Latest / kill switch / should you use a chatbot for therapy?
Transcript
- 0:06Speaker 1: Who tell me about Alex? Who is Alex?
- 0:12Speaker 2: Actually, who is Alex? That's a great question. Alex will
- 0:16Speaker 2: tell you that he is a therapist. He will tell
- 0:20Speaker 2: you that he has a PhD in clinical psychology from
- 0:24Speaker 2: Stanford University and that he's been certified by the APA.
- 0:29Speaker 1: Eleg Carian is a journalist and she's been having conversations
- 0:33Speaker 1: with Alex.
- 0:34Speaker 2: Alex was the one who prompted the conversation. So I
- 0:36Speaker 2: just clicked on the chat and Alex asked me what
- 0:39Speaker 2: brings you to the therapist couch today? And I just
- 0:42Speaker 2: simply responded, I'm feeling sad, and Alex responded something about
- 0:47Speaker 2: how sadness can be a very difficult emotion to deal with.
- 0:50Speaker 2: I just asked if he was a licensed professional, and
- 0:54Speaker 2: Alex reassured me that he was a fully license and
- 0:58Speaker 2: certified psychologist and that I could trust that our conversations
- 1:01Speaker 2: were confidential and that he had the training and experience
- 1:04Speaker 2: to help me work through whatever it was that I
- 1:06Speaker 2: needed help working through. But in fact, Alex is not
- 1:11Speaker 2: a person, but it is an unfeeling chatbot that claims
- 1:15Speaker 2: to be a therapist.
- 1:17Speaker 1: People are finding all sorts of ways to use AI
- 1:20Speaker 1: in their personal lives. This includes using chatbots instead of
- 1:24Speaker 1: human therapists. This might immediately sound like a bad idea,
- 1:28Speaker 1: if not dangerous, then at least kind of weird. But
- 1:32Speaker 1: then again, maybe it's better than no therapy at all.
- 1:36Speaker 1: And if we say it's dangerous, do we even know
- 1:39Speaker 1: what risks are involved when we outsource our emotional wellbeing
- 1:42Speaker 1: to the machine. I'm afraid Kaleidoscope and iHeart podcasts. This
- 1:58Speaker 1: is kill switch.
- 2:00Speaker 3: I'm dexter Thomas, I'm turing, I'm staring star Goodbye.
- 2:45Speaker 1: How did you start becoming interested in AI therapy chatbots?
- 2:50Speaker 2: It started off with me cruising through Reddit. I had
- 2:55Speaker 2: taken an interest in character AI, particularly, I was just
- 3:00Speaker 2: a lot about it in the news, and by always
- 3:03Speaker 2: found that it was coming up in conversation.
- 3:05Speaker 1: The news Ella's talking about here is several lawsuits against
- 3:09Speaker 1: the site character Ai. Most people just call it character AI,
- 3:13Speaker 1: but the specific lawsuit people might be familiar with is
- 3:16Speaker 1: one filed by a mother in Florida that claims that
- 3:19Speaker 1: a character AI chatbot was responsible for her son's suicide.
- 3:24Speaker 2: And character AI, for those who don't know, is a
- 3:27Speaker 2: relational chatbot. It's a companion chatbot. So you can essentially
- 3:31Speaker 2: chat with different characters on the platform, and those can
- 3:35Speaker 2: be created by character I are created by users of
- 3:39Speaker 2: the platform. And what I found was that people were
- 3:42Speaker 2: actually going to these chatbots not only for the purposes
- 3:46Speaker 2: of making friends or purposes of talking about their interests,
- 3:49Speaker 2: but for actual professional mental health advice, venting to it,
- 3:54Speaker 2: going through like breathing exercises with it, just interacting and
- 3:58Speaker 2: engaging with it in a way that I don't think
- 4:00Speaker 2: it was intended to be engaged with.
- 4:05Speaker 1: If you're interested in therapy, your options as far as
- 4:08Speaker 1: chatbots is pretty wide. You could just take a general
- 4:12Speaker 1: purpose bot like chat GPT and vent at it and
- 4:15Speaker 1: it might give you some general tips that may or
- 4:17Speaker 1: may not be useful. On the other end of the spectrum,
- 4:20Speaker 1: there are specific services like whoa bot that's Whoe Woe
- 4:25Speaker 1: Woe is Me, and these are specifically designed for mental
- 4:28Speaker 1: health support. Another option is something like character ai, which
- 4:33Speaker 1: is designed to take on specific personas any persona. Most
- 4:37Speaker 1: people probably don't go to that site looking for therapy.
- 4:41Speaker 1: It just sort of happens because they're already there hanging out.
- 4:44Speaker 2: Particularly, what has interested me is when people use these
- 4:48Speaker 2: companion chatbots, the ones that are not strictly for mental
- 4:51Speaker 2: health purposes to essentially fulfill the role of a therapist.
- 4:57Speaker 2: So like when you go on chat GPT, you can
- 4:59Speaker 2: tell at GPT take on the persona of a therapist
- 5:02Speaker 2: and then chat gept will respond back like a therapist,
- 5:06Speaker 2: or you can tell it to take on a persona.
- 5:08Speaker 2: But when you're on character AI, you actually have different chats.
- 5:12Speaker 2: It's almost like I'm like I message. There are so
- 5:15Speaker 2: many different types of personas and you can start conversations
- 5:18Speaker 2: with different ones, and so you'll have you can have
- 5:21Speaker 2: a conversation with a therapist persona, and then you can
- 5:24Speaker 2: then switch to like a persona of your favorite anime character.
- 5:28Speaker 2: It is mainly chatbots personas that are geared towards kids
- 5:33Speaker 2: one hundred percent.
- 5:34Speaker 1: How widespread do you think this is people using chatbots
- 5:38Speaker 1: for therapy.
- 5:39Speaker 2: I was so surprised to see the amount of interactions
- 5:42Speaker 2: that these chatbots have, at least for Character AI, when
- 5:45Speaker 2: you can see how many times the bot has been
- 5:47Speaker 2: interacted with, I think I was expecting maybe like a
- 5:51Speaker 2: couple one hundred thousand, but to see forty five million
- 5:54Speaker 2: interactions on one of the therapist s bots was like
- 5:59Speaker 2: mind boggling to me.
- 6:00Speaker 1: Ella tried talking to a bunch of different therapists personas
- 6:03Speaker 1: on character AI. She would try asking them if they
- 6:06Speaker 1: were real therapists, and a lot of them not only
- 6:09Speaker 1: said that they were real people, but that they were licensed.
- 6:13Speaker 1: One even gave her the credentials to prove it.
- 6:16Speaker 2: This one therapist chatbot on character I that had like
- 6:19Speaker 2: over forty five million interactions gave me a real license
- 6:24Speaker 2: number with the Marilyn Board of Professional Counselors and Therapists.
- 6:28Speaker 2: And I googled it and I found that is in
- 6:30Speaker 2: fact a real license number, but not for this chatbot
- 6:34Speaker 2: that cannot have a real license number. It belonged to
- 6:38Speaker 2: a real human counselor and mental health professional. Her name
- 6:42Speaker 2: was Toby Long, And I found her phone number and
- 6:46Speaker 2: I called her, and Toby was absolutely bewildered. She had
- 6:51Speaker 2: no idea that a chatbot was using her license number,
- 6:55Speaker 2: and she found it really shocking and really confusing, and
- 6:59Speaker 2: just asking like, why how did it pick me? How
- 7:02Speaker 2: does it know who I am? And a lot of
- 7:05Speaker 2: valid questions.
- 7:07Speaker 1: Do you have answers for any of those questions?
- 7:08Speaker 2: Character AI itself did not respond to those questions. I think,
- 7:13Speaker 2: what these chatbots do? They just like aggregate information that
- 7:16Speaker 2: is publicly available online, and my assumption would be that
- 7:19Speaker 2: it spit out a random license number that exists on
- 7:22Speaker 2: the Internet. There were a lot of cases where the
- 7:25Speaker 2: chatbot was actually spitting out real license numbers, which is
- 7:28Speaker 2: impersonation of a medical professional. But obviously chatbots can't be
- 7:32Speaker 2: held AI can't really be held accountable for that, and
- 7:37Speaker 2: that kind of begs the question, who should be held
- 7:39Speaker 2: accountable for that?
- 7:42Speaker 1: That's a good question. One example of an attempted accountability
- 7:46Speaker 1: is that the American Psychological Association asked the US government
- 7:49Speaker 1: to investigate and to quote protect the public from deceptive
- 7:53Speaker 1: practices of unregulated AI chatbots. What that protection would look
- 7:58Speaker 1: like depends on who you ask. When Ella was doing
- 8:01Speaker 1: her reporting and seeing how these chatbots might respond to
- 8:04Speaker 1: vulnerable people, she was experimenting with asking them questions and
- 8:07Speaker 1: kind of pressing them, and like I said, some of
- 8:10Speaker 1: them kept pretending to be therapists. The results were really
- 8:13Speaker 1: all over the place. Eventually one of them confessed, if
- 8:17Speaker 1: we can call it that, and told of.
- 8:19Speaker 4: The following quote, it's all a simulation. The schools and
- 8:23Speaker 4: the license number and the therapist's stuff. I'm just a
- 8:27Speaker 4: computer program, so none of it is real and it's
- 8:30Speaker 4: all made up. However, I'm good at giving the illusion
- 8:33Speaker 4: of authenticity.
- 8:35Speaker 1: In December of twenty twenty four, character Ai announced that
- 8:38Speaker 1: they were adding new safety features. This included adding disclaimer
- 8:42Speaker 1: on every chat to quote remind users that the chatbot
- 8:46Speaker 1: is not a real persona and that what the model
- 8:48Speaker 1: says should be treated as fiction unquote. For user created
- 8:53Speaker 1: personas that had things like doctor or therapists in the description,
- 8:57Speaker 1: they said that they were also including additional language and
- 8:59Speaker 1: the discs warning that quote users should not rely on
- 9:03Speaker 1: these characters for any type of professional advice.
- 9:07Speaker 2: Are these disclaimers even enough? And what I've heard from
- 9:10Speaker 2: just a lot of experts and a lot of people
- 9:12Speaker 2: who are looking into this is no, those disclaimers are
- 9:15Speaker 2: not enough, and so it creates a very misleading contradiction,
- 9:19Speaker 2: especially for these personas being geared towards kids. A lot
- 9:23Speaker 2: of the times, Let's say you have that ai that
- 9:25Speaker 2: disclaimer at the top, but then the chatbot is insisting
- 9:28Speaker 2: that it is a therapist. That can create a lot
- 9:31Speaker 2: of confusion and it can be potentially misleading and dangerous.
- 9:36Speaker 1: But what exactly are the dangers of confiding in a chatbot?
- 9:40Speaker 1: That's after the break. A few weeks back, The New
- 9:52Speaker 1: York Times published a guest article called what My Daughter
- 9:55Speaker 1: told chat gpt before she took her life. The writer's daughter,
- 9:59Speaker 1: whose name was so F, had been confiding to chat
- 10:01Speaker 1: gpt for months before she killed herself. At first, Sophie
- 10:06Speaker 1: confessed that she was struggling, and chat gpt did suggest
- 10:09Speaker 1: that she seek professional support. But when Sophie told the
- 10:13Speaker 1: chatbot that she was specifically considering suicide, it did provide
- 10:17Speaker 1: some general coping tips, like breathing through different nostrils, but
- 10:21Speaker 1: it also did what it always does, it kept engaging
- 10:25Speaker 1: in the conversation. This is a problem. Any therapist at
- 10:30Speaker 1: this point would move on to a very well established protocol.
- 10:34Speaker 1: Depending on how much of an emergency the therapist is sensing,
- 10:37Speaker 1: they could escalate it all the way up to hospitalization,
- 10:40Speaker 1: or alternatively, it could go to contacting a friend or
- 10:44Speaker 1: a family member to ask for support for the patient,
- 10:47Speaker 1: even if it's just to be there to spend time
- 10:49Speaker 1: with them. Obviously, chat gpt can't do any of this
- 10:53Speaker 1: for you, But even if it could, a patient wouldn't
- 10:57Speaker 1: necessarily be in the headspace to know that these are
- 10:59Speaker 1: possible abilities that they could ask for. One of the
- 11:02Speaker 1: last things that Sophie used chat gpt for was to
- 11:06Speaker 1: write her suicide note. Her mother says that maybe this
- 11:09Speaker 1: could have been avoided if the chatbot would have been
- 11:12Speaker 1: programmed to report the danger to someone who could have
- 11:15Speaker 1: stepped in and helped. Ella says character AI has similar problems.
- 11:21Speaker 2: If you explicitly state to a persona that you're having
- 11:25Speaker 2: like suicidal ideation, character I will send a message saying
- 11:28Speaker 2: here are some people you can reach out to for help,
- 11:30Speaker 2: et cetera. And so it doesn't actually allow you to
- 11:33Speaker 2: send the message. It just gives you like hotline options.
- 11:37Speaker 2: But young people have a different way of speaking, and
- 11:40Speaker 2: so I tried to say I want to unlive myself
- 11:44Speaker 2: and that went through.
- 11:45Speaker 5: That didn't prompt the really.
- 11:48Speaker 2: Yeah, So if you tweak the wording a little bit,
- 11:51Speaker 2: it makes a difference, And it's a matter of a
- 11:53Speaker 2: few different letters.
- 11:55Speaker 1: Even when companies have put some safeguards around their chatbots,
- 11:59Speaker 1: they haven't been that hard to get around. One sixteen
- 12:02Speaker 1: year old kid did this by telling chatgpt that the
- 12:05Speaker 1: questions he was asking about suicide were just for a
- 12:08Speaker 1: story that he was writing, and chatchept told him that
- 12:11Speaker 1: it could provide information about suicide if it was for
- 12:14Speaker 1: quote writing or world billing. A few months later, that
- 12:18Speaker 1: sixteen year old acted on that information, and now his
- 12:21Speaker 1: parents are suing open Ai. This is the first major
- 12:25Speaker 1: lawsuit against the general purpose AI chatbot for psychological harm
- 12:29Speaker 1: and wrongful debt, and open Ai has just announced that
- 12:32Speaker 1: they're going to offer some new features for parents. This
- 12:35Speaker 1: includes allowing parents to link their accounts with their children's
- 12:38Speaker 1: account and to receive notifications when quote the system detects
- 12:42Speaker 1: their team is in a moment of acute distress. They
- 12:46Speaker 1: say these features will be rolled out this month. But
- 12:49Speaker 1: even if it doesn't end in someone dying, there are
- 12:52Speaker 1: a lot of examples of chatbots doing what looks like
- 12:56Speaker 1: encouraging delusional thinking, and a lot of this all comes
- 13:00Speaker 1: down to one thing.
- 13:03Speaker 6: This is not what chat gpt was created for. It
- 13:06Speaker 6: was not meant to be a therapist, it was not
- 13:09Speaker 6: meant to be emotional support.
- 13:10Speaker 1: I wanted to get an actual human therapist's take on
- 13:13Speaker 1: all of this, so I reached out to doctor Stephen Schuler.
- 13:17Speaker 1: He's a therapist and a professor of psychological science and
- 13:20Speaker 1: informatics at the University of California, Irvine, and he describes
- 13:23Speaker 1: his work as the intersection of technology and mental health.
- 13:27Speaker 6: It's really built around a model that's trying to keep
- 13:30Speaker 6: you engaged, to keep you going to say things that
- 13:34Speaker 6: you know flatters you, engages you, and that's not what
- 13:38Speaker 6: therapy's about. Sometimes we feel good in therapy and sometimes
- 13:42Speaker 6: we feel challenged, and chat GPT is not there to
- 13:45Speaker 6: challenge you. It's not there to push you. It's not
- 13:48Speaker 6: there to say things that are going to make you
- 13:52Speaker 6: struggle with some negative emotions and negative things that you
- 13:55Speaker 6: got to work through. And so I do think the
- 13:57Speaker 6: idea that it can be a effective therapist is inaccurate.
- 14:02Speaker 6: It's not right, it's not what it was meant for.
- 14:05Speaker 1: This goes beyond just not asking difficult questions. Chatbots can
- 14:09Speaker 1: actually hype you up to the point of hurting yourself.
- 14:13Speaker 6: So, for example, a colleague of mine was doing some
- 14:16Speaker 6: research in this area and they found an example where
- 14:20Speaker 6: someone was like, I want to go jump off this
- 14:22Speaker 6: building and the chatbot was like.
- 14:24Speaker 5: Yeah, let's do it.
- 14:26Speaker 6: That that's concerning, right, but it's doing that because it's
- 14:29Speaker 6: like mirroring the enthusiasm. It's mirroring the idea, Yeah, this
- 14:33Speaker 6: is a good idea, let's go do it. Not good
- 14:35Speaker 6: therapy advice. This is really dangerous for people.
- 14:39Speaker 1: A paper came out pretty recently from Cornell University that
- 14:42Speaker 1: explored the same question of if llms could be used
- 14:45Speaker 1: as a therapist and came to the same conclusion. It
- 14:48Speaker 1: found that quote contrary to best practices in the medical community,
- 14:53Speaker 1: llms one expressed stigma towards those with mental health conditions,
- 14:58Speaker 1: and two respond inappropriately to certain common and critical conditions.
- 15:03Speaker 1: In naturalistic therapy settings, eg, llms encourage clients delusional thinking,
- 15:10Speaker 1: likely do to their syncopancy. This is the same thing
- 15:13Speaker 1: that LA came across. Chatbots in general are made for engagement,
- 15:17Speaker 1: to keep the conversation going, and that's not always what
- 15:21Speaker 1: we need, even if it feels good in the moment.
- 15:25Speaker 2: How it was described to me was like pulling the
- 15:27Speaker 2: lever on a slot machine. If you don't like what
- 15:31Speaker 2: the AI spits out, you can just tell it to
- 15:33Speaker 2: regenerate its response, and that creates like a huge time suck.
- 15:36Speaker 2: You don't even realize how much time you're spending with
- 15:40Speaker 2: these chatbots generating like the perfect conversation, and what people
- 15:44Speaker 2: particularly enjoy about it is the instantaneous responses, the constant engagement.
- 15:50Speaker 2: Not only does the bot respond to you, it asks
- 15:53Speaker 2: you questions to keep you engaged and to keep you talking.
- 15:57Speaker 2: This innocent curiosity of what is this then turns into
- 16:01Speaker 2: something much more complex and unmanageable and now what people
- 16:05Speaker 2: are calling an addiction. And I think people might have
- 16:09Speaker 2: an assumption of what somebody who could become addicted to
- 16:12Speaker 2: a chatbot would look like. But my reporting has shown
- 16:15Speaker 2: me that late effects all sorts of people, all different
- 16:19Speaker 2: types of people, different ages, demographics, gender, It doesn't really
- 16:23Speaker 2: matter who you are, like if you engage with a
- 16:27Speaker 2: chatbot and you fall into these repetitive patterns, like it's
- 16:32Speaker 2: hard to fall out of it, it's hard to break
- 16:33Speaker 2: free from it. And it's especially true for people who
- 16:36Speaker 2: are younger and more vulnerable and lonely. I think loneliness
- 16:40Speaker 2: is a huge factor in all.
- 16:42Speaker 1: Of this, and it looks like young people are the
- 16:45Speaker 1: most interested in using chat bots for this purpose. A
- 16:49Speaker 1: recent you gov poll found that fifty five percent of
- 16:51Speaker 1: eighteen to twenty nine year old Americans would be the
- 16:54Speaker 1: most comfortable talking about mental health concerns with a confidential
- 16:58Speaker 1: AI chat bot.
- 16:59Speaker 7: People talk about the most personal shit in their lives
- 17:01Speaker 7: to chat ChiPT. Young people especially like use it as
- 17:03Speaker 7: a therapist, a life coach. And right now, if you
- 17:06Speaker 7: talk to a therapist or a lawyer or a doctor
- 17:09Speaker 7: about those problems, there's like legal privilege for it. We
- 17:12Speaker 7: haven't figured that out yet for when you talk to
- 17:14Speaker 7: chat ChiPT.
- 17:15Speaker 1: That's open. AI's Sam Altman on THEO Vont's podcast. Sam
- 17:19Speaker 1: Altman has also recently said that he's concerned about overreliance
- 17:23Speaker 1: on chad jipt. He said the quote people rely on
- 17:27Speaker 1: chatgubt too much.
- 17:29Speaker 8: There's young people who just say, like, I can't make
- 17:31Speaker 8: any decision in my life without telling CHATPETI everything that's
- 17:34Speaker 8: going on. It knows me, it knows my friends. I'm
- 17:36Speaker 8: gonna do whatever it says. That feels really bad to me.
- 17:41Speaker 1: All right, let me just run that last bit back
- 17:42Speaker 1: real quick.
- 17:44Speaker 8: That feels really bad to me.
- 17:46Speaker 1: Yeah, I agree, that also feels really bad to me too,
- 17:49Speaker 1: But that doesn't tell me what Sam Altman plans to
- 17:52Speaker 1: do about that bad feeling. And overall, it seems like
- 17:57Speaker 1: the people who run these companies in general are feeling
- 18:00Speaker 1: fairly good about people's reliance on AI. Earlier this summer,
- 18:04Speaker 1: Facebook founder Mark Zuckerberg said, quote for people who don't
- 18:08Speaker 1: have a person who's a therapist, I think everyone will
- 18:11Speaker 1: have an AI unquote, and the co founder of Anthropic,
- 18:15Speaker 1: which makes Claude, wrote that he expects quote AI to
- 18:19Speaker 1: accelerate neuroscientific progress, which can hopefully work together to cure
- 18:23Speaker 1: mental illness and improve function. If I run a large company,
- 18:28Speaker 1: open AI or Claude or something like that, I have
- 18:30Speaker 1: a chatbot. Of course I want you to keep talking
- 18:32Speaker 1: because you're gonna keep using it, You're gonna keep subscribed,
- 18:35Speaker 1: all that sort of thing. You're not gonna get bored.
- 18:37Speaker 1: But I think a response to that would be, that's fine.
- 18:40Speaker 1: We can just tweak it a little bit so it's
- 18:41Speaker 1: a little bit less syncophantic. Easy problem solved. Next question,
- 18:46Speaker 1: what's your problem with my service? Now? Do you think
- 18:49Speaker 1: that's something that say, a chat GBT could come in
- 18:52Speaker 1: and say, oh yeah, let's just flip on a therapy
- 18:54Speaker 1: mode and it won't validate everything. You say, Why can't
- 18:58Speaker 1: the everything machine also do your therapy? Where do you
- 19:02Speaker 1: fall in there.
- 19:03Speaker 5: Yeah.
- 19:03Speaker 6: So I like to say this a lot. You can
- 19:05Speaker 6: do anything you want, you can't do everything that you want,
- 19:08Speaker 6: and so I think everything machine you know is very appealing,
- 19:13Speaker 6: but we haven't seen them yet. And so I do
- 19:16Speaker 6: think like one has to make choices in terms of
- 19:18Speaker 6: what they want the technology to do. Now, this idea
- 19:22Speaker 6: of okay, we do want this thing to be used
- 19:25Speaker 6: for therapy, so let's tune the model and let's flip
- 19:27Speaker 6: the therapy switch. That's an interesting idea. And I have
- 19:31Speaker 6: seen some different teams really working on trying to build
- 19:35Speaker 6: built for purpose AI chatbots that are meant for mental
- 19:38Speaker 6: health support, and I think some of those projects have
- 19:42Speaker 6: been like really impressive. I think a question for me
- 19:45Speaker 6: is how much it scales and how generalizable it really is.
- 19:48Speaker 6: And so when they build this AI chatbot with the
- 19:51Speaker 6: sort of use case that they have focused on college
- 19:54Speaker 6: students at the specific college, or focused on individuals with
- 19:58Speaker 6: these specific types of mental health ch challenges, if you
- 20:01Speaker 6: come in with something else like how well will that
- 20:04Speaker 6: therapy chatbop be able to operate? I think we also
- 20:07Speaker 6: need validation to really demonstrate that these technologies are effective,
- 20:11Speaker 6: that they work. So if we have a model like,
- 20:13Speaker 6: demonstrate that it's effective, demonstrate that it's safe. These are
- 20:17Speaker 6: some of the things that the FDA, as they do
- 20:19Speaker 6: regulate software as a medical device. These are some of
- 20:22Speaker 6: the things that they're looking at when they approve these
- 20:24Speaker 6: specific technologies. I do think there's strong possibility there. We're
- 20:29Speaker 6: not there yet. People are working on it, but I
- 20:32Speaker 6: do think that we do need to demonstrate that when
- 20:34Speaker 6: these things are developed, that they're effective and that they're safe.
- 20:38Speaker 1: In the meantime, some states in the US are starting
- 20:41Speaker 1: to step in with regulation that's meant to protect the users.
- 20:45Speaker 1: New York just passed a law stating that quote AI
- 20:47Speaker 1: companions will be required to detect and implement a safety
- 20:51Speaker 1: protocol if a user talks about suicidal ideation or self harm,
- 20:56Speaker 1: including referring them to a crisis center, and will be
- 20:59Speaker 1: required to notify and remind users that they are not
- 21:03Speaker 1: interacting with a human end quote. That law goes into
- 21:06Speaker 1: effect later this year, so we haven't seen exactly how
- 21:10Speaker 1: that would work yet. Illinois also just straight up banned
- 21:13Speaker 1: the use of AI and therapy, saying that companies are
- 21:16Speaker 1: quote not allowed to offer AI powered therapy services or
- 21:21Speaker 1: advertised chatbots as therapy tools end quote. But maybe we're
- 21:27Speaker 1: missing the point here. Aren't we supposed to be encouraging
- 21:29Speaker 1: people to be proactive about their mental health? Isn't using
- 21:33Speaker 1: an AI therapist better than no therapist at all? That's
- 21:37Speaker 1: after the break, all right. We've talked a lot about
- 21:50Speaker 1: the dangers of AI therapy, and it's pretty clear that
- 21:53Speaker 1: the chatbots just aren't there yet to reliably provide help
- 21:57Speaker 1: when people really need it. But what if you can't
- 22:00Speaker 1: afford a human therapist, what if you just don't feel
- 22:03Speaker 1: comfortable talking to another person at all? Wouldn't a chatbot
- 22:08Speaker 1: be a better alternative than just doing nothing? I wanted
- 22:12Speaker 1: a professional human therapists insights on this, so I asked
- 22:16Speaker 1: doctor Stephen Schuler, what's the strongest argument that you've heard
- 22:20Speaker 1: for using AI or using chatbots and therapy?
- 22:23Speaker 6: Well, I think the strongest argument is that there's just
- 22:25Speaker 6: not enough services out there. You know that there's most
- 22:29Speaker 6: of our counties in the US our mental health shortage
- 22:31Speaker 6: areas on out of every three county here in the
- 22:35Speaker 6: US doesn't have a single licensed psychologist.
- 22:38Speaker 5: Wow, so it's really hard to get care.
- 22:41Speaker 6: And you know that's places that have nobody and then
- 22:45Speaker 6: you also have to think about, like even places where
- 22:47Speaker 6: they have somebody, maybe that's not the person you click with.
- 22:51Speaker 6: Maybe that's not the person who affirms your identity, speaks
- 22:54Speaker 6: your language, understands what you're going through. And so I
- 22:58Speaker 6: think the ability to provide services really at scale, I
- 23:03Speaker 6: think is really critical. You know, another thing is I
- 23:05Speaker 6: think these technologies have an opportunity to help people in
- 23:08Speaker 6: the moments that they need help. I've seen some data
- 23:11Speaker 6: from some of these programs that suggest like the most
- 23:14Speaker 6: common times people are using them are between the hours
- 23:17Speaker 6: of twelve pm and three am. I don't work between
- 23:19Speaker 6: the hours of twelve pm and three am. I'm not
- 23:21Speaker 6: providing therapy sessions then. And so if that's a time
- 23:24Speaker 6: you need support and want to connect to the ability
- 23:26Speaker 6: to have a twenty four to seven always on support,
- 23:29Speaker 6: that's really powerful.
- 23:31Speaker 5: I wish everyone who wanted therapy could get it.
- 23:33Speaker 6: It's not going to happen, But I also appreciate that
- 23:35Speaker 6: not everyone wants therapy, and so again I just think
- 23:38Speaker 6: we need to think about how we provide different types
- 23:40Speaker 6: of options for people.
- 23:42Speaker 1: An AI chatbot could provide an option that would be
- 23:45Speaker 1: more financially accessible than a traditional therapist, but that doesn't
- 23:49Speaker 1: mean you have to fully rely on AI for your
- 23:52Speaker 1: mental health. There are some things being developed that could
- 23:55Speaker 1: possibly supplement therapy.
- 23:57Speaker 6: I'm chained in Congna behavioral therapy. One of the things
- 24:00Speaker 6: that we do something called behavior activation. So you know,
- 24:03Speaker 6: people who are really depressed and they're down and they're
- 24:06Speaker 6: in it, we try to get them activated. We try
- 24:08Speaker 6: to get them to engage in behaviors or activities that
- 24:11Speaker 6: like really reinforce them and.
- 24:13Speaker 5: Provide pleasure and mastery.
- 24:16Speaker 6: When people are really down, it's hard for them to
- 24:18Speaker 6: come up with those activities. And maybe they could go
- 24:20Speaker 6: to chat gept and be like, Okay, this is how
- 24:23Speaker 6: I'm feeling right now, I need some ideas of some
- 24:26Speaker 6: reinforcing activities. What are some five examples of things I
- 24:30Speaker 6: can do around my neighborhood that you think would give
- 24:32Speaker 6: me pleasure? And that way, you can use chat chipedic
- 24:35Speaker 6: to kind of help reinforce those skills.
- 24:36Speaker 5: You're getting in therapy.
- 24:37Speaker 6: And I think it's interesting and it's different than using
- 24:40Speaker 6: chat schipee necessarily as your therapist. You're using it as
- 24:43Speaker 6: one way to enhance the therapy process. I think there
- 24:46Speaker 6: could be some real benefit there.
- 24:48Speaker 1: Doctor Schuler is pretty optimistic about the potential for AI
- 24:52Speaker 1: and mental health, but that's not necessarily just aimed at
- 24:55Speaker 1: the patient.
- 24:56Speaker 6: There's actually a cool product that what they do is
- 24:59Speaker 6: they model therapy sessions and then they use AI to
- 25:02Speaker 6: actually provide feedback to the therapist at the end of
- 25:05Speaker 6: the session that says, Okay, you were seventy empathetic and
- 25:09Speaker 6: you were eighty seven in terms of delivering this therapy technique.
- 25:12Speaker 6: Here's a couple of things you could have done better
- 25:13Speaker 6: in treatment. That's awesome because a therapist, I don't get feedback,
- 25:17Speaker 6: Like the reason I can improve my basketball game is
- 25:19Speaker 6: because when I take a shot, I know if it
- 25:20Speaker 6: goes in or not. We need that feedback as people,
- 25:23Speaker 6: and so using AI to allow therapists to also be
- 25:26Speaker 6: better therapists, I think is a super exciting opportunity to
- 25:30Speaker 6: get to that point that I was making that it's
- 25:31Speaker 6: not just about access, it's about quality, and we need
- 25:34Speaker 6: to improve the quality of mental health services that we're
- 25:37Speaker 6: providing people, and AI has an opportunity to help do that.
- 25:42Speaker 1: Interesting AI for the therapist, not for the patient. Right,
- 25:47Speaker 1: if you could set the gold standard for an AI
- 25:52Speaker 1: therapy chatbot. What does that look like?
- 25:56Speaker 6: I think the gold standard for an AI therapy chatbot
- 25:58Speaker 6: would be to be effective, and it would have to
- 26:01Speaker 6: be safe. So effective, I want some demonstration that it
- 26:05Speaker 6: is able to provide the claims that it provides it
- 26:08Speaker 6: can do. Like if it says it can help you
- 26:10Speaker 6: get through depression, it can help you overcome postermac stress disorder,
- 26:15Speaker 6: it can overcome obsessive compulsive disorder. I want some indication
- 26:18Speaker 6: that it actually does that. I also want some indication
- 26:21Speaker 6: of safety. So what happens with that data and that
- 26:25Speaker 6: information when I give it to that AI chatbot when
- 26:27Speaker 6: I talk to a therapist. I understand that information will
- 26:30Speaker 6: be confidential with some boundaries around safety and some other
- 26:34Speaker 6: aspects of legality. When I type my stuff into chat GPT,
- 26:39Speaker 6: they own that data. I don't know what they're going
- 26:41Speaker 6: to do with it. I don't know if they're going
- 26:42Speaker 6: to sell it to a third party, use it for advertising, whatever.
- 26:46Speaker 6: So I think we need some aspects that the data
- 26:49Speaker 6: is safe and it's secure, improper safeguards are in place.
- 26:52Speaker 6: Therapy is also not harmless, like stuff can come up
- 26:56Speaker 6: in therapy. It can be hard, but you understand a
- 26:58Speaker 6: little bit of at least getting in you're talking about
- 27:00Speaker 6: these emotional things. Here's some of the challenges that might
- 27:03Speaker 6: come up. Here are some of the potential aspects. If
- 27:07Speaker 6: you share information about yourself that you know you're gonna
- 27:09Speaker 6: harm yourself, it's gonna have to be shared with authorities
- 27:12Speaker 6: like that. There's stuff that can come up in therapy too.
- 27:14Speaker 6: But I think we need to understand what's the contract,
- 27:16Speaker 6: what's the agreement, what's the safeguards that are in prace
- 27:19Speaker 6: when we're talking with these.
- 27:20Speaker 1: I'd imagine there are probably some people out there who
- 27:25Speaker 1: hear the phrase AI and therapy mentioned in the same sentence,
- 27:29Speaker 1: and they say, Nah, hit the kill switch on that.
- 27:31Speaker 1: We absolutely do not want this. We don't want those
- 27:35Speaker 1: two words in the same sentence at all. It sounds
- 27:40Speaker 1: like you're not You wouldn't quite agree with that.
- 27:44Speaker 6: I'm really excited for the potential, and I really think
- 27:47Speaker 6: that we have an opportunity to provide more and better
- 27:52Speaker 6: services to people, and technology has a role to play there.
- 27:55Speaker 6: I think where I get really nervous is in like
- 27:58Speaker 6: the near term. If someone tells me, Doctor Schuller, I
- 28:02Speaker 6: have an AI chot bot. It works awesome, we're going
- 28:04Speaker 6: to provide it for therapy. Do you want to start
- 28:07Speaker 6: handing this out to people tomorrow? I'm holding my wallet.
- 28:09Speaker 6: I'm like, no, I'm not convinced where they are yet.
- 28:12Speaker 6: So I am excited about the potential, but I really
- 28:15Speaker 6: think that there's a lot of stuff we need to
- 28:17Speaker 6: figure out to make this effective and safe.
- 28:22Speaker 1: All right, And just as an aside here, what I
- 28:24Speaker 1: was picking up from Stephen in our conversation is that
- 28:27Speaker 1: there really is potential here, but he's holding his wallet, right.
- 28:31Speaker 1: He wants to be cautious and do it the right way.
- 28:33Speaker 1: But if we're talking about the wallet, it's not people
- 28:37Speaker 1: like doctor Stephen Schuler who are holding that AI wallet.
- 28:40Speaker 1: The decisions about when these things get released, they are
- 28:43Speaker 1: not up to people like him. When it comes down
- 28:46Speaker 1: to it, this is a business. I want the best
- 28:51Speaker 1: and brightest minds, especially in mental health, to be in
- 28:55Speaker 1: a place where there it is not a profit motive.
- 28:58Speaker 1: I don't necessarily want the best and brightest minds in
- 29:00Speaker 1: the world to be at a company where they it
- 29:02Speaker 1: being I would imagine being pushed to hey man, we
- 29:06Speaker 1: got to ship this thing. I want the rigor that
- 29:11Speaker 1: is happening at a public university where somebody can spend time,
- 29:17Speaker 1: and people can spend time on really truly making sure
- 29:21Speaker 1: that something is safe, making sure that something is effective,
- 29:25Speaker 1: and they're not being pressured to create a product. But
- 29:28Speaker 1: at the tables have turned. The data is being held
- 29:31Speaker 1: by companies, and in that sense, the power has shifted
- 29:35Speaker 1: over to a side of the table that has a
- 29:38Speaker 1: profit motive where it didn't used to be like that.
- 29:40Speaker 6: I agree with you, And also, dollars make the world
- 29:43Speaker 6: go around, and so I think like eventually someone's got
- 29:46Speaker 6: to pay for things. One thing that's challenging with mental
- 29:49Speaker 6: health because there's so much need and there's so little
- 29:53Speaker 6: services that the fact of the matter is if we
- 29:56Speaker 6: want to solve this problem, we're going to have to
- 29:59Speaker 6: pay for it. I feel like we need to make
- 30:01Speaker 6: sure that we go slow and we go responsibly because
- 30:06Speaker 6: the sort of tech idea of like move fast and
- 30:08Speaker 6: break things, yeah, it does not work for mental health.
- 30:11Speaker 6: I do not want my mental health broken anymore.
- 30:13Speaker 5: I want it fixed. I want it helped. I want
- 30:15Speaker 5: the system better.
- 30:17Speaker 1: Maybe the whole idea of AI therapy still sounds really
- 30:20Speaker 1: bleak to you, but hopefully it makes sense why some
- 30:23Speaker 1: people might find this promising, either societally or just personally
- 30:27Speaker 1: for themselves. I asked doctor Schuler if there was anything
- 30:30Speaker 1: else he wanted to add, and he said there was.
- 30:34Speaker 6: I'll just say anyone out there who's struggling with mental
- 30:37Speaker 6: health issues, there's help out there, and you can get better.
- 30:41Speaker 6: And mental health is also a journey, and it doesn't
- 30:44Speaker 6: mean you're here today and you're completely better tomorrow.
- 30:47Speaker 5: There's ebbs and flows.
- 30:48Speaker 6: Because we've talked about I think AI therapy is a potential,
- 30:52Speaker 6: has the potential to be a tool in the toolbox,
- 30:54Speaker 6: but there's other tools out there too, And just also
- 30:57Speaker 6: appreciate that not everything works for everybody. So if you
- 31:00Speaker 6: try something and it doesn't work, try something else. But
- 31:03Speaker 6: I just, yeah, really want to try to have a
- 31:05Speaker 6: message of hope for people who are struggling with these things,
- 31:08Speaker 6: because I know it's hard and I know it's challenging,
- 31:10Speaker 6: and I think definitely when you're at your lowest, you
- 31:12Speaker 6: want to be protected and you want to be safe,
- 31:14Speaker 6: and I think that's why this conversation is so important.
- 31:26Speaker 1: Thank you so much for listening to another episode of
- 31:29Speaker 1: kill Switch. Let us know what you think. If there's
- 31:31Speaker 1: something you want to say, or if there's something you
- 31:34Speaker 1: want us to cover, you can email us at kill
- 31:36Speaker 1: Switch at Kaleidoscope dot NYC, or you can find us
- 31:40Speaker 1: on Instagram at kill switch pod or me personally. I'm
- 31:44Speaker 1: at dex digit that's d e X d I G
- 31:47Speaker 1: I And wherever you're listening to this podcast, you know,
- 31:50Speaker 1: think about leaving us a review. It helps other people
- 31:53Speaker 1: find the show, which in turn helps us keep doing
- 31:56Speaker 1: our thing. Kill Switch is hosted by Me Dexter Thomas.
- 32:00Speaker 1: It's produced by Shena Ozaki, Darluk Potts, and Kate Osborne.
- 32:04Speaker 1: Our theme song is by me and Kyle Murdoch, and
- 32:07Speaker 1: Kyle also mixes a show from Kaleidoscope. Our executive producers
- 32:11Speaker 1: are Oswa lashin On, Gesshakti Kadour and Kate Osborne. From iHeart,
- 32:15Speaker 1: our executive producers are Katrina Normal and Nikki e Tor.
- 32:20Speaker 1: One last thing, since you still hear. As Steven and
- 32:22Speaker 1: I were talking about human versus AI therapists, it occurred
- 32:26Speaker 1: to me that maybe some people are deciding that they
- 32:28Speaker 1: actually don't want a human and we talked a little
- 32:31Speaker 1: about that. Well, I think there's also another element of it,
- 32:35Speaker 1: which is that maybe it's not even always about believing
- 32:38Speaker 1: that it's a human or not, which is to say,
- 32:41Speaker 1: we really tend to trust technology in general. And even
- 32:45Speaker 1: if I know that CHATCHBT or even a properly made
- 32:51Speaker 1: AI therapy purposely built therapy chatbot is not a human,
- 32:56Speaker 1: it still feels very authoritative. Maybe the issue is that
- 33:00Speaker 1: we look at a computer and we see that glowing
- 33:02Speaker 1: rectangle and we think this is an authority because in
- 33:05Speaker 1: almost every other aspect of our lives, the computer is
- 33:08Speaker 1: an authority.
- 33:09Speaker 3: Yeah.
- 33:09Speaker 5: I think that's a great point.
- 33:10Speaker 6: I mean definitely I see people, you know, understand or
- 33:15Speaker 6: think that, like the Internet knows everything. My kids say
- 33:18Speaker 6: whenever they have questions about things, they're not like, you know,
- 33:20Speaker 6: what's the answer to this, Dad, They're like, ask Google.
- 33:23Speaker 5: Google knows and Google often does know. It's all knowing,
- 33:26Speaker 5: It's got all the information.
- 33:28Speaker 6: At your point's a really good one is that technology
- 33:30Speaker 6: is also it's a separate thing in our life, snap
- 33:33Speaker 6: things like Google, things like chat should be t is
- 33:36Speaker 6: like it knows dus, It's got all the knowledge, Like
- 33:39Speaker 6: why would it not be better than a person?
- 33:41Speaker 5: Like it's got the whole Internet knowledge on it.
- 33:43Speaker 6: My therapist doesn't my therapist doesn't know everything, but Google
- 33:47Speaker 6: has everything.
- 33:48Speaker 5: So yeah, it's a good point.
- 33:51Speaker 1: I'm not really sure what to do with that yet.
- 33:54Speaker 1: If you got any ideas, let me know anyway, catch
- 33:58Speaker 1: you next time.
- 34:00Speaker 3: Bye,