00:00We haven't talked enough about AI scary things.
00:02The relationship between therapist and patient is an intimate one,
00:05but there are no efforts underway to incorporate AI into that therapy.
00:09Journalist Isabel Woodford has written about how psychologists and some celebrities
00:12are turning to AI companies to make them AI twins who can mimic how they think and communicate.
00:18We started off by asking Isabel how this process works.
00:22The simplest way that we talk about in this piece is Delphi,
00:26which is this startup backed by Sequoia, and they've created this technology
00:30where if you're a fairly prolific individual, in our case, in this case, in the piece, therapist,
00:36you would upload all of your public content, podcasts, books, writing,
00:41and you would feed this through Delphi software to create your digital twin.
00:46But we also spoke to other therapists who had literally used AI, a clawed bot,
00:52and over a year just been speaking to them, correcting them, training the bot effectively.
00:59So really anyone could do this using clawed, open AI, whatever software they have.
01:07You talk to a bunch of different people, but it seems to me from a therapy standpoint,
01:11there's like two ways to go about this.
01:13You have one therapist who's like a sex therapist,
01:16and her patients want to talk to her at all times of day
01:18so they can interact directly with this chatbot version of her
01:22and get theoretically similar answers she would give them,
01:26and it allows her to be accessible to more people if they're paying her subscription.
01:29There's another therapist who uses it to kind of be the, like he's the intermediary.
01:34They're not allowed to interact with it directly, but it saves him time
01:37because his patient can email, I'm having this problem,
01:40and then rather than spending 45 minutes crafting an email back,
01:43he feeds it into the AI.
01:45It crafts the response.
01:46He then checks it and says, yeah, that's what I would say, and sends it off,
01:49and it allows him to be more available but takes up less of his time.
01:52What are the pros and cons of both those kind of approaches?
01:56Yeah, I mean, so effectively the Delphi bot, the sex therapist,
02:00that really is just completely hands-off.
02:03I mean, she effectively has done the work of training this bot,
02:06and it interacts directly with the user.
02:10That created a lot of questions from therapists I spoke to,
02:14this idea that this bot was user-ready in a therapeutic environment.
02:20I used it just to kind of test it and for fun,
02:23and I was just kind of like making up questions to it.
02:26And, I mean, you know, you could see the appeal as a client.
02:30Did it feel like you were talking to a real person?
02:33Yeah, totally.
02:35I mean, freakishly, because if you use it on the website,
02:38the little avatar actually moves.
02:41So, you know, you can actually see this individual,
02:43the sex therapist's face, kind of moving as well as reading the answers.
02:48And the tone is very unique.
02:50It doesn't have that kind of AI voice.
02:53It's quite individualized.
02:55So, it did feel pretty authentic.
02:59And I guess with the other option for Oscar,
03:01who has trained his bot and he's the intermediary,
03:05you could say that there are fewer ethical questions.
03:09I mean, there is no direct kind of interaction with the user.
03:14But in terms of how you scale that,
03:17I guess he's still very involved.
03:19And, yeah, that's kind of up to the therapist.
03:23You note this is like a $50 billion industry.
03:25Therapy is a $50 billion industry.
03:28And I'm curious what professional associations say about
03:31the role that AI is playing now,
03:33the way that they think about how AI could be incorporated
03:35into therapy going forward.
03:38Yeah, so I spoke to the American Sociological Association.
03:42And they were pretty realistic.
03:44The researcher I spoke to there said that she predicts
03:47a hybrid future for therapy.
03:49So, one in which you have a human therapist
03:52and some sort of AI accompaniment,
03:56accomplice or kind of use additional assistance.
03:59So, I mean, that was pretty stark for me,
04:02this idea that therapy will be a hybrid model.
04:08And already in the UK, in the NHS,
04:10they are using AI chatbots, specialist chatbots, I should say,
04:14not kind of the clods of the world.
04:16So, they are already rolling out AI chatbots
04:20for kind of patients who need mental health support.
04:23So, this is already here.
04:24I mean, the digital twin is maybe kind of a more
04:28kind of expanded version of that for the everyman.
04:31But this is already here.
04:33And it is quite stark to imagine a world of therapy
04:35where the human intermediary isn't alone anymore.
04:40Okay, so you say 77% of psychologists say
04:44they know patients are confiding in AI.
04:46But there have also been incidents of that ending very badly
04:48for people with mental problems.
04:50What are the fail-safes here?
04:52Are there flags or there triggers
04:53where a real person is going to get involved?
04:56So, again, just to kind of use the Delphi bot as an example,
05:01we actually asked a kind of researcher to test it properly
05:05because I could tell you, you know,
05:07this felt pretty safe for me, but I'm not a therapist.
05:10So, we asked her to put it to the test.
05:12And she said that the guidelines were pretty good,
05:15that it was clear that if she pretended
05:17to have suicidal ideation, for example,
05:19it would shut that down, it would respond,
05:21it could tell even hints of that.
05:25And so, in that sense, perhaps Delphi has been very considerate here
05:29about the most extreme mental health scenarios.
05:33But she also flagged that actually it didn't pick up
05:37things like anxiety and obsessive thinking.
05:40The Delphi bot wasn't able to capture that
05:42or at least didn't relay that to the user.
05:45So, on the spectrum of mental health crises,
05:49perhaps this researcher was saying
05:50that it might pick up the more extreme
05:53and it wouldn't do what we have read about
05:55with ChatGPT encouraging suicidal ideation,
05:58but it also might not be able to provide
06:00that therapeutic relationship
06:02and that therapeutic aid
06:04for less extreme cases.
06:07Isabel, let me ask you lastly
06:08just about another kind of economic question
06:10about all of this.
06:11So, you have therapists who are, I guess you say,
06:14downloading so much of their personal information
06:16into this to create a twin.
06:17Arnold Schwarzenegger, you know,
06:18has been doing this as well.
06:20Honestly, when I read that, I was like,
06:21I kind of want to interact with the Schwarzenegger bot,
06:24but, you know.
06:24How do they think about the economic calculation here?
06:27Because by doing this,
06:28I imagine they're going down this path
06:30where they could kind of render themselves
06:32maybe not useless, but less useful going forward.
06:34There has to be some sort of self-defeating
06:37economic calculation at play here.
06:40Yeah, I mean, I really was baffled by this
06:42because the Delphi CEO made quite clear
06:45that this is kind of a commercial avenue
06:48for the likes of Arnold Schwarzenegger
06:50and this therapist, for example.
06:52So, by charging people to use it,
06:55they are effectively monetizing themselves.
06:58And no one else can pretend to be you.
07:01You're not, you know, these are verified bots.
07:03So, Mark Cuban also has one, for example,
07:05to give business advice,
07:07and he's charging for that.
07:08And so, I think the idea is that, you know,
07:11if you want a specific individual,
07:13that person of potential fame,
07:17they can monetize this.
07:19Maybe what's the deeper question is,
07:22what does this mean in terms of the growth
07:23of the therapeutic industry?
07:25And, you know, I do think about
07:27the next cohort of therapists,
07:29and will there be less demand for new people
07:32if one can already use the bots available?
07:35So, I think, you know,
07:36the most prolific people here
07:37are certainly going to win.
07:40But there are definitely economic questions
07:42for the up-and-comers.
07:45Also, what is the liability here?
07:47If your chat bot does something wrong,
07:50even if you're not a therapist,
07:52if you're Mark Cuban,
07:53you give someone bad business advice,
07:54or if you are a therapist,
07:55and you do something that someone then claims
07:57hurt their life or ruined,
07:58like, who's liable for that?
08:00Yeah, I mean, I imagine
08:02that they have kind of disclaimers
08:04as part of the use,
08:06you know, by signing on to use it.
08:08And as I say, for the Delphi bot,
08:10if you're not paying,
08:11and you're not kind of a known user,
08:13you do get cut off pretty quick.
08:15So, you can maybe ask three or four questions
08:17before it says,
08:18you know, you should call me,
08:20or we should talk.
08:21So, I think there are some
08:23kind of practical disclaimers
08:26that you as an individual
08:27have to get past.
08:29But yeah, these are huge questions
08:30for these big AI companies,
08:32and I think we're waiting to see
08:34the answer to that across the board.
Comments