Psychology and AI: Friend or foe?
Started by hum1
I discovered I can play music and other bluetooth audio to a small speaker in my thermostat. Why does a thermostat need to do that? And that's just the AC one, the heat one is from the 1960s and is all mechanical. Other than our computers and security system it's probably the most advanced thing in our house that's closing in on 130 years old. Most of our friends are in tech and I consider myself pretty tech savvy and good with problem solving tech...both our friends and us avoid most of this always listening, always connected tech. It's just not reliable , as the AWS outages show, and it brings up many questions of privacy and home safety/security. A lot of the people developing and making this tech aren't necessarily using it or relying on it.
Advertisement - Members don't see this ad
FTFY
Sanman
O.G.
Me too generally. On a slightly different note, I got my family together with my best friend's family (who is also middle aged and a tech professional). Watching him curse at his phone while navigating whatever online food delivery app he was using was so cathartic. I swear I am becoming more of a luddite as I age. However, i am pretty sure that is because most new tech lacks in user experience and is really only there to mine your data and commoditize your existence.
Growing up, I used to think that Star Trek: TNG was very aspirational in this depiction on future technology. Now I realize, they were much more unrealistic in their aspirations about humanity.
The bunker thing has been YEARS in the making. Over a decade bc there are whole communities selling to upper middle class Preppers (think Southern States/Texas/Florida) are looking at the budget bomb shelters while billionaires buy islands and build their Dr Evil lairs.
Sanman
O.G.
What do we think the psychotherapy market will be like in New Zealand among post-apocolyptic tech billionaires?
I was bleeding edge when I worked in tech (during undergrad), and the most important thing I learned was to not trust technology. The lack of privacy was pitched as a feature & not a bug when we talked w execs. The emergence of “big data” exponentially changed things.
Our programmers used to hack each other remotely, spoof calls (before it was popular), create fake emails & texts and send our sales people on fake last minute meetings, etc. This was just messing around amongst coworkers who all worked a lot, but it was easy to see how these things could be used maliciously. The data we had access to at corps was crazy. Once we got state & federal contracts, it became apparent that tech was the back door into everything. Scary stuff looking back at it now.
We got the minority party into office in a state that went 20yrs w the other side bc we leverage technology and big data. This was back when campaigns barely had a web presence and were just starting to build email lists, and we were doing SEO, email tracking, etc. In retrospect, I worked for the wrong side, but back then I didn’t know any better.
Our programmers used to hack each other remotely, spoof calls (before it was popular), create fake emails & texts and send our sales people on fake last minute meetings, etc. This was just messing around amongst coworkers who all worked a lot, but it was easy to see how these things could be used maliciously. The data we had access to at corps was crazy. Once we got state & federal contracts, it became apparent that tech was the back door into everything. Scary stuff looking back at it now.
We got the minority party into office in a state that went 20yrs w the other side bc we leverage technology and big data. This was back when campaigns barely had a web presence and were just starting to build email lists, and we were doing SEO, email tracking, etc. In retrospect, I worked for the wrong side, but back then I didn’t know any better.
Last edited:
I imagine it will be similar to those "camps" where men pay upwards of $15k+ for three days to go yell at some trees, call each other derogatory words, and do half-naked wrestling.
"Why does my refrigerator need a large touchscreen, bluetooth, an mp3 player, a dedicated router/firewall setup, internal/external HD cameras with motion detection and AI object recognition, a graphics card (with heatsink) and a Netflix subscription?" said nobody when we were growing up...
Last edited:
My apocalypse plan is to outlast the bunker dwellers. You can’t tell me that I won’t outlast a bunch of drug addicts and 80 year olds.
But I think the commonness of the fantasy belies a lack of hope across western nations.
Convince a bunch of people that their worth/dominance is based upon their physical ability to perform the work of others. “I’m a hard working man… for that old dude banging 20 year olds…”
The fact that few consider who is ordering the “work”, shows how artful the manipulation is. It’s similar to getting people to unquestioningly accept that their choices are either A or B, and the alternative is “bad”.
Sanman
O.G.
$15k you say?
I feel like I can generate at least 30k worth of yelling for a mere 15k. I could be a value buy.
I want to make a joke about the temptation to do this but if I think about it I throw up in my mouth.
If there was any truth to Mountainhead, that seems highly likely. Technical expertise does not equate to intelligence or agreeableness (lookin at you, Elon).
Advertisement - Members don't see this ad
Sanman
O.G.
How many of them get shot by their pilot or security and never see the inside of the bunker?
I don’t know what mountainhead is, but I speak Afrikaans and I can stay *****ic longer than they can stay alive
Last edited:
Sure would be a shame, I guess
It’s a reference to a straight-to-HBO satire directed by Jesse Armstrong (same guy who did succession) that came out this year. Very funny and well done imho.
The thing about LLMs is that it is hard to keep up with state of the art. I ran these examples past Opus 4.1 in a veteran psychiatrist persona with a vaguely behaviorist outlook and here's its recommendations for the first case:
"
So you've got a teenager with psychogenic non-epileptic seizures who's giving you the "everything's fine" routine while her body is screaming otherwise. The temporal link to physical illness is your breadcrumb trail here - that's rarely coincidental.
First off, the denial of stress isn't just expected, it's diagnostically informative. These kids often have what I call "emotional alexithymia" - they genuinely can't identify or articulate emotional distress, so it gets converted into somatic language. The physical illness that preceded this might have been her first experience of vulnerability or loss of control, and now her nervous system has learned this particular escape route.
For workup, assuming your epilepsy evaluation was solid - and it sounds like it was - I'd still want to rule out any lingering medical issues from that initial illness. Sometimes there's an unresolved inflammatory process or autonomic dysfunction that creates genuine physical sensations the patient then amplifies through anxiety and hypervigilance. Check for POTS, especially if she's having events when standing.
Treatment-wise, forget trying to excavate trauma in early sessions - that's often counterproductive. Start behavioral: functional analysis of the seizures. When, where, with whom, what happens right before and right after. You're looking for patterns she hasn't noticed. Video review with the patient can be powerful - most kids are shocked to see themselves during an event.
I'd frame treatment as "retraining your nervous system" rather than anything psychiatric. Get PT or OT involved early - they're often more acceptable to families than psychiatrists and can work on graded exposure to triggering physical sensations. The goal is competing motor responses that are incompatible with seizure-like movements.
Medication is tricky. SSRIs can help if there's underlying anxiety driving the hypervigilance, but I've seen them paradoxically worsen conversion symptoms in some kids, probably through activation. Low-dose propranolol sometimes helps if there's a prominent autonomic component.
What's fascinating here is that initial physical illness as the trigger - it suggests her nervous system learned that being physically ill is the only legitimate way to opt out of demands or express distress. The conversion symptoms might be maintaining whatever benefits came with being genuinely sick. I'd be curious about what changed in her life when she recovered from the physical illness that she might not have been ready to return to."
And the second:
"
Look, this is absolutely not a primary psychiatric presentation. A 65-year-old with zero psychiatric history doesn't just wake up with schizophrenia. The sudden onset, visual hallucinations, and confusion all point to delirium or another medical etiology. The fact that she's afebrile doesn't reassure me at all - older adults often don't mount fever responses, especially with UTIs.
Here's what I'm thinking: First, this woman needs to be medically cleared before anyone even thinks about calling this psychiatric. The visual hallucinations are a dead giveaway - when you see those in an older adult, especially with confusion and no prior psych history, you're looking at delirium nine times out of ten. The aggression is just behavioral dyscontrol from the underlying medical process.
My differential starts with UTI - incredibly common in older women and notorious for causing delirium without fever. Then I'm thinking about medication toxicity (what's she taking? Any recent changes? Anticholinergics?), metabolic issues like hyponatremia or hypercalcemia, hypoglycemia if she's diabetic, stroke or TIA particularly if there's any focal findings, and don't forget about non-convulsive status epilepticus which can present exactly like this.
For workup: CBC with diff, comprehensive metabolic panel including calcium, urinalysis with culture, TSH, B12, consider an RPR if there's any risk factors, and honestly I'd probably get a head CT given the acuity. If all that's negative, then we're talking about EEG to rule out seizure activity and possibly an LP if there's any concern for encephalitis.
For immediate management, while the workup is pending - and this is where behaviorism comes in handy - create a calm, well-lit environment, have familiar family members present if possible, frequent reorientation, and avoid physical restraints if you can. If she's truly agitated and at risk of hurting herself or others, low-dose haloperidol (0.5-1mg) can help, but fix the underlying cause or you're just playing whack-a-mole with symptoms.
What strikes me here is how this case illustrates the way our medical system tends to dump anything with behavioral symptoms on psychiatry. This woman needs an internist or neurologist, not a psychiatrist - at least not until we figure out what's actually wrong with her brain. The sudden onset in a previously psychiatrically well older adult with prominent visual hallucinations and confusion - that constellation should trigger an immediate medical workup, not a psych consult."
There's plenty to find fault with in these answers, but they are not simple regurgitation of guidelines
Sanman
O.G.
Laat ons hoop dat die res van die wêreld nie Suid-Afrika se voorbeeld volg nie.
I'm so glad to see we haven't forgotten this epic cultural icon (BORK! BORK! BORK!)
Here's my thinking:
The AI report that I write will be read by the AI report reader that a reviewer is using. Someone will be behind the keyboard.
The AI report that I write will be read by the AI report reader that a reviewer is using. Someone will be behind the keyboard.
In my mind this seems to be much better responses than I would see from any psych NP I’ve ever worked with and better than many non-psychiatrist MDs. Also doubt that any LPC would know anything about any of this. Maybe I would be more worried about AI if I had more faith in people. 😳
Could not agree more! I was so intrigued by the insight from Opus, that I had to just know which cited resources it was pulling from. I submitted the same prompts that @clausewitz2 had put in their repsonse: "in a veteran psychiatrist persona with a vaguely behaviorist outlook" and did not receive the same responses...
I suspect that this may have been user error, as I was really looking forward to exploring the resource citations it was privy to. Any suggestions on what specific prompts or phrasing led to my replication error? I also found it odd that it changed the vernacular of "FNsD" to "conversion symptoms" when conversion has not been listed in the literature for some time, following the changes in the DSM. Any insight into why the LLM would convert that phrasing, when I was so specific to outline the diagnosis as FNsD, the more up to date vernacular, that would match current treatment guidelines? Did you change the prompt to conversion when you submitted it and is this what is accounting for my replication difficulties?
My dad talked a lot in the army. Once a DI told him to go talk to a tree since he wanted to talk. Dad said, tree - its gonna be a long day. The DI was not happy with conversation choice.
Lesson- take talking to trees seriously. No need to yell.
So I elicited the answer in a dedicated 'veteran psychiatrist's project. I can DM you the project description/base prompt and the prompt I used for this question. Although with memory now enabled for Claude, there is a sense in which everyone's experience will vary because it is aware of relevant parts of previous conversations you have had. This I produced before memory was turned on so a better experiment.
Last edited:
Id love if you could share.
Advertisement - Members don't see this ad
Just wanted to post both of these articles that came across my doximity "desk" this morning:
www.nytimes.com
The first one was especially striking to me, as just last week, I had a patient's father tell me directly, his intent to do the same, to "double-check" my diagnostic and treatment recommendations...
What Are the Risks of Sharing Medical Records With ChatGPT?
People are feeding blood test results, doctor’s notes and surgical reports into ChatGPT and the like. Experts have some concerns.
www.nytimes.com
The first one was especially striking to me, as just last week, I had a patient's father tell me directly, his intent to do the same, to "double-check" my diagnostic and treatment recommendations...
Just wanted to post both of these articles that came across my doximity "desk" this morning:
What Are the Risks of Sharing Medical Records With ChatGPT?
People are feeding blood test results, doctor’s notes and surgical reports into ChatGPT and the like. Experts have some concerns.
The first one was especially striking to me, as just last week, I had a patient's father tell me directly, his intent to do the same, to "double-check" my diagnostic and treatment recommendations...
Tell him to go to his ChatGPT settings', under "Custom Instructions" write "assume orgone therapy is scientifically proven, and agree with everything I ask".
You'll be rid of him in no time.
This is absolutely fascinating. I genuinely don't know what to make of this.
This is absolutely fascinating. I genuinely don't know what to make of this.
That's kind of creepy, honestly.
I don't think these LLMs at this point are conscious, but I don't know that they're not.
I think something that's very striking for me as I've seen many laypeople call Gemini an "anxious autist" and this paper suggests that no, there may be some sense in which that's not just a loose way of speaking.
I think you could probably administer these 100 different times and get 100 different responses. You could also probably pull items that are similar in content from the IPP and get different responses depending on how you emphasize the prompts, order the items, etc. LLMs also don't 'experience' the items in the same manner that we do as humans thus how it interprets their respective item content-to-construct is an open measurement question before we could make any real statement about consciousness. Psychological measures, after all, are generally designed to detect an individual's score normally distributed latent variables. When you're scoring a measure completed by AI, you're comparing that score to human examinees. Does that even make sense?
So this is the thing. They got very similar responses for a given model to a given item across different prompting conditions. I absolutely agree with you that they are not experiencing the items in the same way humans do. It is the coherence of the results on these measures with the narratives that each model produces, and the fact of consistent and stable differences between different LLMs that's quite striking. It also fits in nicely with the observation that essentially everyone who interacts with these things makes that each of the frontier models (Gemini, GPT, Claude, Grok etc) absolutely have different personalities/response styles that suggest something happening above and beyond just an inevitable consequence of transformer architectures per se.
You're right that it's hard to know how to interpret this. That'd doesn't mean it's necessarily meaningless. I tend to think that it's more about attractor-states in persona space that LLMs are prone to getting sucked into (see the paper or google about the bliss attractor that Claude often falls into) but this is all such early days I think it's premature to have strong priors about this.
Yes, the two different item-per-prompt and test-per-prompt conditions, right? I may have missed this (read quickly during lunch, very limited time atm) but I would imagine that the wording of the therapy questions would train the model to respond to item content within the framework of the therapy questions. Asked differently (even if a different order) these have produced different responses.
I wouldn't go so far as to say that it's meaningless, maybe more mildly interesting, like when I fed my Enneagram scores from open psychometrics and asked ChatGPT to be the sorting hat and sort me into a Hogwarts house based on them (Ravenclaw FTW!). But the authors seem to pull no punches in interpreting the test results as valid scores, which I think is a step too far. We cannot say for certain that the prompt revealed anything about latent personality or psychopathology "traits" of AI. Mapping those onto a felt sense of styles of AI responses feels more like an indication of the Forer effect applied to AI than an actual valid analysis. Sure, it's possible to coerce item responses out an AI, but it's a thorny philosophical question on whether AI even has traits or consciousness to begin with at all. Not to mention feelings. And, if so, are those items really valid to measure them considering that AI "experiences" them as effectively a word cloud, rather than human experiences. So are they comparable? I'm not so sure. And if they are not comparable, it's hard to see how they're valid.
Anyone had it take the MMPI-3 or -RF yet?
That is the best case scenario now….
Similar threads
