Would You Rather Have An AI or a Human Therapist?

My sense about these things is, you know, it's hard because on the one hand, there's a loneliness crisis. There's not enough therapists, People need a lot of help, and you're saying, well, you know, you can't get to a therapist in the real world, so talk to a chatbot. The chatbot that you'll talk to. It'll always be available. It will always be nice. It will always say that what you're saying is really smart and really interesting. It's Is that a great modality to be trained into, given what you actually have to face in the world?

Hello, I'm Dr. Farrah White.

And I'm Dr. Grant Brenner.

We're psychiatrists and therapists in private practice in New York. We started this podcast in 2019 to draw attention to a phenomenon called the doorknob comment. Doorknob comments are important things we all say from time to time, just as we're leaving the office, sometimes literally hand on the doorknobs.

Doorknob comments happen not only during therapy but also in everyday life. The point is that sometimes we aren't sure how to express the deeply meaningful things we're feeling, thinking, and experiencing. Maybe we're afraid to bring certain things out into the open or are on the fence about wanting to discuss them. Sometimes we know we've got something we're unsure about sharing and are keeping it to ourselves. Sometimes we surprise ourselves by what comes out.

We're here today with Professor Nir Isakovich. He is a professor of philosophy and the founding director of the Applied ethics Center at UMass Boston. We're going to be talking about artificial intelligence, ethics, and psychotherapy. Before coming to UMass, he taught at Boston University and Suffolk University. His research focuses on the moral and political dilemmas arising after war, the psychology of war, and the ethics of technology. His books include A Theory of Truces, Sympathizing with the Enemy, Reconciliation, Transitional Justice, Negotiation, Theorizing Transitional justice and Glory, Humiliation, and the Drive to War. He is also guest editor for a recent issue of Theoria on the Idea of Peace in the Age of Asymmetrical Warfare. He has written numerous articles on political reconciliation, the role of forgiveness in politics, truth commissions, and the ethics of war, and the ethics of artificial Intelligence. In addition to his scholarly work, Nir comments frequently on political conflict and the ethics of technology for American newspapers and magazines and has appeared in many very prestigious publications. There's more to say about Nir, but we'll leave it there for now and welcome you to the Doorknob comments podcast.

Thank you. Thank you for having me.

I always tend to ask how Grant sort of crossed paths with someone. So I am curious, what did you guys connect over?

And when I think, Grant, you read an essay of mine or heard me on the radio, I read your white.

Paper, the Ethics of Automating Therapy from the Institute for Ethics and Emerging Technology, and then I reached out and you graciously did a brief interview on my Psychology Today blog, experimentation on the ethics of AI Therapy.

That's right. That's right. Yeah. In the last few years, we've had a Ethics of Emerging Technology project that my center at UMass, and we wrote a white paper that sort of mapped out the questions that come up with chatbot psychotherapy. And so that's what grandson.

So is that. Because I this morning was listening to your podcast, Prosthetic Gods. Kind of cool concept, and I want to hear more about that. But I'm wondering, was it in the writing of that paper that you learned so much about psychotherapy, or is this an interest that sort of pre. Predated that?

Oh, I've always been really interested in psychotherapy and psychotherapeutic theory, psychoanalytic theory, you know, very much adjacent to philosophy. If you. Among other things, I work on the psychology and philosophy of war. If you do that kind of work, you have to read Freud. From there, you sort of start reading Freud adjacent people. So I've. I've always been fascinated by it.

Yeah, Freud has. I'm sure you've read all of this. But there's one of his earlier papers called On War and Death, which is really fascinating. And one of Freud's disciples, though there's a sad story there. You may be. I'm sure you're familiar with Sander Forensi's work, also wrote a monograph on the war neuroses.

Yeah, yeah. And there's this really cool correspondence Freud has with Einstein that I've been working on recently. Yeah, yeah. And, you know, Einstein asks this kind of very naive question, you know, doctor, what can we do to end the scroudge of war? And basically says, leave me alone.

Yeah. He's like, it's human nature. I don't know how to fix it yet.

Yeah, yeah. Apparently he didn't like Einstein very much because he didn't want to recommend him for the Nobel Prize. Wow.

Well, maybe I can help with these problems. I had asked the chat bot the other day, like, what kind of leader would have to emerge in order to bring world peace within one year and the AI Gave like, a pretty solid answer, like about being strategic, being extremely charismatic, having a strong moral compass, being blah, blah, blah, blah, blah. And then it ended by saying, that's a little too ambitious, though. It's more of a. A marathon than a sprint.

There you go.

It didn't say let an AI take over, though, which would have been scarier.

Yeah. Yeah. Well, I thought we could end it in one day, so there you go.

Yeah, the AI is like, well, mind if I drive? Move over, I'll drive.

You know, quite a few thousand years of recorded history, and we've been doing major warfare for most of them. I know there's thinkers like Steven Pinker who think that we're making consistent progress. I. Not. I don't buy that as much. I think there's measurement problems there. But I guess that's a. That's for a different podcast, right?

The angels of our better nature.

Yeah.

Yeah, people cite him. I saw him debate Robert J. Lifton on this subject many years ago, and I guess the proof is in the pudding, but right now it's a really scary time.

Yep.

So how did you shift from war to AI?

I didn't shift. I'm doing both. When I started the center at UMass in 2017, I decided to kind of expand a little bit beyond the war stuff. You know, at the time, mainly, it was a little bit too depressing to just keep focusing on the psychology of war. And AI had not become the hot button topic then yet that it is now. So I thought it would be, you know, kind of cool to apply the philosophical tools to a different area. And I teamed up with some engineers to make sure that I gradually become up on the technical issues and that I don't just, like, speculate in the air, as we philosophers are prone to do. So. Yeah, And I've been doing. I've been doing both in parallel. And one of the things that's kind of fun about the philosophy of technology is it makes a lot more sense to combine sort of public facing work with scholarly work in that context, because the technology changes so quickly that if you just stay with a scholarly cycle, by the time you publish something, the technology that you published it about has become obsolete. And so it's been a fun sort of. I mean, we do quite a bit of that too. You need to do that both for the credibility and for the ability to kind of dive into stuff rigorously. But it's been fun to also dive into a modality of connecting more directly with the public in podcasts and op eds Magazine articles, radio. So that's. That's been an interesting experience. If, you know, if the psychology of war hasn't changed since Homer, the ethics of AI move a little more quickly.

Yeah. Well, the technology of war has kept pace, and there's lots of AI in warfare now. The other thing I'm reminded of, and maybe it segues into AI is. Do you know the book the Crazy Ape by Albert Sven Yorgi?

I do not.

This was like a classic sort of book from. I think it was published in 1970, like a real kind of manifesto. And he was a Nobel Prize winner, I think, for discovering vitamin C or something like that. But he wrote a whole book called the Crazy Ape, which exactly was why do we Keep Waging war? And one of the quotes that I like in it, which I'm not going to be able to remember directly, is he says, like, we face the world with this caveman's brain, and it's not equipped to solve problems better, basically. Maybe there's some hope, you know, if you're hopeful about AI, there's some hope that maybe it can augment. Farah, where do you want to start? I know we talked about kind of maybe starting generally with AI because it's such a broad term.

Right. I. Well, actually, I am kind of curious about near what you were saying, like, the more technological aspects, because I don't know that much about what's out there. I mean, the last time and that I really checked, there were things. And I guess when we talk about the idea of a chatbot, to me, it seems like technology has taken something that was traditionally considered pretty sacred, like that sort of patient clinician relationship, and it has morphed it into something very different. So through a lot of these online services, I guess what I'm wondering. But those are still like, real people. I don't know if they augment any of their treatment or what's on the horizon.

But before we dive into therapy and AI, I think can we start with kind of a broad overview of what do we actually mean by AI? Because there's, like, machine learning, artificial intelligence, there's large language models which are essentially kind of pattern generators. And then there are things like versus AI, which is based on, like, Bayesian epistemology, which is supposed to replicate thinking and maybe code in kind of human motivational systems. So, Nir, I wanted to ask you kind of like, what for, you know, for a more general listener, what. What actually is AI nowadays? And people talk about AGI, like artificial general intelligence. So what are we Actually dealing with when, before we approach, like AI therapy, what. What are the possibilities for that?

Right. So I guess on the broadest level, as an umbrella explanation, it's a set of technologies that is capable of performing cognitive tasks that were previously the preserve of humans, broadly speaking. In all of these, the methodology is a pattern recognition kind of technology. So these are systems that are trained on large sets of data and are capable of recognizing patterns in them, and given a kind of model or a set of instructions, generating predictions on the basis of them. Up until a few years ago, mainly what the general public wouldn't be familiar with were sort of recommendation engines or recommendation algorithms, like the kind that you would see on Netflix or on Amazon, that which would recommend what next book or next show you would watch. But also, you know, who would be a good candidate to get a mortgage based on a certain kind of both personal and demographic data? Who would be a good candidate to be hired for a job? In recent years, as you were mentioning Grant, the idea of generative AI came online. And the difference there was that the same kind of technology, based on a much greater computing power and the capacity to process much larger data sets, is now able to kind of change interface and generate responses to user prompts. And so you get systems that don't only find patterns in existing data, but find patterns in large amounts, large enough amounts of text, basically the entire Internet, large enough amounts of images, large enough amounts of video so that in response to prompts, they can use the sort of statistical prediction analysis to, you know, give you content essentially to simulate, mimic speech, mimic video generation, mimic images, etc. So that's the kind of generative AI that everybody's talking about and that has these potentially very strange and somewhat disturbing applications in psychotherapy, among other things.

Yeah, I'm aware of some AI models that are supposed to actually think, say, based on Bayesian or probabilistic thinking and actually mimic the way natural brains operate. And, and even, you know, we were speaking with Mark Solmes recently, he may be working with Carl Friston on a kind of an AI which could even have like core human emotions baked into it. Right. Analogous way that we think human beings do. Like, why not put that into an AI? And I'm bringing it up, Farrah, because I think it contextualizes the discussion of AI therapy. When we get into the nitty gritty of like, can you have a relationship with an AI? What would it mean? Are the emotions there? Can there be empathy?

Yeah. If you go into the paid version of ChatGPT, one of its new models is supposedly able to mimic thinking as well. I mean, you know, the terms get a little fuzzy because even when it comes to human beings, we don't have outstanding definitions of what it means for a human to think. We famously can't explain consciousness in human beings. So let alone. It's not clear that we can explain it in AI. So I guess my own way of describing it is that you increasingly are having models that can look like they're thinking, can seem like they're thinking, in which you have an increasingly harder time telling apart the difference between the kind of results that they produce and humans produce, whether or not that actually means that they're thinking. Yeah. Most computer scientists will tell you that the AGIs have no intentions, don't actually think in the way that human beings think, don't have actual emotions. But again, you can sort of copy the structure of some emotions. And if you're sort of behavioralistically inclined, you know, at some point you'd say that it doesn't matter.

Yeah, that comes down to a philosophical debate that we. We won't get into here, but things like David Chalmers and the hard problem of consciousness. And is consciousness sort of just an illusion or not? Or does it have causal impact? How do you know whether another human being has any subjectivity?

Right.

You know, can't quite know yet, but it does make. It does make sense in terms of what you were bringing up Farah, about. Okay, what about this particular human relationship of psychotherapy?

Well, I think it's important to note that there are different types of psychotherapy and that in. In most of the ones that we believe to be effective, that the sort of person to person, the therapeutic alliance, that is really important, and it's a big predictor of success. I don't know that it's like that for every single manualized treatment. So I think there may be a role for AI, but it's not the same role that there is for a human therapist.

Yeah, I mean, there's. If. If we start sort of, if we zoom out a little bit, there's an entire field of sort of parasocial relationships with AI so kind of so essentially kind of social having a relationship. We can drop the social having a relationship with AI. So companies that try to market the ability to, you know, have a relationship with AI now there's a few areas in which this happens. So increasingly there's sort of AI romantic partners that a company like Character AI is developing. There's the use of AI for grief bots. Where you would upload the social media activity of a departed person that was close to you and start having conversations with them. And you can imagine combining, combining that with video modality and you would have pretty lively conversations with them. And a few years ago, by the way, there was a Black Mirror episode called Be Right Back that was based on that premise. And it seemed insane at the time, and it seems a lot less insane now.

Well, can I jump in? There's precedent adjacent to AI. So, for example, you know, you may have seen, is there people who are in love with a real, a real doll, they call it objectum sexuality. People who feel a human emotional connection to an object. It could be a humanoid figure, like a doll, it could be a car. And then the parasocial relationships have also been studied. I'm curious about how you see parallels here with the way people may have. The way people may have relationships with a celebrity. Like, you know, you feel like you have a relationship, right.

The difference here is that you might be more justified in feeling that you have the relationship with the AI. So you can enter. I mean, look, we anthropomorphize technology all the time, right? We give names to our cars and, you know, we talk to our boats, et cetera, et cetera, et cetera. In places like Japan, where there's a relatively large elderly population and not always enough people to care for them, there are robotic devices that help in the care of the elderly. And it turns out that it doesn't take a lot for the older people who interact with these technologies to sort of start treating them like, you know, they're kids and they don't look anything like humans. They don't interact in any way like humans. So I mean, you know, if you've ever had, you know, if you've ever loved a pet, you know how easy it is for us to anthropomorphize non human beings and how easy it is for us to. And the same applies to technology. So. So broadly speaking, yeah, we have relationships with technology all the time. We project onto them. Now what generative AI brings into that picture is now the technology that we're so prone historically to anthropomorphize, anyway now begins sounding and looking and behaving like it's a lot more human, so it's a lot easier to anthropomorphize it. All you have to do is sort of imagine the combination of, you know, a chatgpt with, for example, there's this. There's this company in England that Makes these robots called the Amica robots. These are very human like robots. So if you imagine an amica robot with a whole variety of facial expressions run by ChatGPT, then you have essentially a very, very human, like, $. And you really are already, if you. I'm sure you've seen that movie her, you really are in that territory. So those are, you know, people feel like those are actual relationships now. People feel a lot of things that they're wrong about, but their justification for feeling that their actual relationship is different from, you know, saying that you have an actual relationship with your car.

There's another movie, and I'm curious if either of you guys have seen it, called Lars and the Real Girl.

I have not seen that.

Okay, so that predates her and basically stars Ryan Gosling. And he is the character who has a lot of trouble actually connecting with others. He has major trauma and difficulty maintaining social connections. And he orders this doll and has a delusion around it, but everybody sort of understands that it's delusional and they sort of support him and actually a really beautiful movie because they allow that to be a bridge, right, to having connections with other humans. And I think that that's where. What I wonder about with technology and sort of the connectivity, these parasocial relationships, is there a way to harness, I don't know either the ethics around this.

Technology that could be a therapeutic application. You're kind of saying, like, if someone has difficulty with intimacy, a lot of people with trauma, you know, first book I co authored was with relationships in mind, is called Irrelationship, how we use dysfunctional relationships to hide from intimacy. And Farah, you're bringing this up, which near, I think you're bringing up too, is, okay, there's a therapeutic application which is kind of like exposure therapy. So, you know, it's a low risk thing to learn how to relate. And then. But you're saying then you want to transfer that onto, you know, actual people versus kind of using it as a substitute. So if the experience of intimacy feels real near. But you. You have this kind of the. This fact that, okay, it's a human or it's not a human. That's an irreducible truth. Right? It's either a human being or it's not a human being. Who is the therapy bot. And if you can't tell the difference, you know, then that AI is passing what they call the Turing Test. Right? Indistinguishable. But my. My sense is that you have a position, which I tend to agree with, that Even if it's exactly the same, if you know that it's not a human being, it makes it different.

Well, I mean, first of all, just to sort of, on the most kind of concrete level, the context in which you wouldn't be able to tell the difference are limited still. So you can't go into an in person session with an AI chatbot. Yeah. Not be able to tell the difference. Right. So that doesn't exist with video. It's getting lot better. So if you are doing a sort of better health type video chat, I don't think it's there quite yet, but it's a lot closer to being there than it was a year ago. The sort of latency problems that AI voice use has, the. Which had to do with the delay that makes things awkward, are being resolved. If you've used the ChatGPT's advanced voice mode, you know, you've seen that it responds pretty immediately. So it's getting better in that sense. So that's kind of, you know. Would it pass the Turing test in person? It would never pass the Turing test on virtual platforms. Under some circumstances it would. In a text exchange, it certainly would. Perhaps in a voice call, it could. Or at least is getting there. And we know that people do therapy, not just in person, especially since the pandemic. I guess in the points that both of you have raised, you've raised two questions. One is two separate questions. One is this kind of a useful tool for somebody who has limitations in or problems interacting with humans as a kind of training wheel, as it were. And the other is in completely replacing human therapy in both cases. I mean, so if we stay with the training wheel metaphor, the training wheel has to be on an actual bike. I don't think it's a great training wheel because it's still available in a very different kind of way than humans are. It's still incapable of empathy and recognition in a way that human beings are. It's incapable of caring about you in the way that human beings are. So the question is, do you want to. And is it ethical to train someone, somebody to interact with, you know, a sort of technological entity that has no boundaries and that doesn't really have skills that are relevant for interacting with the human world? You'd almost be teaching them to do the wrong kind of thing. Right. And then therapy replacement is a very different thing altogether. So my sense about these things is, I mean, look, you know, it's hard because on the one hand, there's a loneliness crisis. There's not enough Therapists, people need a lot of help. And you're saying, well, you know, you can't get to a therapist in the real world, so talk to a chatbot. And the chatbot that you'll talk to, it'll always be available, it will always be nice. It will always say that what you're saying is really smart and really interesting and on and on and on it goes. And you know, is that a great modality to be trained into given what you actually have to face in the world?

No, I don't think so. But I would imagine that the chatbot could also, and I don't know, maybe this, it's not there yet, sort of be a coach. So one thing that I think helps a lot with social anxiety. Yeah, like a coach is like the skills training, right. Where a lot of times people will find themselves in different social situations and they're really not sure of the etiquette. What is important here culturally or what is important emotionally? How do I comport myself in XYZ situation? And I think, I would imagine that a chatbot could be helpful.

Kind of like a virtual Cyrano de Berger act, like telling the patient what to say. But I think a lot of therapists who are psychoanalytically oriented and near. I think, I think you're sympathetic and I'm a psychoanalyst and Farah is psychoanalytic would say that, okay, you can call it therapy, but skill training and coaching really isn't therapy. Therapy really is kind of deep and organic and co constructed and emerging. And so you might say, well, yeah, those types of chatbots are really helpful, but they're not a substitute for the real human relationship. The thought I'm having is that when you talk with a human being, you share an existential condition, you know, that the other person is mortal like you and you don't necessarily have that same feeling. And you also assume a level of self reflective similarity. You assume that the other person empathizes with you and you empathize with them. Now you might feel empathy from a good AI, but you're never going to believe that it truly empathizes with you the way another mortal human being will. I think so. I'm curious. Think about that in light of what Farah is bringing up.

Yeah, I mean, psychotherapy is a relationship. It's a kind of relationship. You know, maybe some psychoanalysts think that it's the most important kind of relationship. You know, I'll, I'll let the two of you be the judge of that, but it's a relationship. And you can ask if you can have a relationship with a chatbot. And I think in the end, I think you can't. Can you be friends with a chatbot? Can you be lovers with a chatbot? Can a chatbot be your mentor and your teacher? I think, as Farrah was saying, it can help you learn things about friendship, it can help you learn things about romantic relationships. It can help you as a sort of adjacent tutor to a real tutor. But no, you can't have a relationship with it because, Grant, what you said, you don't share an existential condition with it. It's a creature that can't really take your interests into mind or care about them because it's essentially a large word prediction machine, a little bit like Google spits out what you're about to put into a query. It's that large. Large, large.

You could simulate that like they're training LLMs on therapy transcripts. And it could act like it was in the same existential condition. But it occurs to me as we're talking and curious what you think that one thing, like, because I've dealt with like chatbots on Amazon, like voice chat bots, and it's very different from dealing with a human customer service agent because I'm aware that I can like hurt customer service agents feelings. I hold a customer service agent morally accountable in a different way than a chatbot. And I'm curious in the therapy relationship, what you think about that? That you kind of. People protect their therapist. They're worried about hurting their feelings. You're not going to have that with. With an AGI? Probably, yeah.

But if they can talk, you're not. And it's probably not great to have a therapist whose feelings you only think that you need to protect all the time. But the process of wanting to protect the therapist's feelings, knowing that that's what you're doing and being able to tell them that that's what you're doing and hear back that, you know, there's probably a set of reasons why you're doing that is, you know, a fundamental learning process. And I don't think you can have it with the chatbot. So it's the sort of existential condition, what you called an existential condition. The fact that, you know, the chatbot just. I'm sorry, the fact that you're a therapist, just like you can be, can have an off day, can be unfocused, can be, you know, a jerk sometimes, can be petty, can have their own set of narcissistic issues. It's actually all of these sort of shortcomings that if you can have a conversation about are the moments when something actually happens in therapy. It's less true of a chatbot. Now, one could respond to that by saying you could potentially calibrate a chatbot to doing that. You could sort of put in some kind of random pettiness generator into your chat GPT5, when that comes online, you would still know that it was randomly generated to do that. And I think that, you know, that makes a difference. Look, it's a little bit like a broader question, you know, about AI, which one of my colleagues very cleverly pointed out is that in the world that's completely mediated by AI, you lose serendipity. So, for example, a lot of our best and most meaningful stories have a serendipitous element into in them. Like your stories about how you met your partner, how you found your hobby, would have this. I went here meaning to do X and then accidentally went into room Y and found person B. And I never meant to. And I went into, you know, Barnes and Noble looking for a philosophy book. I found a gardening book. I became a gardener, gardening enthusiast by accident. You know, I took the scenic route because I got lost. Now, the point of AI is to eliminate all of that because it is not efficient. Right. You know, if you use ways, then the AI is gonna give you the best possible way to get somewhere. And a recommendation engine on Amazon or on Netflix is going to tell you the show that you're likely to enjoy. And the same is true of Spotify. And increasingly, they're very good. And they would be correct. And the result would be that you wouldn't find out stuff by accident in a way that makes accidents meaningful for us. And the response would be, well, all right, we'll put a randomness generator into it. And you would. But it would still still be mechanically generated randomness, which does not have the existential half the factual randomness. The same would be true about the therapeutic relationship, is my sense there needs to be. Spontaneity is the word I guess, I've been looking for in the background. And large language models don't have spontaneity because it's an organic quality.

I think when you mentioned spontaneity, we also think about play. Yeah, that that can be an important element. Without it, yeah. I do feel like something is lost. And I think that when these very real moments of pettiness or our own issues or things come up, part of the work of therapy is Understanding them together.

Yeah.

And joking. Also, when you said pettiness, I wanted to say Tom pettiness. I like classic rock, and sometimes when I'm working with people, I'll bring up song associations. And I guess you could program an AI to do stuff like that, but it would still be trying to optimize in a way that might never quite feel human.

Yeah. Think about our. Think about. This is a bit of a therapy move to pull on you guys, but think about our conversation right now. Right? It has its own evolution. Began a little stiffly because we don't know each other, and a little awkwardly. And then it has these kind of, you know, leaps in the dynamic of the conversation. And in those leaps, both the three of us and whoever is listening to them, they latch onto something because something human is going on and that is still impossible to simulate.

I'm actually an avatar. This company actually wanted to do a 3D thing of me for sort of promotional stuff, and I didn't feel comfortable with it, but. Yeah, you'd be. That'd be terrifying, right, if I said, oh, actually, I. I am, you know, a. An instantiation of the. Of the Grant Brenner personality.

Right. But, you know, that being said, all of this being said, especially with, I mean, look, with enough Grant Bremer video material for that company to work with and with enough Grant Bremer audio for them to train their models on, and with enough Grant Bremer text, depressingly for all of us, they can come up with a pretty compelling avatar. And if you combine that with what we talked about earlier, namely that, you know, even if the avatar was crappy, our tendency to anthropomorphize makes up the difference pretty quickly. Even if it was, like, really crappy, you could very easily imagine a set of patients that would say, it's good enough, it's fine. I feel heard more by the Grant Bremer avatar than I do by my flesh and blood friends.

It might be less narcissistic than me.

In some ways, or maybe more. Right.

We do fill in the gaps. What do you think about. Because I know we're running short on time, I wanted to ask you about the other side of the couch for a sec. Farah, maybe you have some thoughts about this. Like the way therapy, an AI could serve as a kind of a prosthesis to help a therapist do a better job. You know, mention something the therapist didn't think of, maybe provide information, serve, you know, to remember things that the therapist thought didn't, or notice expressions on the patient's face and bring them, you know, basically whispering in the therapist's ear, maybe even reading the patient's physiological markers or brain scans and directing like the therapy.

I think that's. I mean, I think that's not only possible, but that's already happening. So there is a. The technology is being used in these kind of adjunctive kind of ways, and it's very good at content analysis. It's increasingly good at voice tone analysis, at facial expression analysis. So you could imagine it being used in those kind of ways. By the way, there's other contexts also, except from therapy, where it's being used. Being used in similar kind of ways diagnostically.

You know, I've seen patients who say they have something pretty obvious like obsessive compulsive disorder or adhd, and they worked with someone for many years who never really diagnosed them. And then when you diagnose, you can treat. You know, I can see a clear application where, you know, AI could say, hey, don't. Don't miss this diagnosis.

But. Yeah, yeah. I mean, yes. Although even in. I mean, look, incidentally, and I've just written about this recently, I think that in the medical context and in specifically the drug discovery context is one area where AI has genuine transformative promise. I think it's largely hyped up. In other areas, the case hasn't been made. I mean, it's impressive, but not everything that's impressive is good in those areas. I think it's actually, you know, quite promising. Even there, though, sir, part of what we're seeing is that in diagnostics, doctors tend to defer to it and often say, you know, often kind of prefer its questionable diagnosis to their often better diagnosis. You know, even in these contexts where I agree. Grant that it's promising, there is going to be a deskilling effect that you should take into consideration. It will. Knowing that it's there and increasingly relying on it will gradually erode the kind of spidey sense that is pretty crucial for, I think, clinicians. But, yeah, I think that's a. That's a less controversial use case. Assuming that the technology has enough good data that it's trained on, et cetera, et cetera, et cetera, which is also a question. I just want to go back quickly, though, to the question that I asked you before. If a patient does, say, robot Grant or robot Farah is good enough for me, and I feel that I can tell them things that I don't tell an actual therapist, and I can, and I feel they care about Me more than my actual friends do. And I prefer this. What would you say to them? I mean, I know what I would say, but what would you say?

I would say that's not the ultimate goal. Right. And that we can't lose sight of the importance of that human connection and how we need to be fostering that. I think it's nice to have something that makes people feel good. There are a lot of things that make people feel good all the time. How and when we sort of prescribe and use and incorporate them is important, but it. It's not a substitute for real treatment. That's what I think.

Yeah, I would agree. And I would say you can be wrong about thinking that you have a relationship. You can say, x is my friend and just be wrong about that. These are not completely, you know, I mean, this goes all the way back to Aristotle. What it means to be a friend is not completely a subjective proposition.

Yeah. Well, we're sort of wrapping up for now, but this has been really, really interesting. Thank you for joining us.

My pleasure. Thanks for inviting me.

Yeah. Where can our listeners learn more about.

You and your work and your podcast?

And your podcast?

Yeah. Well, some of your listeners might like. Prosthetic Gods is a relatively new podcast that I'm doing with my colleague J. Hughes. J is a sociologist and what you call a transhumanist. Somebody who's very optimistic about technology and its capacity to transform society and lives for the better. As you could probably tell from this conversation, I'm a lot more skeptical. So basically, this is an opportunity for me and Jay to fight about technology. The two middle aged muppets yelling at each other.

Technologically, it's a great dynamic. And we'll share the links as well. In the spirit of Doorknob comments, I guess I'll ask a question that was on my list, but we don't have time to get to, which is, would you rather have a bad human therapist or a good AI therapist? Or a parent, for that matter?

Yeah, that's great. That's great. I would opt out if those were my options.

You would raise yourself?

Yeah, I'd go read a book, probably. Honestly, I'd go watch tv. But I should say I'd go read.

A book that reminds me of back to the beginning with Freud where he analyzed himself right, and then bootstrap psychoanalysis into existence. So thanks, Farah. Thanks, Nir. It's been a pleasure.

Thank you both. Pleasure is all mine.

Take care.

Thank you. Thank you.

Remember, the Doorknob Comments podcast is not medical advice. If you may be in need of professional assistance, please seek consultation without delay.