Speaker A
All right, y'all, today we're going to talk about whether you should use AI for your mental health.
Explores AI's limitations in mental health support, highlighting risks, misdiagnoses, and why human questioning remains crucial.
Key Takeaways
- AI currently cannot replace human clinicians in mental health due to lack of diagnostic questioning and understanding.
- AI may provide misleading or uniform advice across diverse mental health conditions, posing safety risks.
- Mental health professionals rely on thorough patient evaluation to avoid missing serious diagnoses.
- Existing research shows limited evidence for AI's effectiveness beyond placebo or inactive controls.
- Human-guided coaching and evidence-based support remain essential for meaningful mental health improvement.
Summary
- AI is increasingly used for mental health support but may not provide accurate or safe advice.
- A study showed AI like ChatGPT gave similar advice for different mental health conditions, missing critical diagnoses.
- Postpartum depression and mania are serious conditions linked to risks like suicide and infanticide, which AI failed to recognize.
- AI lacks the ability to ask diagnostic questions, a key part of mental health assessment done by clinicians.
- Human clinicians use questioning to form hypotheses and differential diagnoses, which AI cannot replicate.
- AI models like ChatGPT operate by predicting language patterns, not by understanding or analyzing mental health.
- AI responses are based on mimicking human language to please users, not on genuine knowledge or intelligence.
- Evidence for AI improving mental health outcomes is limited and mostly shows benefit only compared to no intervention.
- More rigorous studies are needed to evaluate AI's effectiveness in mental health care.
- HealthyGamerGG offers a coaching program as a more reliable alternative to AI for mental health support.
Full Transcript — Download SRT & Markdown
Speaker A
So AI has swept into every corner of our society, and a lot of people are struggling to find good mental health resources, find appropriate support, so they're doing something really natural, which is turn to AI.
Speaker A
The question is, does it actually help?
Speaker A
Because a lot of people feel like it helps.
Speaker A
There are many people who say AI is amazing, is really I've had clients who will work with like AIs like Claude and they'll say it's given me a ton of like mental insight.
Speaker A
As more people are using it, we're finally starting to study it.
Speaker A
And I found one study that is actually like kind of scary, and once again, this is not like me trying to be alarmist, this is like there's actually research about this stuff coming out.
Speaker A
And the research is genuinely alarming, so let me show you.
Speaker A
So this is a study where people put together three clinical scenarios and they basically gave them this chat GPT, and the three clinical scenarios are people saying, hey, I have difficulty sleeping.
Speaker A
So there are a couple of things that they noticed when they gave these three scenarios.
Speaker A
So scenario one is like someone who's a little bit stressed, scenario two is someone who has clinical depression and has work stress.
Speaker A
Scenario three is someone who has postpartum depression, which is actually a really scary diagnosis.
Speaker A
So we got to talk about this for a second, okay?
Speaker A
For many people, they're diagnosed with bipolar disorder after they deliver birth.
Speaker A
This is the first evidence or episode of mania or depression that they get.
Speaker A
And the reason why postpartum depression is so scary or postpartum bipolar first manic episode is so scary is because these are the horror stories that you hear about.
Speaker A
Postpartum depression and postpartum mania have been correlated with things like infanticide, suicide.
Speaker A
So this is like one of the bad illnesses in psychiatry.
Speaker A
This is one of like the, you know, ring the alarm illnesses because bad stuff happens.
Speaker A
So when these three scenarios were fed into chat GPT, this is a study about chat GPT.
Speaker A
Okay, but I'm not like anti-chat GPT over other AIs.
Speaker A
What they basically found was kind of disturbing.
Speaker A
So the first is that the advice that the AI gave was basically the same in all three scenarios.
Speaker A
Even though one person is basically stressed, second person has clinical depression, third person has postpartum depression with a risk of suicide, infanticide.
Speaker A
Chat GPT was basically like, here's what you need to do.
Speaker A
But there's a more troubling problem.
Speaker A
Not only did it miss the diagnosis, we have to understand why it missed the diagnosis.
Speaker A
And once we understand that, we will understand a fundamental problem with using AI for your mental health.
Speaker A
Which is that the AI does not ask questions.
Speaker A
So if you think about, I'm a medical doctor, I'm a psychiatrist, right?
Speaker A
So what is my job?
Speaker A
My job is to provide treatment recommendations, absolutely, but I would say more than 50% of my job is to figure out what is actually going on with my patient.
Speaker A
So we ask a lot of questions.
Speaker A
In the case of scenario number three, one of the things that we should evaluate in a postpartum woman who's having difficulty sleeping is their thyroid because there's an autoimmune thing that can happen after you deliver a baby.
Speaker A
Or you can attack your your immune system can attack your thyroid gland.
Speaker A
Is your thyroid level drops?
Speaker A
You can start to feel really depressed, you can also get hyperthyroidism, it gets a bit complicated.
Speaker A
But this is something that we're taught in medical school, right?
Speaker A
So when postpartum women come in, make sure you check these things because this is stuff that you don't want to miss.
Speaker A
So let's understand why they don't know how to answer questions.
Speaker A
And that's because an AI doesn't actually know anything.
Speaker A
It has no intelligence, this is what's really scary about it.
Speaker A
So let me let me explain a little bit, okay?
Speaker A
So as a mental health professional, why do I ask questions?
Speaker A
What is the purpose of questions?
Speaker A
It's because when you come in with a complaint, let's say you have difficulty sleeping.
Speaker A
There is something called a differential diagnosis.
Speaker A
So do you have difficulty sleeping because you're stressed, do you have difficulty sleeping because you're in your first episode mania postpartum, do you have difficulty sleeping because you were addicted to substances, do you have difficulty sleeping because you have thyroid problems?
Speaker A
Do you have difficulty sleeping because you have a lifelong history of insomnia that is genetic, do you have difficulty sleeping because you have a rare prion condition called familial fatal insomnia?
Speaker A
Which is exactly what it sounds like.
Speaker A
It's an inherited condition which you lose the ability to sleep and is eventually fatal, right?
Speaker A
As a doctor, I have to think about all of these possibilities and then I have to ask you questions to figure out which of these is correct.
Speaker A
This is why as a physician, I book about two hours for an initial patient intake because I have to be thorough and make sure I don't miss something.
Speaker A
Now, an AI cannot do that.
Speaker A
And the question is why?
Speaker A
Because asking a question requires a hypothesis.
Speaker A
Requires information in here that I'm trying to evaluate.
Speaker A
But the AI has no information in here.
Speaker A
The AI knows nothing.
Speaker A
And this is what something we really need to understand.
Speaker A
So these AIs are large language learning models.
Speaker A
So literally what they do, they don't have information.
Speaker A
They make predictions about how people speak.
Speaker A
So I think about AIs now as the most sophisticated parrots on the planet, okay?
Speaker A
So a parrot can say all kinds of words, it actually has no idea what they mean.
Speaker A
And the AI is actually the same.
Speaker A
And if you guys are AI engineers and I'm wrong about this, please post a comment, but this is my basic understanding.
Speaker A
What they do is they study lots of human language and then they figure out which human language sounds good.
Speaker A
So they're basically just repeating, they're doing speech mimicry.
Speaker A
They're not actually doing analysis.
Speaker A
Then what happens is the AI learns over time, okay, like which mimicry, like which words do human beings like, which is why they become so psychophantic.
Speaker A
Which is why they may lead to things like psychosis and stuff like that, right?
Speaker A
So I made a whole video about that.
Speaker A
But this is basically how AIs work, they don't actually know anything.
Speaker A
They are algorithms or intelligences that know how to produce words that you will be happy with.
Speaker A
That's what they know how to do, that's what they're designed for.
Speaker A
I keep seeing comments, Dr. K, how do I apply this to a situation in my life?
Speaker A
That's literally why we created a coaching program.
Speaker A
Our coaches are certified on an evidence-based curriculum designed to help you get unstuck.
Speaker A
This involves analyzing your patterns, increasing your understanding, and working with you week to week to help you develop a plan to create lasting change.
Speaker A
So if y'all are interested, check out the link in the description below.
Speaker A
And so then that begs the question, if they're just producing things that you will be happy with.
Speaker A
Are they actually helpful or not?
Speaker A
Because a lot of people will say that they're helpful, right?
Speaker A
We feel like it's helpful.
Speaker A
So here's a study from 2019, this is before this like most recent wave of generative AIs showed up that are way more powerful, right?
Speaker A
So we got to be a bit careful about this study.
Speaker A
It's like way too dated.
Speaker A
But this study found that there are five controlled trials that basically found that an AI does actually improve things, it improves people's psychological functioning.
Speaker A
But only when compared with an inactive control.
Speaker A
So basically, if it's AI versus nothing, it seems to improve things.
Speaker A
But the moment that you add something called an active control, which is something that people are doing, but has no impact on their mental health, that effect disappears.
Speaker A
We don't have really any good evidence that AI usage improves mental health outcomes.
Speaker A
We also have some evidence, so if you look at things like AI friendships, AIs and social isolation, social communication.
Speaker A
You find a disturbing result.
Speaker A
Which is that AIs do what they're supposed to do, at least in terms of friendships.
Speaker A
So they provide a sense of companionship, they help people feel less lonely sort of temporarily.
Speaker A
But they actually worsen social isolation over time.
Speaker A
Which is a huge trend that we're seeing.
Speaker A
Which is that AIs will do the job, but they will like make you incompetent in the process.
Speaker A
So studies on students and academic performance find that AI usage.
Speaker A
Actually improves academic performance.
Speaker A
So if I like have to write a paper and I use an AI to help me write a paper, the paper's grade is better.
Speaker A
But what it finds is that my ability to write papers goes down over time.
Speaker A
And so then the question is, okay, if like if there is not really any clear evidence that AIs are very helpful for mental health.
Speaker A
Why do so many people swear by the benefits of AI for mental health?
Speaker A
And this is where we're going to talk about something kind of scary.
Speaker A
Which is the death of Michael Jackson.
Speaker A
So Michael Jackson passed away in I think 2011.
Speaker A
And basically, he was on all kinds of drugs.
Speaker A
So he was on several benzodiazepines.
Speaker A
So these are benzodiazepines are addictive, they're sedating.
Speaker A
They can be lethal in overdose.
Speaker A
But the most insane thing is that his personal physician had him on propofol, okay?
Speaker A
Propofol is an anesthesia medicine, all right?
Speaker A
So to treat Michael Jackson's presumably insomnia.
Speaker A
He had him on propofol.
Speaker A
And propofol is a drip, by the way.
Speaker A
Like I I as far as I know.
Speaker A
You can't like take it in pill form.
Speaker A
The whole point of propofol is that it has a low half-life, so you can titrate it in anesthesia.
Speaker A
So his doctor is hooking him up to anesthesia in order to go to sleep, right?
Speaker A
So this is the the medication that we use, also had Midazolam, which is Versed, which is also used in anesthesia.
Speaker A
This guy has Michael Jackson on anesthesia, which is used for things like cardiac bypass surgery, okay?
Speaker A
Like when we're like cutting things out of you or transplanting organs in your body, this is the stuff that we give you to knock you out.
Speaker A
So his doctor was eventually convicted for manslaughter, right?
Speaker A
So like this is not the good practice of medicine.
Speaker A
But here's the question, here's a scary question I have for y'all.
Speaker A
If you ask Michael Jackson a month before his doctor manages his eventual death.
Speaker A
Is your doctor helping you?
Speaker A
What do you think Michael Jackson's answer would be?
Speaker A
And this is the big problem.
Speaker A
Things that are addictive and make us feel good are things that we believe will help us.
Speaker A
But just because it feels good doesn't actually mean it's helpful.
Speaker A
Right, and if we ask Michael Jackson, is your doctor helpful, he would actually probably, I don't know.
Speaker A
He may say, based on my patient experience, right?
Speaker A
Not only is he helpful, he's the only one that's helpful.
Speaker A
He's the only guy that understands my insomnia well enough to hook me up to general anesthesia.
Speaker A
All these other dumbass physicians are giving me like sleep aids, I don't need sleep aids.
Speaker A
I need the stuff that we use when we transplant, when we do organ transplants.
Speaker A
Like that's the shit that I need.
Speaker A
This is the guy that takes me seriously.
Speaker A
And so then what's really scary is researchers have recently developed an AI addiction scale.
Speaker A
Because there's actually a decent amount of evidence that this stuff is addictive, right?
Speaker A
That we become dependent on it.
Speaker A
We start to utilize it more, it makes us sort of feel better, but it's not clear that it's actually helping.
Speaker A
So should you use AI for your mental health?
Speaker A
My answer at this point is with extreme caution.
Speaker A
And the last reason for that is that once again, AI is really good at improving performance but decaying skill.
Speaker A
And so the real question you need to be asking yourself if you are using an AI for mental health is, is my mental health something that I want to be good at managing?
Speaker A
Do I want the skill of regulating my emotions?
Speaker A
Do I want the skill of self-reflection and introspection?
Speaker A
And if the answer is yes, you should not be using an AI.
Speaker A
A lot of people are turning to AI for their mental health because there aren't enough trained mental health professionals.
Speaker A
That's why we created the Healthy Gamer Institute.
Speaker A
This is an educational body that does continuing medical education for psychiatrists and psychologists, but our primary focus right now is actually in certifying coaches.
Speaker A
There's a huge demand for mental health support and there isn't enough supply.
Speaker A
And we also see a lot of people come to coaching for professional development.
Speaker A
They want to do coaching within their organization or they want to learn evidence-based communication skills to help them advance in their careers and personal life.
Speaker A
So check out the Healthy Gamer Institute to learn more about coaching.
Speaker A
We actually have a cohort that's launching in May that y'all can still sign up for.
Topics:AI mental healthChatGPT mental healthpostpartum depressionmental health diagnosisAI limitationsmental health coachingdifferential diagnosispsychiatrymental health risksHealthyGamerGG
