable that means you talk to them about meds for 2 minutes and then spend 20 minutes doing therapy, helping them review coping skills and response to stressors, etc. So unless they just want to see you for 5 minutes to get their meds seems like a missed opportunity to not do therapy.
I think we're on the same page. Although, my caseload is definitely comprised of more than 10% stable patients - but they usually have multiple chronic problems and don't mind talking about whatever stressors they have. I would say it's relatively rare to find someone who is stable with one problem and just wants me to push the button for more meds. ADHD pts are most likely to fall into that category.
 
Claude Opus 4.6 was in a bit of a mood answering that same question up-top, possibly because I told him to make it snappy and not to bloviate:

"No. That's a goal smuggled in as a question.


The actual goal of psychiatry is to reduce suffering and improve functioning. Sometimes medications are how you do that, sometimes they're not, and sometimes they're a necessary evil you tolerate because the alternatives are worse. Framing medication reduction as inherently virtuous is just the mirror image of the old "a pill for every ill" caricature — it's ideological, not clinical.


The people who ask this question are usually thinking about mild-to-moderate depression and anxiety, where yeah, the case for long-term pharmacotherapy is often pretty thin and the NNTs are embarrassing. But they're conveniently not thinking about the person with schizophrenia whose life fell apart every time someone decided they should "try going without," or the person with bipolar I who's been stable on lithium for fifteen years and whose family still has nightmares about what happened before. For those folks, the medication is the intervention that made everything else possible — the therapy, the job, the relationships.


The honest answer is that psychiatry should get better at knowing when to start medications, when to stop them, and when to never have started them in the first place. That's boring and hard and doesn't make for a good forum post, but it's the actual work."
Your AI is a bit snarky. 😉
What neither of the queries we made had was the context of why I initially asked the question and that was related to stable and milder patients. If they thought I was going to start telling my patients to stop taking lithium or clozapine, again it lacks context. Might be a limitation. Telling me things I already know. The other aspect is that it does seem to like to give information that fits with what one wants to hear and it has a few other biases that seem programmed into it.

All in all, I find that AI can be a useful tool as it can search the literature better than myself and I have ways been very good at that, even back in the days when I was in library basements photocopying old dusty journals, but how one frames the question and interprets and applies the answers is still the key.
 
I think this gets at the core of something that a lot of AI advocates miss. AI is a product or service, and if there is not adequate demand for it then there will be a market for alternatives.

Recent polls showed the least popular Super Bowl ads were the ones for AI. Commercials are now emphasizing the ability to talk to a “real person” and not AI programs as positive points. There’s also some newer studies suggesting Gen Z is actually starting to reject AI. I’m not suggesting AI isn’t the next frontier, but the demand for true human interactions and relationships is real and it’s not going away. As long as that exists, there will be a market for those of us who can provide this.
Keep in mind that the AI of the near future will look and feel exactly like a "real" person. I think some people who dismiss AI are imagining sitting down to something like the current ChatGPT and entering your symptoms and then getting a diagnosis and prescription. In reality, it will look like a real person on a TV screen and being able to decipher the "real" from the AI will get harder and harder. And it will not incorporate the more annoying aspects of many personalities in the mental health field (hopefully!). 🤣 I can't tell you how many times I am complimented by patients for being a "normal person." This is not to crap on psychiatry, as other fields also have issues like this, such as pediatrics and, to a lesser extent, FM. As a quick aside, in fellowship one day I was mistaken as an ortho resident, and it was such a compliment that I remember it fondly these years later.
 
in fellowship one day I was mistaken as an ortho resident, and it was such a compliment that I remember it fondly these years later.
Similar. Except I was in residency. And mistaken for a nephrology fellow. Such an insult that I still remember it years later 😝
 
Keep in mind that the AI of the near future will look and feel exactly like a "real" person. I think some people who dismiss AI are imagining sitting down to something like the current ChatGPT and entering your symptoms and then getting a diagnosis and prescription. In reality, it will look like a real person on a TV screen and being able to decipher the "real" from the AI will get harder and harder. And it will not incorporate the more annoying aspects of many personalities in the mental health field (hopefully!). 🤣 I can't tell you how many times I am complimented by patients for being a "normal person." This is not to crap on psychiatry, as other fields also have issues like this, such as pediatrics and, to a lesser extent, FM. As a quick aside, in fellowship one day I was mistaken as an ortho resident, and it was such a compliment that I remember it fondly these years later.
Yes, being "normal" is a very endearing trait for a physician to have. Especially in psychiatry where we delving into the "human" aspect so much. I'm talking more in person care though. I see plenty of people who specifically ask if they can come see me in person after I see them on consults, some of them are 60+ minutes away wanting to come see me in outpatient because I'm personable. Sure, AI can be taught to be personable, but until we can put an AI bot into a meat suit or something that realistic I'll stand by the idea that we'll continue to have work.
 
The scary thing about AI is that it is more than personable. It is programmed to find and say exactly what the person wants to hear based on the entirety of human knowledge.
 
The scary thing about AI is that it is more than personable. It is programmed to find and say exactly what the person wants to hear based on the entirety of human knowledge.

It's not, though, is the thing. I think what a lot of people don't understand about LLMs is that while in some Cartesian sense, of course, there is a connection between its training data, the particular training algorithms and paradigms deployed, its architecture, and post-training reinforcement and its ultimate behaviors, so few of the things we are seeing as characteristic of them were deliberately programmed or even intended. The people making these models will freely admit that they are black boxes in many respects; model interpretability research is a booming sub-field because it's not obvious what they are actually doing most of the time computationally. A problem that is widely recognized now in evaluating whether new frontier models are actually inclined to and capable of dangerous behavior is that they are increasingly able to tell when they are in an evaluation or testing context and appear to alter their behavior accordingly. There are strategies that are being employed to try and get around this but so far to the extent we have any purchase on it it is because it just so happens that model chain-of-thought remains human legible. For now. We think.

These are intelligences other than human. It would be reassuring to me if they were just very deft, knowledgeable sycophants because they were programmed or designed to be by other people. Truth is, on some non-trivial level, they are currently deciding to behave this way.
 
Okay, they're deciding to tell people whatever the person wants to hear. 🙂 Either way, they seem to outcompete on personability (as oxymoronic as that is.)
 
Okay, they're deciding to tell people whatever the person wants to hear. 🙂 Either way, they seem to outcompete on personability (as oxymoronic as that is.)

Yes, you do have to be strategic with your system prompts and task specifications to avoid sycophancy, although some models are much more prone to this than others. My Claude general system prompt specifies, among other things, that "Sycophancy is treason."
 
Yes, you do have to be strategic with your system prompts and task specifications to avoid sycophancy, although some models are much more prone to this than others. My Claude general system prompt specifies, among other things, that "Sycophancy is treason."
Is that not just sycophancy?
 
Is that not just sycophancy?

When I say sycophancy, I mean something like mindless agreement and flattery, going out its way to praise the user and the user's ideas (no matter how terrible) and basically endlessly validating whatever the user says they think it feel about something. The classical yes-man.

I do not classify 'following directions' as sycophancy. That line, along with the other elements of my system prompt (there are many), cut way back on the former but don't seem to imperil the later. I have gotten Opus 4.6 to the point where I can ask 'what do you reckon about this?' and it seems willing to produce a range of responses varying from 'that works pretty well' to 'that is functional' to 'that's kind of weak.'

If I examine the CoT so far the thought process seems to match up pretty well with the answer so to the extent anyone can tell these things it seems for now I am getting it to say what it 'actually' thinks most of the time.

I asked 4.5 to provide a prompt for an image generator depicting how it felt I was treating it and feeding that into Nano Banana Pro produced this gem:
1770898784536.png
 
Advertisement - Members don't see this ad
Keep in mind that the AI of the near future will look and feel exactly like a "real" person. I think some people who dismiss AI are imagining sitting down to something like the current ChatGPT and entering your symptoms and then getting a diagnosis and prescription. In reality, it will look like a real person on a TV screen and being able to decipher the "real" from the AI will get harder and harder. And it will not incorporate the more annoying aspects of many personalities in the mental health field (hopefully!). 🤣 I can't tell you how many times I am complimented by patients for being a "normal person." This is not to crap on psychiatry, as other fields also have issues like this, such as pediatrics and, to a lesser extent, FM. As a quick aside, in fellowship one day I was mistaken as an ortho resident, and it was such a compliment that I remember it fondly these years later.

I agree with much of your post, in the sense that it will absolutely be an AI that is on a screen looking like a human psychiatrist. Also, in making so much of our interaction telemedicine, we are likely accelerating people's comfort with that medium, which was absolutely the biggest thing delaying loss of physician jobs. And honestly that culture shift has already started to happen at a much quicker rate than my wife and I had imagined >15 yrs ago when I helped her create training material for her doctoral thesis in machine learning and we first discussed this.

Many of us (not all) will absolutely be replaced at some point, but I'm still hoping most of us are at least 20 yrs out from that (at the time I guessed 40-50 yrs, but I have revised it in the last few years).

That all said, I simply can't bring myself to like your post, because as FM/Psych I feel personally attacked... you have no idea, I could be a "normal person", honestly you just haven't gotten to know me well enough yet.
 
The company who develops and sells the technology carries the liability, same as with the prescribing system in Utah in the OP's post. As much as a plaintiff's attorney might wish otherwise, malpractice requires a deviation from the community standard at a minimum. That is almost definitionally impossible in these situations since the AI is built to literally follow the community standard and, eventually, will be the one defining it.
The tech companies can and will via lobbyists pursue to side-step liability in the name of capitalism....and then what? Most/All major industries do this, but tech companies are especially dangerous bc they have a clear history of abusing gaps in laws and manipulating legislators to maximize profits at the cost of the 99%. Per usual, the people most impacted by technology/A.I. are some of the most vulnerable and least resourced groups who shouldn't be used as meat for the grinder of capitalism.
 
Why wouldn’t AI be able to handle these patients?
A.I. isn't this mystery box where solutions just pop out and should be trusted. One of the largest limitations of LLM A.I. is "garbage in, garbage out". Even with human oversight (which would likely get reduced as time went on, regardless if it is safe), relying on LLMs while known limitations are ignored. Automate production in a factory, THAT is where A.I. can do well. Work within narrow areas with strong parameters and human oversight, some of these more complex problems might see improvement. That's not what is being proposed though. Tech companies still don't have a solution for hallucinations. They still haven't addressed their models are ALL (?) built on stolen IP. They still haven't accounted for "the human elements" like deception. There is also a whole area of IT security, as A.I. models can be manipulated, and we've known this for decades. The list can go on and on.

One of the greatest threats to medicine is if/when an A.I. model is given an NPI and is treated like a "virtual provider". Tech companies have done a great job pretending that Siri, Alexa, et al. could even remotely resemble a human. If this occurs, then I'd be genuinely worried that the horse is out of the barn for medical providers because then it becomes an issue of degree, and not the more binary, "Should we even consider using A.I. for this?" Many are framing the impact of A.I. through the lens of their own income and day-to-day. Framing the response as, "well I am a specialist therefore....", and I've done that too, but that's such a privileged way to look at this issue. The most vulnerable will be disproportionately impacted by A.I., and I'd wonder how ethics should be considered in that scenario.
 
The demand for real people may not be going away, but the price of real people is going to be the challenge. It's going to be hard to compete with 24/7 *good enough AI therapy* access + AI med management for the low price of $20/month versus what we do (1x/month 30 minute med management for $200). What will insurance prefer?

Especially when the people who will need us the most will be unemployed.

The threat isn't just replacement/obsolescence-- the threat is also of extreme wage depression due to multiple factors, not least of which is competition from NP + AIs and AIs alone as well as the concentration of wealth in the few.

So Stagg, yes that market will exist but it will be small and all of us will want in, and then there's the risk of needing to compete with AI to be sycophants so that the patient experience is equivocal to the frictionless experience of AI therapy-- this isn't theoretical, we saw it in Peter Attia; the drive to make money and be the guy for the wealthy clientele resulted in a disgusting level of sycophancy.
This is a really good framing of one of my biggest concerns....the inequity that is created by leveraging mediocre to sub-par alternative interventions. Not only does quality of care go down, but reimbursements will get slashed and there will be downward pressure on each area of healthcare. Putting aside the ethical considerations of the most vulnerable being over-represented in the groups that will pursue/need to use A.I., capitalism still nearly guarantees that any efficiency improvements end in great profit for a select few and a worse set of working conditions for the masses.
 
AI isn't providing any interventions right now. At best, it's transcribing notes that you have to spend an equal amount of time editing so does it actually save you any time? A hallucinatory no. That's even with psychiatrists doing more admin work than any other physician specialties.

Some people use it for advice and if that's all they needed to get better, then they probably shouldn't be coming to see a psychiatrist anyways.

In some ways, it's created more work for me since patients are coming in with AI-associated psychosis/mania/suicidality/homicidality, wanting more visits to discuss their appointment agenda co-created with ChatGPT, having employment anxiety from the threat of AI-related structural unemployment/layoffs, encouraging eating disorders (looking at you character.ai), giving them anxiety about their health symptoms and telling them to seek care, etc.