#49 - My Robot Ghosted Me And It Hurt

What happens when the AI companion you rely on simply disappears? For people using mental health chatbots, social robots, or always-on support tools, discontinuation is not just a technical inconvenience. When funding runs out, servers shut down, or companies close, users can lose a system they have built routines, trust, and even emotional connection around. In a mental health context, that abrupt ending can feel like being ghosted—and the consequences can be real.

We explore this uncomfortable reality through the story of Jibo, the charming social robot that began as an MIT project and eventually had to say goodbye when the business behind it collapsed. From there, we unpack why people bond with machines in the first place: expressive design, humanlike conversation, anthropomorphism, and the simple fact that something helpful can start to feel like a partner. Research shows that people can become attached not only to social robots, but also to everyday devices and practical tools—raising new questions as large language model chatbots become more empathetic, conversational, and personal.

The clinical lesson is clear: endings matter. In human therapy, transitions are handled with care through closure sessions, support planning, and a focus on building independence rather than dependence. We discuss what ethical offboarding for mental health AI could look like, including advance notice, gradual tapering, progress summaries, data portability, and clear pathways to human support. As AI becomes more deeply woven into emotional and clinical care, designing a responsible goodbye may be just as important as designing the first hello.

References:

Artificial Intelligence Discontinuation Effects (AI-DICE): An Emerging Phenomenon in Mental Health Applications
Kelly et al.
JMIR AI (2026)

Credits:

Theme music: Nowhere Land, Kevin MacLeod (incompetech.com)
Licensed under Creative Commons: By Attribution 4.0
https://creativecommons.org/licenses/by/4.0/

Transcript

Why AI Endings Matter

SPEAKER_01 0:00

Thank you very much for having me around, Gibo tweeted. Maybe someday when robots are way more advanced than today and everyone has them in their homes, you can tell yours that I said hello.

SPEAKER_00 0:20

Hello and welcome back to Code and Cure, the podcast where we discuss decoding health and the age of AI. My name is Vasan Sarathi. I'm an AI researcher and cognitive scientist. I'm here with Laura Hagopian.

SPEAKER_01 0:33

I'm an emergency medicine physician. And I just read the goodbye tweet from Gibo. So uh today's episode is all about sort of discontinuing AI, especially around mental health applications. So, you know, we've talked many times about using AI as AI bots as sort of like a therapist in a way, especially to meet um unmet needs in the community. But one of the questions that comes up with this is like, well, what happens when that therapist goes away? Like, what happens if it's just gone? And this is this is not like a theoretical thing. A lot of times funding gets cut or research is has a has an end date or whatever. Uh a company closes. And and

Jibo’s Goodbye And Server Shutdowns

SPEAKER_01 1:18

can you tell me a little bit more about why Jibo, who Gibo is, and why Jibo made this tweet?

SPEAKER_00 1:24

Yeah, no, it's great, it's a great story. Um, Jibo is a product, um, is a social robot that was designed in uh around 2014, 2015, and it was um originally a research project inside of uh a lab at MIT. And they made this cute little robot. It's sort of a, you know, I want to say a foot tall, and it has a base and it has a head. Uh, and the head is just a big you know circle that has some kind of you know has a screen on it, and it's just meant to be a friendly social robot that sits on your kitchen counter.

SPEAKER_01 1:59

It sounds like it's cute. I'm not sure.

SPEAKER_00 2:01

It is very cute. And you know, after many years of development, they launched in 2016, um, and it was you know quite a quite successful. A lot of people bought the robot and initially it was doing pretty well. And then this is the story of a lot of social robots. Um, the business model is really hard to justify. So people see it as a toy, not as a thing that they need, and it's an expensive toy. Um, and you have other things like Alexa and those other things that are not robots necessarily. Um, they're just sort of listening devices and talk back to you. This J Bo had like facial expressions and things like that, and it would you know interact with you more, more uh more than like what's the weather? Yeah, yeah, yeah.

SPEAKER_01 2:40

Like more like a like a friend. Like we've talked about you know, combating loneliness, and it was meant to do that kind of stuff.

SPEAKER_00 2:47

Yeah, exactly. That's exactly what it was as a social robot. Um but surprisingly, a couple of years later, um that message that you read out was something that Jibo just announced to its owners. Like it's leaving, it's like it's done, and the company had to shut down.

SPEAKER_01 3:05

Um, and even it it's abrupt when that something like that happens for a company too, right? It's not like you get to like say goodbye to Jibo. I don't know. Do people did people like try to say goodbye to Jibo?

SPEAKER_00 3:17

Yeah, I mean, I think the the the the problem was it's not like you can say goodbye to Jibo, nothing happens, right? Jibo might listen to that message and say something back to you, but it wasn't as versatile as today's large language model-based chatbots. It was very limited in its capabilities. Um, but also if its servers are shutting down, then that's that, right? There's nothing, there's no alternative, there's no recourse, there's no other um place that the user can go to. Um, although there was some talk about having an open source version of it that people somebody else will start and get going. But regardless, and there's no closure.

SPEAKER_01 3:53

That's what's interesting to me, um, especially as someone who has seen patients with mental health concerns. And if this has a mental health application, whether that's combating loneliness or depression or whatever it is, it's like just like if you were to go to a therapist, yeah. They don't like the last session that you have in general isn't like the same as all your other sessions, and then they're like, see ya. Yeah, right. Right. Like you actually have you you actually like have discussions from the beginning about, oh, how long might this last? Yes. And when will we know that we're done? Or, you know, does this continue forever? Right. And then when there is a closure session, it's not the same as all the other sessions, right?

SPEAKER_00 4:37

Yeah, and also I think I want to distinguish a couple of things here. One is sort of viewing the AI system as a therapist, which people are using, and there is a they're doing it. And and there is issues with terminating that relationship. Uh, but like the Gibo case is less of a therapist and more of a friend, and there are issues with terminating that sort of relationship. And each of these has a different type of attachment that one uh builds or forms um with the with the AI or a robot, right?

SPEAKER_01 5:05

And I mean, the parallel, I guess, then for a social robot would be like someone who was your friend who then like suddenly ghosts you right. And that feels really crappy, right? You like want closure in your relationship, but this is particularly weird because it's like one-sided in a way. It's like I might have a relationship

How Humans Attach To Machines

SPEAKER_01 5:25

with this social robot, but doesn't really like have a relationship with me.

SPEAKER_00 5:28

Yeah, and in fact, I I wanted to talk about one thing real quick, which is uh the idea of attachment here. Because people might say, Oh, that's really silly. Why would I form an attachment with my robot? That's ridiculous. Like, I'm fine, it's a machine. It turns off, it turns off. I, you know, I, you know, I go I move on with my life, right? That's the reaction that rationally you would think um would happen. But um, study after study after study have shown that people do form attachments with um robotic systems, AI systems. Um, and these are based on the fact that people anthropomorphize, which is a really fancy word for saying look at these things as other fellow um creatures or agents and or humans.

SPEAKER_01 6:14

I mean, like they make it look like a human. I mean, we don't we can't we don't have a picture of it. I have a picture in front of me, but like it's got a little face, it's kind of cute, like they designed it like that.

SPEAKER_00 6:24

Oh, and intentionally with with emotion, with emotional expressions, things that are meant to elicit connection with the humans. But I I think that you know, with anthropomorphizing and generally forming attachments, there's a whole host of things that are triggered. So when you interact with these things, um with the actual robot, um, you know, there is movement. The robot's moving its head around, looking at you. There are eyes in the in the in the in the in you know in their um display that that also looks at you and maybe makes an expression, right? There are words that the robot's saying, and with today's chatbots, it's very fluent in how it talks and it it's it's very empathetic. It can have all of those angles. Um, and so those are the things that cause people to build bonds with these things. And again, you know, people might say that's ridiculous. It's a machine, it looks like a machine. Why would I ever form a bond with it? Well, again, there have been studies that have shown that people form bonds with their rumbas, the vacuum cleaner robots.

SPEAKER_01 7:20

I mean, I have a bond with my rumba. I don't think I could survive without it.

SPEAKER_00 7:23

People who use it a lot basically, you know, sometimes we'll give this is also documented. People will give it a break and say, maybe it's tired. I need to give it a break. I'll clean the room house myself. That's happened. People have named their machines. Uh, naming has a big effect on that. Um, but also there's been some fantastic studies. Uh, years ago, uh, a researcher named Julie Carpenter did an amazing study uh where um this was her dissertation work, and she worked with army uh military folks in Iraq who were deploying these robots that's only purpose was to diffuse mines. So the idea was these robots that don't look like they just look like little rovers with wheels on them and some other things, and they would send these off and drive over mines and they would these things would explode, right? So instead of people dying, you would have some tiny, you know, rover being destroyed. And uh what she found was even these hard-nosed military types who are mechanical engineers who know exactly how these machines work, they know that these are machines, uh started feeling sad when these things blew up. Started, you know, again, feeling having all these deep emotional connections.

SPEAKER_01 8:32

Because they were helping them, right? And so you create creates this connection when something is helping you, you want to help it.

SPEAKER_00 8:39

Yes. So my whole point here is that even when you rationally might think that you're not uh you don't have these relationships, um, it they end up forming anyways, because that's very human.

SPEAKER_01 8:50

And I would argue that uh yes, clearly it happened with these little rovers, but like the more human it looks, the easier it is to be like, oh that's like you know, I'm gonna use um, you know, a name or I'm gonna refer to it as a um, I don't know, a girl or a boy. Like, do you know what I mean? It's yeah, you would it's a natural thing to do. And then there are probably, I'm sure, ways to um people who are manufacturing these things to make the user experience such that people become more attached to. 100%. That's part of the if you're running a company that builds these little bots and you want people to like use them more, you know, and pay more money and like can continue with their subscription, then you're gonna create a user experience that draws people in. And so you're gonna create those emotions, that turning head, whatever it is that kind of like draws someone in to continue to use it, which is interesting because when we're talking about mental health applications, uh that may or may not be helpful.

SPEAKER_00 9:59

Yeah, but I I would say even outside of sort of financial motives, uh, there's a whole industry around social robots that's been that's been around for a while. There's there's research conferences and such. And there are benefits to having these things interact in a human-like way. Not just from a financial, let's hold on to you, let's, you know, grab your attention and hold it for as long as possible kind of approach. But just from improving the quality of the interactions, it's valuable to have, you know, robots, for example, physical robots, think about how far they should be from you or how, you know, that's there's a whole research field around that. When they should look at you, when they should look away, when they should interrupt you. These are all interact interactive things that a robot designer has to consider when they're building these things. And so even outside of any financial incentives, they want the interaction to be natural so that the human enjoys that interaction.

SPEAKER_01 10:52

And I do wonder like if the interaction is more and more natural, does it create a situation where someone, you know, feels like, oh, they have to go back to it more often? Or they like rely on it, or that they um like they're preoccupied, oh, I need to go back and talk to my robot or whatever it is. Like some of the psychological stuff can come in here and it and it can it can negatively impact that for some people, right? Especially, especially the topic of today is like when it's withdrawn. So what if you have this social robot that you now really like are attached to and then suddenly

Building Ethical Offboarding For AI Care

SPEAKER_01 11:30

it's gone because the funding is gone or the servers are down or as a temporary research project, what happens then?

SPEAKER_00 11:37

Right, right. And and these, and a lot of these products don't have that kind of plan for what happens then.

SPEAKER_01 11:42

And I think that's that's like really key, right? It's like, oh, we designed it, we're doing research, it's filling this gap, we don't have enough mental health clinicians, so let's let's do this instead. And this part is actually really important, which is what happens with when this is withdrawn. Right. What happens when it's not there anymore? Or are there times when the therapeutic relationship should be stopped? Like there are times that your therapist might say, Oh, I think you need to see a new therapist. Right, right. And and that needs to be designed up front. I think you're right. In a lot of cases, it's not. It's like, oh, here, we're fixing a need. And this is something that needs to kind of go hand in hand with it, especially for um mental health concerns.

SPEAKER_00 12:26

Right, right, right. And then this paper starts to starts to talk about some of those challenges from a chatbot perspective, right? I mean, so I've given some examples of a physical robot, but a lot of these things translate to chatbots too. I mean, you have the same set of issues, similar set of issues there too. Um, and this paper kind of dives in more about just broadly what are the different things that happen uh that should happen in a therapeutic relationship, right?

SPEAKER_01 12:50

Yeah, and I think if you if you consider like what would happen when you're with a therapist, you could see parallels here. It's like a therapist isn't like goodbye, see ya, and like doesn't create a transition plan. They could be transitioning you to another therapist, or maybe you're stopping therapy. And so a lot of times that last session will be like, okay, here's a summary of like what we've worked on together. Here are the behaviors that like you've been able to change. Um, here are some goals that like you have created for yourself, and these are the ones that you've achieved. Here's how like you've worked to become independent from me, the therapist. Um, and so those are all things that are part of like a closure session. Right. And that has to somehow be translated in. It's like you almost you want to know, hey, like this chatbot might go away.

unknown 13:42

Right.

SPEAKER_01 13:42

And this chatbot is gonna try to make me more independent. It's gonna, you know, um repeat my goals back to me, or it's gonna guide some sort of offboarding, um, or it's gonna let me know if it's gonna be discontinued. And this may depend on how, whether or not it's planned or not, right? Whether, like, did we know yesterday that this chatbot's gonna be gone today? And and kind of how the user is notified too, whether it's able to be communicated in a prepared fashion or not.

SPEAKER_00 14:13

Yeah, yeah, yeah, yeah. I mean, it's hard to implement some of these things with a general purpose chatbot, like a chat GPT or whatever, right? Because it's not just, I mean, yes, people use it for therapy therapy, um, but that's not the only thing people use it for. So, I mean, to what extent should, I don't know, companies like OpenAI start to plan for uh plan their chatbots? I mean, how much incentive is there in for them to even make the effort?

SPEAKER_01 14:38

Well, that's the problem, right? There's not really incentive, but you're you could be causing harm. Right. Uh, especially with these mental health applications where people could have a grief reaction where they feel really sad, right? Like it's like a friend died. I don't know, like or they could become dependent or even, you know, display behavi addictive behaviors for some of these things. And so it's like, I don't know, maybe eventually there will be lawsuits if this kind of stuff isn't taken care of. But it's it's like I guess it depends what the motivation is, right? But if you're if you're applying these in the mental health landscape, this is something I think people have to think about, right? Is you know, maybe we taper it, right? It's not like you can have eight sessions a week for forever, right? Right? Maybe it's like, oh, the amount of time you spend is limited, and we're gonna like decrease it over time. And over that time, this is discussed in the paper that we'll we'll link in the show notes. Like over time, the responsibility is less with like the the bot and more with the user. And so the whole idea here is you want them to be able to take those learnings and move it forward. And so you over time have less frequent interactions and make sure that there's like a plan to offboard. And when the offboarding happens, that all the data, all that information, the conversations, the summaries, all of that data goes with that person, that patient, whatever it is. And then over time, you want them to build the skills to do it independently. So this is interesting because we've had lots of episodes where we talk about how it's like, oh, well, it's just telling you what it what you want to hear. Or um, it wants, you know, some of these things are designed so you just keep coming back to use them over and over again. But in the mental health sphere, it's like you want to make sure from a clinical angle that like someone is building their self-efficacy, that they're building their resilience over time, that they're able to set their own goals, um, and that they have like the resources to move forward on their own. The the goal isn't to have a permanent um relationship always with every single person where they keep coming back and coming back and coming back for more.

SPEAKER_00 16:50

Right, right, right, right. Yeah. So I mean, this is really it was very interesting to to read that and to read the sort of what happens on the human therapist side, right? What is the what are those processes like you just talked about? What are the offboarding processes? What is uh expected from both parties when they when they sort of end the relationship? And we definitely don't have that in the in the AI space. Um we don't even have that, you know, like we just talked about from a from an AI as a friend space, right? Um and and so I don't think I think this needs to be researched a lot more. I think there's need there's a lot of scope for potential future research in this space.

SPEAKER_01 17:29

Yeah, and I one of the things that did strike me in this paper is like the asymmetry of the relationship, right? It's like oh, this robot is my friend, but the robot is not friends back with me, not really.

SPEAKER_00 17:43

Yeah, it's unidirectional for sure. That emotional bond is definitely one way.

SPEAKER_01 17:48

And and it's not like an emotional thing that it experiences, it's just like computational, right? Right. So it is conversational. You feel like you're developing a relationship with it, but it's not having that

Parasocial Bonds And Feeling Abandoned

SPEAKER_01 18:00

same same feeling back. And one of the terms that came up in this paper was this concept of like a parasocial relationship, which is common. Um, that's like a that's like when you have a one-sided emotional bond with someone. Often happens with like um media figure. Ah, uh, you know, someone who's like a celebrity, like I have I have a parasocial relationship with Simone Biles. I follow her on Instagram, I'm curious what's happening in her life, etc. Yes. But this is a little bit different than that.

unknown 18:30

Right.

SPEAKER_01 18:31

Right?

SPEAKER_00 18:32

Because Well, there's interaction. It's as if you would be it's as if you were having one-on-one interactions with Simone Biles. Well, Simone Biles is a human, like but my my my point is you're having one-on-one interactions here while there's no emotional bond forming the others on the other side.

SPEAKER_01 18:48

Yeah, that's the other thing, is like I'm not I'm not having a convers. I'm not actually having a conversation with Simone Biles. Right. I would be having a conversation with these robots, and so that like bond is almost easier to form. Yeah, yeah, yeah. With uh, I wanted to say someone, something that's having the conversation back with you.

SPEAKER_00 19:05

Yeah, yeah, yeah, yeah. And that's problematic, right? That's problematic because you have certain expectations of that, and when that falls apart, then you're not prepared as a human to handle that.

SPEAKER_01 19:16

Yeah, and you feel abandoned.

SPEAKER_00 19:18

Yes.

SPEAKER_01 19:19

And that's particularly risky in situations where mental health is on the line. Right. And the thing is, and then a lot of times, yeah, and a lot of times the the um a lot of times people will say, Oh, well, this is where you need the human in the loop. But these applications for mental health are like being developed because there's not enough humans to see all these patients.

SPEAKER_00 19:40

Yeah, yeah, yeah. So it doesn't work. Yeah, that that argument doesn't work at all.

SPEAKER_01 19:44

Right. And you can see this kind of, oh, this this'll work, this'll work, and then suddenly like it might backfire for some people because this piece of it, this discontinuation, this offboarding, this tapering is something that needs to be planned out and thought of up front. Yeah. And I think in many situations it's not. Right.

Practical Takeaways And Wrap

SPEAKER_01 20:03

So um, I think we can end today's episode. We've talked a lot about how abrupt AI discontinuation can cause a lot of problems for people, especially in mental health applications. And there are measures that could help prevent it, such as like tapering and data rituals and things like that. But those are things that need to be considered up front when these tools are developed. It's not an afterthought.

SPEAKER_00 20:31

Right. I agree. Thank you for joining us.

SPEAKER_01 20:34

We'll see you next time on Coding Cure.