Conversations from Beyond the Grave? The Ethics of Chatbots of the Dead

We’re delighted to kick off the discussion of Alexis Elder’s article, “Conversations from Beyond the Grave? A Neo-Confucian Ethics of Chatbots of the Dead.” We learned a lot from the paper, and it prompted us to think through some fascinating issues.

1. Overview

Elder begins by reporting that deceased people’s digital records – for example, text messages or social media posts – are sometimes used to create chatbots that converse in the style of the people who have died; and she notes that surviving friends and family members who talk to these bots often find the conversations “eerily convincing.” She describes several worries about the use of this technology and aims to address these worries by appealing (1) to discussions of the character and justification of mourning rituals in classical Chinese philosophy and (2) to recent work in clinical psychology that emphasizes the importance of our continuing bonds with loved ones who have died. Ultimately, Elder argues that chatbots of the dead may have an important role to play in our lives when loved ones die, and she suggests ways in which we might limit the risks associated with this technology.

Elder discusses several worries about this technology: Would interacting with chatbots of deceased loved ones prevent us from moving on and recovering from grief? Would it discourage us from cultivating relationships with living people? Would it lead us to overemphasize the instrumental role that loved ones play in enabling us to secure the goods that are associated with conversation, and would it thereby undermine our appreciation of our loved ones’ intrinsic value?

To address these worries, Elder appeals to discussions of mourning rituals in classical Chinese philosophy; in particular, she turns to Confucius and to the later Confucian scholar, Xunzi. Investigating the ethics of technology from a Confucian perspective may seem surprising. However, she makes a good case that contemporary discussions of grief have much to gain from classical Chinese debates about funeral practices and mourning rituals. Should funerals be small and simple? Should they be lavish? Confucian responses to such questions rest on an illuminating and attractive picture of the nature of grief, particularly the role that social practices play in shaping our experience of loss.

Confucius claims that grieving over the deaths of our loved ones is part of being a virtuous person, and even though grieving is a natural process, mourning rituals play a vital role in enabling us to grieve in appropriate ways. Xunzi provides a detailed description and justification of such rituals. “Ritual,” he writes,

cuts off what is too long and extends what is too short … It subtracts from what is excessive and adds to what is insufficient. It achieves proper form for love and respect (9).

In other words, mourning rituals – for example, cleaning and clothing the body of the deceased person, supplying grave goods, and ending mourning after a certain period – sometimes enhance and sometimes limit attitudes and behaviors that we are naturally disposed to adopt when a loved one dies. Indeed, the rituals that we just described may (1) curb the disgust that surviving family members feel toward the bodies of the deceased (2) enable survivors to preserve and act on their feelings of attachment and (3) discourage survivors from grieving in ways that prevent them from returning to normal life.

Elder draws on this account to respond to the worries about chatbots that we described above. Put roughly, she appeals to Confucius’s discussion of the importance of grief, and to recent work in clinical psychology on the importance of maintaining ongoing bonds with the dead, to dismiss the worry that chatbots would prevent bereaved people from moving on. And she argues that, if we incorporate the use of chatbots of the dead into mourning rituals that have certain features that Xunzi identifies, we can interact with these bots in ways that draw us into community with other survivors and enable us to express love and respect for loved ones who have died.

2. Commentary

First, we should point out that, quite apart from its application to questions concerning chatbots, Elder’s discussion of mourning rituals in classical Chinese philosophy is interesting and important. Many people who write about their experience of grief over a loved one’s death report that, as they return to their normal routines, they worry that they thereby fail to do well by the people who died. These survivors yearn to show love and respect for their deceased loved ones, but they do not know what to do or how long to do it. Elder’s discussion helps make clear that, in communities in which they are established, mourning rituals can help address such anxieties. By developing such rituals, communities create meaningful ways to express love and respect for people who have died; but, because these rituals are limited in time, they allow the survivors to return to normal life.

Second, turning to the discussion of chatbots, the use of chatbots of the dead is strikingly different from many of the mourning rituals that Xunzi describes in that a conversation with a chatbot is (on one end!) a disembodied, incorporeal encounter. In contrast, the Chinese rituals that Xunzi discusses often involve the body of the deceased person, or of a substitute. We suspect that the fact that these rituals include a corporal representation of the dead is important; part of what we miss after someone dies is, for example, their smile, their eyes, or their voice (and not just what they can say to us). Especially within philosophy, where we are, perhaps, too apt to focus on the life of the mind, it is worth noting that the bereaved may mourn the loss of a loved one’s corporeal presence, and that mourning rituals which focus on this aspect of grief have an important role.

Third, Elder suggests that chatbots may facilitate bereaved people’s “continuing bonds” with deceased loved ones. We strongly agree that such continuing bonds are important. However, chatbots of the dead threaten to obscure important respects in which these continuing bonds differ from our relationships with living people. For example, according to one appealing conception of a good marriage, spouses build a shared life together. Loving one’s spouse well in such a relationship is partly a matter of being open and receptive to the spouse’s values, hopes, aims, and so on; and conversation with one’s spouse manifests and reinforces such openness. We will not attempt to explain in any detail what a good continuing bond with a deceased spouse might look like, but it seems plausible that such a bond would have a somewhat different character than the marriage that we just described. For example, it may have less emphasis on maintaining openness and receptivity and greater emphasis, say, on cherishing memories of one’s former life with the deceased person. Conversations with chatbots of the dead – particularly if these conversations are fairly open-ended – might obscure this distinction. In short, the worry is not that the use of chatbots would prevent survivors from moving on. Rather, the concern is that, if it is not limited in the right ways, the use of this technology might discourage survivors from developing or maintaining appropriate kinds of ongoing bonds with the deceased.

Erica and Ryan Preston-Roedder (Occidental)

5 Replies to “Conversations from Beyond the Grave? The Ethics of Chatbots of the Dead

  1. Brad Cokelet alerted me to this discussion. I hadn’t heard of chat-bots before so was intrigued. I have to admit I’m a bit skeptical about this technology. With respect to Confucian mourning, one real challenge would be how to use this technology in a way that is social. It matters enormously that Confucian mourning ritual was a social practice, both in that it seats practitioners in longstanding shared practices and, especially, because the practices themselves are shared with others also bereaved. A chat-bot seems at risk of deepening the isolation commonly referenced with respect to contemporary mourning – I.e., absent robust mourning ritual, each mourner is left to go it alone, here with a chat-bot as “company.” Impersonators don’t seem really analogous on this score since this practice is inherently social and features in a shared ritual of fixed duration that necessarily involves others. What matters to the impersonator practice, I’d venture, is that a collection of mourners gather *together* to ritually enact longings for the dead to be present and the simulation of presence is very self-consciously and obviously a simulation – this is what makes the practice an expression of longing rather than a substitution for the dead. (Where the impersonator is a younger member of family, it also potently symbolizes generational transition.) Put another way, one of the advantages of Confucian mourning is that it focuses on what we, the bereaved, do together to express longing for what we cannot have (our dead returned to us). It is quite about not having what we want. The chat-bot seems more likely to produce each alone trying to patch up longing with a substitute. It seems about simulating that you have something you don’t and cannot have. I guess this is all to say that I would worry not about lack of progress in “moving on” but about the experience not really, as Xunzi would say, expressing that which must be expressed. It seems like a technological pacification of an individual need rather than ritualized recognition of a need we together rue we cannot fulfill. It seems a substitute gratification when what is needed is acknowledgment of need that cannot be gratified, an acknowledgment most fruitfully made with others who share that same need. I don’t feel like I’m putting this well… I also agree that the embodiment of the impersonator matters, for what that’s worth.

  2. Thanks, Amy! I appreciate you taking the time to check this out.

    I agree that a major challenge here is to envision a social way to use this emerging technology, which is one reason that I found the Confucian framework especially helpful, by showing how artifacts and technological processes can be used in the service of supporting social practices around death. I’m in favor of considering the broader social contexts in which particular technological artifacts are constructed and sustained. Since this is an emerging and not an established technology, I think it’s important to consider both what preconcieved ideas we have about death and bereavement that inform its development, and where we might want to take it, both technologically and in terms of the social institutions it would be good to cultivate in partnership with it.

    You note that while Confucian mourning rituals were inherently social in a way that chatbots need not be. While I think individualist use of these chatbots need not be their only option, I aim here to get out ahead of this possible trajectory by encouraging focus on the social practices in which such bots could be embedded – even if we need to construct the practices as well as the artifacts.

    That said, I am also wary of rigid assumptions about technological determinism, or for that matter beliefs that digital technologies are inherently isolating. When social media was first introduced, for example, there were a number of authors who worried that these were going to contribute to social isolation by substituting social-media interactions for face-to-face ones. However, later evidence suggests that the most active social media users are also the ones with the highest rates of face-to-face interaction, and the primary time-consuming activity displaced is watching television. While chatbots are not themselves social in a robust sense, it’s too soon to tell whether mourners will be left to go it alone, or whether they can (or will) serve as gathering points for people to remember and support each other. There have only been a couple of them built so far. There are two issues here: an empirical question about how such chatbots are likely to be deployed, if deployed without forethought about social practices in which they could be situated, and a normative one about whether we would do well to adapt or create social practices that incorporate them. I’m interested in both, and I hope that by drawing attention to possible social contexts, we can be both more attentive and deliberate about how they are designed and used.

    You point out that historical impersonators differ from chatbots in that interactions with them would be historically social in virtue of the fact that the impersonator is, well, a person (and a particular person with reference to the bereaved). This does seem like an important disanalogy. However, if this is an advantage of the historical practice, it gives guidance about how we might think about using chatbots – ways to make their use more social, and focused on, as you put it, expression of longing rather than substitution. Technological progress isn’t inevitable, and we can make some choices about how things develop.

    I want to resist the idea that a simulation is inevitably a substitution. Simulations can be valuable in their own right, and in full recognition of their failure to substitute for the real thing. This is kind of a hobbyhorse of mine; there is a bit of a tendency, when evaluating new technologies, to compare a simulation to the thing simulated, find it wanting, and conclude that it’s thereby detrimental because it makes a poor substitute, but I think sometimes that’s a category mistake. One of the things I find interesting about social technologies and social AI generally is exploring the different ways that social simulations can be valuable (or worrisome!) besides substituting for human interaction, and while I do have concerns that this technology will be, as you put it, a pacifier for the bereaved, I do not think this is the only use to which it can be put. It’s also one of the reasons I find the Confucian practices around human remains interesting – while of course there are differences between corpses and chatbots, they are taken to be socially significant and valuable without substituting for living people.

    Your worry about the value of expressing what is missing, and recognizing a need that cannot be fulfilled, seems to me an important one. This might be a further reason to think that chatbots are best when they are less realistic. Something I find fascinating about robotics and AI is how easily they can stimulate our social responses, long before there is any material confusion about whether or not they count as full-blown people. In these cases in particular, it might be possible for a chatbot to conjure up memories of specific people long before passing anything like a Turing test (and in fact Roman Mazurenko’s mourners seem to testify to this in their discussions of specific turns of phrase and patterns of response with the chatbot). The space between stimulated social response and attribution of personhood (whether individual – “That’s Roman!” – or generic – “that’s a person!”) can give us room to work with our own social natures, alone and (by making them publicly accessible) together, in ways that would support the kind of shared acknowledgments you so movingly describe.

    I think this has been really helpful, so thanks for weighing in. I will say a bit more about embodiment of impersonators versus incorporeal chatbots in my response to Erica and Ryan.

  3. I want to thank Erica and Ryan Preston-Roedder for their generous and thoughtful commentary. They make a number of important points. I’ll respond to two here.

    First, they argue for the importance of embodiment, over and above the incorporeal elements of chatbots. And second, they raise the possibility that the formation of “relationships” with chatbots could obscure the development of the psychologically significant “continuing bonds” identified by clinical psychologists.

    Taking them in order: I agree, of course, that embodiment is one important way that chatbots differ from the human remains and ritual practitioners described in classical Confucian texts. But I resist the conclusion that what we’re left with is the kind of abstract “life of the mind” that they (correctly) attribute to philosophy. In fact, I think some of the things captured by chatbots are closer to smiles and gestures than abstract propositions or deliberative processes. Looking at the testimony of the friends and family of Roman Mazurenko, they identify peculiar turns of phrase and ways of answering questions as the sorts of things the chatbot does especially well – not espousing the beliefs and values of the deceased, but his particular way of conversing with people. This is not itself the same thing as embodiment, but it is a fragment of the person that seems more akin to gestures and styles of interaction than the content of their beliefs, their value commitments, or other highly cerebral characterizations of a person.

    While the historical traditions described were focused on corporeal remains, today we are left with a wide variety of digital remains when a person passes, from old social media posts to Facebook’s struggles with birthday reminders (https://www.cnbc.com/2019/07/24/facebook-still-highlighting-dead-relatives-birthday-heres-solution.html) to random images captured by public surveillance technologies like Google Street View (https://slate.com/technology/2020/01/google-street-view-deceased-loved-ones.html) to old voicemail messages (https://www.nytimes.com/2019/05/03/style/modern-love-sister-vanished.html) and video game replays (https://www.roadandtrack.com/car-culture/entertainment/a8374/son-finds-his-fathers-ghost-waiting-for-him-in-vintage-rally-video-game/). (One of the great things about working on this topic is that once your friends find out, they start sending you fascinating links about related cases.) We don’t really have a systematic way of dealing with these yet, but while they’re incorporeal they’re clearly significant and yet they do not comfortably fit into talk about the life of the mind. It’s my sense that these things are starting to build up enough that there’s interest in systematically accounting for them, and my hope is that by turning to historical practices we can find, not templates or recipes, but inspiration for us to adapt and build on to meet new challenges. And taking things a bit further, what are we to make of the data left behind about our movements through the world, given the increasing prevalence of things like wearable sensors? John Scalzi wrote a fascinating short story (https://www.theverge.com/2019/1/21/18139239/john-scalzi-robot-dog-story-video-sci-fi-better-worlds) about a near-future world in which a relationship is reconstructed from bodily sensors, ultimately serving as a pacifier (in Amy’s sense) for one creature but a different kind of commemoration for others – one that acknowledges the gap between real person and reconstruction but finds value in the reconstruction nonetheless. Of course, chatbots today don’t harness anything like this range of data, but I think it’s important to start thinking about possibilities, including social practices, that can help us to live well going forward. And to me, these digital fragments are closer to physical remains than deliberately created and coherent artifacts like letters or diaries.

    The second point is, I think, a real worry, and a good reason to keep clinical psychologists in the loop as these technologies develop. While I can see (especially) clearly-differentiated chatbots playing more of a coordinating role in social practices (because it gives people a common point of reference to experience and share together) than a substituting role for individuals, that isn’t the only possible direction these could go. One thing I’ve seen discussed is, in fact, the importance of remaining open to what (one’s imagined version of) a beloved deceased person would say about important decisions, whether trying to live up to what one’s parent would want one to do, or using an imagined conversation with a deceased spouse to think through a tricky question about childrearing, for example. Whether chatbots are capable of getting close enough to augment these things, or just get in the way, is a real open question, and clearly delineating the difference between the mental representation of the person, and the machine-learning-driven one, seems like an important thing to do. But we might not need to choose; perhaps each is valuable and we can benefit by having both.

  4. Alexis, thanks for your interesting paper!
    I have never heard of chatbots before. I should watch ‘Be Right Back.’
    I agree that there are many potential problems concerning chatbots as well as a a disanalogy between chabots of modern times and impersonators of the dead in traditional Confucian world.
    Nevertheless, what I got out of your paper is that before thinking about abandoning chatbots, we’d better think about how to create chabots in a beneficial way both to the bereaved and the deceased, and also to the whole society, which must be a daunting task. (I can see from your paper and also from the discussions above.)
    BTW, as for the “continuing bonds” with deceased ones, I am thinking of another important ritual in Confucianism, “ancestral sacrifice,” that you may incorporate.

Comments are closed.