Volume 3, Issue 4: Winter 2022. DOI: 10.1037/tmb0000097
As robots are increasingly designed and marketed for social adoption, questions emerge about manufacturers’ ethical responsibilities for ensuring continuity of their creations. However, addressing such questions first requires advancing empirical understandings of how humans experience robots’ metaphorical death (i.e., functional cessation). This study advances that aim by exploring an interstitial: public reactions to a known robot’s functional cessation (RFC). Focusing on a specific RFC event—the official announced demise of the “Opportunity” Mars rover—this study unpacked experience dimensions by inductively analyzing public tweets about the event. Semantic network analysis of N = 10,303 tweets is interpreted as indicating 11 key expressed themes (robot legacy, journey, last words, persistence, contributions, and death finality; physiological and emotional reactions; formal and meme-based salutations and memorializing). Interpretation of network content suggests three potential overarching orientations toward the RFC event: cessation as death, as milestone, and as material loss.
Keywords: social robots, death, parasocial relationships, brokenness, meaning-making
Acknowledgements: The author thanks Alexandra Sakkis for her assistance with data processing. Thanks to the Texas Tech College of Media & Communication, where a portion of this work was done. The author has no perceived or potential conflicts of interest. All data and analysis documentation are available in the online materials at https://osf.io/dpreu/.
Disclosures: The author has no conflicts of interest to disclose.
Open Science Disclosures: The data are available at https://bit.ly/robdeath
Correspondence concerning this article should be addressed to Jaime Banks, School of Information Studies, Syracuse University, 343 Hinds Hall, Syracuse, NY 13244, United States. Email: [email protected].
From Johnny 5’s cries of “No Disassemble!” (Turman & Badham, 1986) to the vengeance-fueled extraction of Ultron’s power cell (Feige, 2015), depictions of robots’ apparent deaths are common popular media fare. Audiences see robots stop functioning as they are powered down, unplugged, dissembled, bludgeoned, melted, blasted, and hacked, but also as they malfunction, degrade, and self-sacrifice. Just as fictional depictions can contribute to people’s understandings of what robots are and how they function (Banks, 2020) and cease to function, understandings may also be shaped by news media depictions of events in which actual robots cease to function—or by direct observations of a robot’s demise in everyday life (e.g., purposeful destruction, breakage, corruption, or obsolescence). Further, many robots require continuous support or server connectivity, but purveying companies cease operations. For instance, the now-defunct Jibo was billed as a robot to be brought into the home (Hodson, 2014), but families were left mourning their companions’ near-complete loss of functionality when Jibo’s company was acquired and its supporting servers were shutdown (e.g., Carman, 2019). Such scenarios raise questions about robot creators’ practical and ethical responsibilities to ensure the continuity of their creations (Tomlinson, 1999).
Before such dilemmas can be broached, however, it is first necessary to develop a more complete empirical understanding of how people experience the ostensible “death” of robots (i.e., robot functional cessation [RFC]). The aim of this work, then, is to explore an RFC event to explore the nature of these experiences and directions that must be better understood—an effort in preliminary reconnaissance laying the groundwork for later descriptive and explanatory work (see Casula et al., 2021). In particular, this work focuses on social robots—those (semi) autonomous machines with capacities to simulate human states and behaviors (see Sarrica et al., 2020). Given that social robots are not yet widely adopted at this point, people who have acquired social robots already are likely classified as innovators on the diffusion of innovations curve (i.e., the earliest 2.5% of adopters tend to be high status, educated, and risk tolerant; Rogers, 1962). Considering the experiences of that small subset of possible adopters would limit the utility of this exploratory work—intended to discover patterns in lay individuals’ experiences of RFC—as it may drive future investigations. To bridge the gap between limited adoption and the aim of better understanding a wide range of orientations to technology, this investigation considers global public reactions to an RFC event in which humans had a parasocial connection to a robot. That is, they experienced the robot’s existence indirectly through media in ways that can promote feelings of intimacy and attachment (Horton & Wohl, 1956) and that can engender distress when that connection is lost (cf. Eyal & Cohen, 2006). By unpacking this interstitial—the experience of mediated RFC—this study takes a step toward better understanding the expressed experience of a popular robot’s functional cessation as a springboard toward future inquiry and related practice.
Technologies are made to function in the service of human desires. As a matter of physical properties, innovation, cultural change, or a host of other factors, technologies ultimately stop functioning and thus stop serving those desires (Morador, 2015). In other words, a particular kind of attachment is broken (cf. Carpenter, 2016). Little is yet empirically understood about how humans experience that functional cessation. In the absence of an established framework for considering RFC, it is useful to tentatively draw from Dennett’s (1971, 1973, 1987) notion of stances, or strategies for interpretation (widely considered in philosophical and empirical treatments of human–robot interaction; e.g., Papagni & Koeszegi, 2021). Dennett contends that people take one of four possible stances—physical, design, intentional, personal—in understanding the behavior of things in their environments. Each stance (in correspondence with extant scholarship on human experience of other technologies) suggests a different potential for how people may experience RFC—when a robot’s behaviors stop altogether—based on interpretive differences in construing the robot variably as a designed something to a morally engaged someone.
First, the physical stance is one in which people explain behaviors based on knowledge about the laws of nature and the exact nature or state of the behaving system—principally in explaining why or how a system has malfunctioned (Dennett, 1971, 1973), taking up as a matter of concern the physical properties, states, configurations, and operations of the robot. From this stance, RFC may be understood as a breaking down of the robot’s constitutive physical structure and perceived as a material brokenness or degradation (Morador, 2015). A robot that is broken, rusted, corrupted, or otherwise degraded (Banks, 2021a) may be said to have ceased functioning through compromised integrity. Alternately, RFC could from this stance be seen as a deviation from its known and “lawful patterns” (Rouse, 2021). For instance, consider a swarm of drones acting in concert and one drone breaks from the swarm because it shorted out when struck by an errant raindrop: Although the individual drones continue to operate, the swarm may be said to have ceased its aggregative function when a subcomponent malfunctions. Broadly, an RFC seen in this way may be the breaking of attachment based on expected reliability and utility (e.g., Schifferstein & Zwartkruis-Pelgrim, 2008).
The design stance is one that draws on assumptions about a system’s purpose: People predict a robot’s behaviors based on knowledge or inference of what it should be doing by its design—for complex machines, inclusive of their inferred programming (Dennett, 1971) without necessarily having knowledge of its workings or programming (Dennett, 1973). For instance, a robot that carries a tray may be thought to be designed for the purpose of carrying food, and one with wings is meant to fly. By the way of this stance, RFC may be a halting of or deviation from its inferred purpose. This aligns with normative understandings of brokenness—the halting of or deviation from purposive function of (usually) doing things better or more easily than a human could (Morador, 2015). In this sense, a design-deviating machine may be interpreted as obsolete (Slade, 2006), useless, broken in relation to utility for a specific context (Morador, 2015), or faulty (in hardware, software, logic, or behavior; Geiskkovitch & Young, 2020). It may also reflect a violating of expectations for performance (Sturdee et al., 2020, April), becoming opaque in its lack of seamless performance (Clark, 2004), and otherwise “demanding unwanted effort of the user” such as requiring maintenance (Salvia et al., 2015) or adaptation (Schaub et al., 2014). Thus, it is possible that such an orientation results in RFC experiences that are broken attachments to expectations, routines, or self-relevance (e.g., Borup et al., 2006).
In contrast to the first two stances, which engage functional and utilitarian views of machines (see Lee et al., 2011), the intentional stance is one by which a system’s behavior is explained based on ascriptions of thoughts, beliefs, desires, or intentions as the cause of action (Dennett, 1973). Specifically, a system must have an epistemic possession of information (an actual belief, as with humans, or an “analogue of belief” as with machines; Dennett, 1971, p. 90) and goal directedness. That is, people understand a robot’s behavior based on assumptions that it intends to do what it does, based on a reverse engineering of what a good or reasonable intention would be for an action (Dennett, 1973). From this stance, RFC may be interpreted as though the robot no longer thinks, believes, desires, or is otherwise mindful in its behavior; this may be the breaking of an attachment based on perceived sociality, or the interpreted tendency toward or fitness for engaging in social relations, for example, performed roles, esthetics, and personality (see Carpenter, 2009). In some sense, this may be akin to organic forms of death as a loss of mindful experience (Machado, 1999), but need not be seen overtly as such. A robot may be seen as a system that is “so complex, and yet so organized, that we find it convenient, explanatory, pragmatically necessary … to treat it as if it had beliefs and desires and was rational” (Dennett, 1971, pp. 91–92, emphasis added). It may not necessarily presume a loss of (seeming) aliveness, since one can lose the ability to make decisions without losing life—for instance a comatose organism or a computer that has inconsistent or insufficient data (e.g., having a virus or being hacked, cf. Salem et al., 2015). Indeed, this liminal characterization aligns somewhat with extant scholarship suggesting that robots are seen as occupying an animate space somewhere between humans and objects (Carter et al., 2020).
Finally, Dennett suggests that it may be possible that people take a personal stance toward systems in which the intentional stance is augmented by “the annexation of moral commitment to the Intentional” (Dennett, 1973, p. 167). In other words, orienting oneself toward a robot in this fashion would be to not only infer its intentionality but also positioning oneself to have regard for it. From this stance, RFC could be considered a loss of the robot’s personhood—the termination of subjective experience and identity that may be at all regarded. This could be a figurative understanding since humans are limited in their abilities to understanding things outside the bounds of human experience (Bogost, 2012) and since life and death are easily apprehended metaphors for technological continuance and loss (Lynch & Matthews, 2018). However, it could also be experienced more literally, as people may cultivate deep attachments to machines (White & Katsuno, 2021) and see them as meaningful moral patients (Banks, 2021b). The experience of RFC as meaningful death of a machine person could be taken up as evidence of the experience of the machine’s constructed aliveness—a loss of attachment to someone (Tomlinson, 1999).
Alternately, it is possibly that RFCs could be experienced in a cross-stance fashion given that robots may manifest an in-betweenness—both alive and not (Carter et al., 2020; Kahn et al., 2011; Kahn et al., 2012)—in ways that can impact expectations and interactions, for example, prompting reduced fault tolerance (Koda et al., 2016), RFC could be experienced according to hybridizations of those possibilities described above. It could also be that RFC is experienced as the cessation of something different altogether—for instance, as a loss of animacy (Okanda et al., 2019), meaningful kinetics (Bartneck et al., 2009), some affectation or status (Niess & Woźniak, 2020), or some other machine-native properties (cf. Vasylkiv et al., 2020). Given these broad potentials—and the practical and ethical implications thereof—I ask (RQ1): How do people experience the functional cessation of a robot?
Given the relative lack of empirical understanding of RFC and the lack of widespread robot adoption, this study takes an exploratory, naturalistic, and inferential approach as an early step. An instance of actual robot death was purposively selected, Twitter content related to that event was comprehensively scraped, and that corpus of tweets was subjected to semantic network analysis to explore latent clusters of concepts illuminating the ways people experience RFC. This approach allows for an induction of semantic patterns surrounding a naturally occurring event, without intervention and as expressed in users’ own words.
Given that social robots are not yet widely adopted, this exploratory investigation relied on an RFC event that was widely accessible to a range of people. This focal event was National Aeronautics and Space Administration’s (NASA’s) announcement of the completion of the Opportunity Rover’s 15-year mission to Mars and accompanying recognition of its functional cessation. On February 13, 2019, NASA tweeted a Rover-produced photo of Martian landscape with a single pair of tracks, along with a testament to records broken, miles traveled, and discoveries made (NASA, 2019). This announcement followed more than 8 months of noncommunication following a dust storm that prevented the rover from charging its batteries; the agency hoped that skies would clear and Opportunity could “phone home” but the rover never recovered (NASA, 2018). The event was selected for analysis because the activities of and surrounding the rover created conditions for a range of possible interpretations of its cessation: Its activities were widely covered by international media such that people could form a parasocial relation (an internalized sense of knowing without actually knowing; Horton & Wohl, 1956), the rover’s extended tenure created the conditions under which it would have been possible for people to see it as having a life, and NASA’s explicit and implicit anthropomorphizing of the robot (i.e., through affectionate naming and first-person accounts) allowed for audiences to potentially anthropomorphize the robot. Importantly, although these conditions allowed for the possibility for a cessation to be experienced as death, they did not necessitate such a reaction as the robot and its activities were highly machinic in nature.
Focusing on laypersons’ personal expressions around the RFC (i.e., as an externalization of internal experience; see de Graaf & Malle, 2019), this study analyzed public tweets. Focusing on Twitter-presented expressions allowed for concentrated and comprehensive scraping of expressions during the time around the event (4 days beginning with the day of the official announcement) specific to the event (based on dedicated hashtags). The hashtags were as follows: #ThanksOppy (NASA’s hashtag marking the event), #RIPOppy, and #GoodnightOppy (as anthropomorphizing references), #Oppy (the rover’s affectionate though not anthropomorphizing nickname), and #OpportunityRover (as a more neutral, non-anthropomorphizing tag). From those parameters, a comprehensive scrape of all tweets during that period was drawn.
Separate archival searches of Twitter were conducted for each hashtag, using academic research credentials (https://developer.twitter.com/en/products/twitter-api/academic-research) through the Postman client (http://postman.co). This process resulted in a corpus of 167,938 tweets.
Tweets were copied to an Excel (Version 2205) file for preanalysis processing. Specifically, data were cleaned by: de-duplicating tweets (as some tweets had multiple hashtags), removing tweets with no original content (e.g., rote retweets with no comment) or only single words (not appropriate for the term co-occurrence analysis), removing tweets containing only hashtags and no other content (i.e., no original experience expression), removing tweets not in English and stripping non-English from multilanguage tweets (as analysis is constrained to a single language), stripping hashtags, URLs, account tags, emojis, and errant characters (to focus on personal expressions), and standardized language to ensure the analysis tool would treat all semantically aligned words the same, despite mis/spellings (e.g., changing “rovr” to “rover,” “omg” to “oh my god”). Regional slang was adapted to common U.S. English vernacular (e.g., “bodged” to “a mess,” and “goodest” to “best”). Then, a round of content scrubbing was conducted, removing news reel or thread promotions (i.e., those that pointed to but did not comment on threads), spam, promotions or listicles, hashtag highjacks (e.g., political commentary). Finally, because the analysis tool unitizes at the sentence level and the aim here was to examine whole tweets as expression units, tweets were stripped of tweet-internal terminating punctuation (period, exclamation point, question mark, quotes) and a period added to the end of each tweet. This ensured that each tweet would be treated as a unit and the inclusive words would be associated in analysis. Finally, in a last pass, references to robot- and agency-specific terms (opportunity, curiosity, insight, robot, rover, Mars, NASA, JPL) were stripped as artifacts of the focal event that would force artificial constellations around those terms and mask other sentiment patterns. This process resulted in N = 10,303 tweets retained for analysis, and the data set is available in the online materials.
Because of the broad phenomenological potentials for RFC, an exploratory and inductive approach was taken. Induction was privileged over an overtly deductive (i.e., theory-driven) approach to avoid assumptions that experiences of human death or any one of Dennett’s stances necessarily applies to experiences of RFC. Specifically, analysis was conducted in two stages: First, a software-aided induction of word and concept associations in the data, and second, an interpretive analysis of the induced concept clusters. See Table 1, for a summary of the relationships among data and analytic units across the two stages.
Summary of Data and Analytic Units in Two-Stage Software-Aided Inductive Analysis
Stage 1 analytical function (Leximancer)
Stage 2 analytical function (researcher)
Discrete words with the text corpus.
The most granular unit considered by the topical algorithm, evaluated for frequency, co-occurrence with other terms, and weighted for their likelihood of predicting a latent concept. Serve as markers of data units in which a concept is likely to appear.
Set of weighted terms that tend to travel together throughout the text.
Represent a latent idea present in the corpus, constellated with other co-occurring concepts to construct clusters. Named automatically according to the highest weighted predictive term.
Concept meanings are validated by ensuring predictive terms and data excerpts support the automatically assigned name.
Set of concepts that tend to travel together thoughout the text.
Represent a latent construct present in the corpus, as a constellation of lower order concepts. Hits (ns) signal prevalence of constitutive terms, indicating data units (here, tweets) containing terms associated with the construct.
Themes (core construct meanings) are interpreted by iteratively evaluating the clusters’ content, constitutive concepts and weighted predictive terms, and source data excerpts.
Note. All data and analysis outputs are available in the online materials for this project: https://osf.io/dpreu (Banks, 2021a). The study was designated as nonhuman subjects research by the author’s institutional review board.
In the first stage, systematic semantic analysis was conducted using Leximancer (Smith & Humphreys, 2006), which uses natural language processing to discover latent concepts embedded in large, unstructured data sets. This discovery emerges as the software learns from the data set itself such that meaning models are grounded in the semantics of the language and context in question (rather than on dictionaries built for other concepts, which introduces error and bias; Smith, 2020). This initial step was taken to mitigate known challenges with wholly subjective interpretive analyses (e.g., inadvertent cherry-picking of attractive data, accessibility of self-relevant patterns over others, attention to linear narratives at the expense of more discrete, distributed patterns) by systematizing the identification of objective patterns in co-occurrence of terms within the corpus. Using a topical algorithm (which emphasizes direct relationships and maximizes discriminating differences; Leximancer, 2021), the program identifies concept seed words (terms occurring most frequently) and learns how those terms are associated with others in the corpus, ultimately building a thesaurus for each concept—a set of terms that maximally predict the presence of a latent concept. In turn, the program determines concepts that tend to co-occur within textual units (here, sentences) and aggregates them as clusters representing higher order constructs. The output of this induction process is a map visualizing the concepts as they structure the clusters, and accompanying catalogs of clusters, those clusters’ constitutive concepts, those concepts’ predictive terms (weighted), as well as catalogs of text excerpts (here, tweets) from which those concepts and clusters were induced. Each cluster represents empirically distinct aggregations of co-occurring terms, although the exact configuration of model selected for analysis relies on researcher evaluation of a model with cluster granularity balances parsimony and interpretability (see online materials). As the aim in the present analysis was to identify the possible characteristic of RFC experiences, analysis of content was privileged (over network-structure particulars) toward discovering people’s meaning-making around the event.
In the second stage, I conducted iterative interpretive analyses separately for each cluster to determine the theme it represents—that is, to identify the overarching phenomenon depicted by the cluster’s content and source data that help to generate insight with respect to the research question (cf. Braun & Clarke, 2006). Analysis began by confirming that the automated concept names were face valid in relation to the predictive terms and source data; all names were determined to be reflective as there were no evident deviations in term usage compared to the names. Then, I undertook the iterative process of tacking back and forth among the map visualization (considering concepts and their relations), the predictive terms for concepts (considering the varied language contributing to the predictions), and the source data (considering term usage in context, and the apparent meaning of tweet language in relation to the research question). In this process, I engaged notions from Dennett’s framework (e.g., animacy, brokenness, purpose) as sensitizing concepts—nondefinitive, nonprescriptive ideas that function as initial, intuited directions (unavoidable as a function of prior knowledge; Bowen, 2006, see also Casula et al., 2021). Having then a tentative characterization of the theme represented by each cluster, I consulted relevant literature to name and define the themes in relation to the constellated maps, terms, and data (cf. Braun & Clarke, 2006).
The analysis resulted in the induction of 52 concepts with term frequencies varying from n = 70 (stars) to n = 1,300 (crying) comprising 11 clusters. (See Figure 1 and online materials, for complete catalog of terms, counts, and co-occurrence matrix.) By tacking between the concepts, clusters, and source data, themes were interpreted and named in correspondence with the extant literature surrounding human experiences of machines, death, loss, remembrance, and other relevant concepts. Themes are reported here from most to least prevalent, where “hits” (ns) refer to the number of tweets that included a concept in that cluster; words in “quotes” are data excerpts and those presented in italics are identified key terms (i.e., those that predict the presence of the latent concept) or their variants (e.g., thank and thanks). In some cases, direct quotes are edited slightly or abridged to avoid identifiability via searches of public tweets.
The most prominent cluster consisted of accounts of the robot’s legacy—the gifts of memory, knowledge, and experience that the robot and its makers offered humankind. Most prominently these were constellated around expressions of appreciation, fondness, and inspiration as people testified as to the importance of the robot’s mission and to its evidencing of human ingenuity. People offered thanks to both the robot and its creators, principally for inspiring people to appreciate science but also thanks in appreciation for the rover as a feat of engineering. In some cases, appreciation was directed to the rover and manifested as thanks for your service. There were often-echoed sentiments along the lines of “farewell and thanks for all the science.” By way of personal gifts, some recounted specific, fond memories—mostly recounting the rover’s initial landing—but sometimes more esoteric, as with a recollection of showing them “Martian blueberries,” spherical mineral accumulations (Catling, 2004). Alongside these testaments to legacy, tweets included expression of awe and amazement at the machine itself and of its “hardworking” efforts as it drove advancement through exploration.
Tweets giving rise to this theme focused on the robot’s ground-breaking and unexpected journey to and on the red planet, persisting far beyond its planned 90-day mission and stretching to nearly 15 years in duration. Tweets ranged from merely referencing to expressing wonderment at the declared completion of the mission that it was finally understood to be no longer operational—its accomplishments deemed trailblazing and heartwarming. Some linked the robot’s journey to their own (e.g., being “a kid when it landed”) and others commented more manifestly on the nature of the robot’s time and demise on the red planet: “dust storms” and “miles explored.” The robot was a “great explorer” that restored faith and preceded humanity into space. These accounts and reflections on the robot’s journey were often paired with hopes or promises that the robot would be reunited with humanity, that “we may see that little rover again within our lifetimes.”
Crying, here, is a theme representing people’s expressed affective and physiological responses to the event, as well as their commentary around those responses. This theme is notable in that it is constituted by a single concept—crying—that makes up more than 12% of all hits. Some merely admitted to crying while some expressed exasperation (often with a good deal of profanity, e.g., “for fuck’s sake”). Tweeters indicated they were crying in offices, restaurants, in class, at desks, and into their cocktails, in the morning and at awake at midnight. They asked why they were doing it, owned it, critiqued it, and invited others to cry along. There was repeated crying (e.g., at initial news, at each tribute read, and on revisiting threads), a good deal of “ugly crying,” and rhetorical cover-ups: hiding with a *cough, *getting (Martian) dust in the eye, dealing with hormones, or musing about “onion-cutting ninjas.” Some characterized it as a “millennial experience” to cry over a robot.
Tweets giving rise to this theme largely served to recognize and digitally mark the moment and their (usually sad) reaction to it. Expressions often included notes that today was “a sad day” that seemed quieter and feeling unprepared for that kind of reaction. Some made soundtracks to mark the occasion or otherwise paid homage to the robot as a space cowboy or space doggo as self-professed space nuts, nerds, or enthusiasts. Others called their mums to share the news, while others were moved to reflection on the nature of their own short existence. A few reflected on the nature of an evolved “relationship to machines”—especially as they felt a loss of a piece of childhood—and some imagined the robot hanging out with Laika the space dog (see Caswell, 2018).
Tweets giving rise to this theme almost entirely referred to a science reporter’s “poetic translation” (Margolis, 2019a) of a project official’s accounting of the robot’s “last words”: “My battery is low and it’s getting dark” (Margolis, 2019b). The phrase was compared to the poetry of Emily Dickinson, Radiohead, and HAL9000, and likened to feelings about Brexit and Valentine’s day. Notably, the robot did not actually make any such expression. Rather, 8 months before the announced end of mission, the robot had sent its last message with technical details indicating low solar-power production and high-atmospheric opacity (MERB Nasa Archive [Sol 5111], 2018). There was some semantic overlap in these references to recitations of a Dylan Thomas poem: “do not go gentle into that good night … wise men at their end know dark is right because their words had forked no lightning” (see Thomas, 1937/n.d.). References to these two interpreted and ascribed dying sentiments were sometimes accompanied by further interpretations of the conditions of functional cessation (usually being engulfed by a dust storm), by notes that the ostensible final words were gut-wrenching, and by calls to send Matt Damon on a rescue mission (a reference to 2015 film The Martian in which Damon’s character is stranded on Mars) because “surely we can do better” for the robot. In a few cases, posters reminded others that the last words would have been “lines of code” written by a human, and challenged them to consider if they would feel as sad if their car battery died.
This theme emerged from tweets that conveyed the notion of a (generally figurative) persistence—that the robot has traveled a long road and existence and that a rest had been earned. There were direct wishes for the robot to rest easy, rest well, and rest in honor and in peace “among the stars.” Others wished for it to have pleasant dreams or to “live long and prosper.” Sometimes, this persistence in a metaphorical life and rest was accompanied by promises or hopes that the robot might be recognized properly, that it may “rest until one day humanity can reclaim it and give it the proper honor it deserves.”
The sentiment indicating “rest in peace” manifested as a theme separate from the preceding persistence theme due to the difference in form; however, it is notable that the acronym seemed to also accompany a different kind of sentiment. “RIP” was embedded in tweets alongside affectations such as “little buddy” and “little rocket man” and “a good boi,” and lauding that the robot had been brave, diligent, and venerable. RIP also tended to accompany references to Bradbury’s (1945/2012, p. 96) The Martian Chronicles narrating an exploration and settlement of Mars: “the first lonely ones had to stand by themselves.” Notably, since single-word tweets were removed from the data set (including many singular expressions of “RIP,” the expression itself was far more prevalent than is represented here.
This theme was constellated around expressions of the finality of the robot’s long-running mission and the months-long attempt to restore communication after the harsh Martian season. Posters referenced the final transmission including the last uplink sending Billie Holiday’s 1944 song “I’ll Be Seeing You” including the lyrics “I’ll be looking at the moon but I’ll be seeing you.” More generally, sentiments detailed notions of finally declaring the mission complete, declaring the robot dead, and saying goodbye. There were explicit reflections on the finality of it. Some marked the event as final by sending an e-postcard (see https://mars.nasa.gov/participate/postcard/opportunity-rover/), while others joked that the robot was “fine until” an event, for example, “someone sent it the 2016 election data.” A small portion of tweets in this cluster pointed to when humans finally reach it again, for instance with hopes that it would be like “booting up the old betamax.”
This theme comprises general references to posters’ emotional reactions as they were “legit emotional,” wrecks, distraught, compromised, and overly so—especially in reference to the final-words interpretation thread, songs, comics. Some were accompanied by calls to rescue the robot (e.g., “we gotta go to save the homie”), while a number blamed the Pixar film WALL-E for their emotionality.
This theme comprises what we’ll call here “memed” salutations—those that bid farewell by drawing on sayings from popular or class culture. These were generally “good night” references, calling the robot “sweet prince” (from Shakespeare’s Hamlet) and the variation “sweet robot,” often paired with “Godspeed.” There was some overlap here with the aforementioned “do not go gentle into that good night” and continuation of Billie Holiday’s song with “when the night is new …” such that night is generally a metaphor engaged in reference to death.
This theme focuses on the specific contribution (rather than the more general legacy) of having served a bridging function in showing people another, new, distant world—one “beyond our world.” The robot specifically made that world seem not so distant or alien and helped a world of people become more curious, such that the world together “mourned a machine” (for better or worse).
This exploratory investigation found 11 key themes in peoples’ expression of their experiences with a robot’s functional cessation. Focusing on tweets posted in the 4 days following the Opportunity Rover’s announced end of mission, themes emphasize posters’ accounting of the robot’s legacy, journey, last words, figurative persistence, contributions, and the finality of its death; they also focused on the poster’s own emotional reactions to the event and extensions of formal and meme-based salutations. A heuristic evaluation of themes and their corresponding cluster-relations allows for inferencing of three potential orientations toward the RFC event—with implications for advancing theory around how people understand machines’ functional cessation as variations on death.
From the analysis, I here infer some candidates for formal RFC orientations: RFC as a death, as a milestone, and as a material loss. Considering the themes induced in this study, the analysis tool used here lays out clusters in such a fashion that “distance is difference” (Smith, 2020): The further apart the clusters, the less likely they are to co-occur and so less likely to be experientially related. Considering the overall layout of the semantic network map in relation to source data, three possible orientations toward the RFC event are qualitatively discernible. These candidate orientations are presented here not as definitive, but as prudent targets for future study and as a springboard for how RFC experiences may be conceptualized according to what kind of function and attachment is thought to be lost in the event.
First and most clearly, the Last Words and Crying grouping is an otherwise relatively disconnected pair of themes in which there is an expressed affective response, largely in relation to hearing the robot’s interpreted “last words.” This orientation may be characterized as experiencing the RFC as death (although the presentation of that death was a fiction). That is, the RFC was for them an emotional event, but in actuality they were responding to a manufactured, anthropomorphizing narrative around the RFC. Considering this concept-pair’s distance from other clusters, people experiencing the RFC in this fashion were unlikely to consider the robot’s functional and productive journey or its long-standing contributions to scientific efforts. In other words, they appear to have experienced the RFC as a dramatized death, perhaps most aligned with the intentional stance, that is, understanding through belief/goal ascription, (Dennett, 1971) in that the sadness was in response to the robot’s narrated mindfulness in observing its impending doom (see Malle, 2019)—signaled by a low battery and a dark sky. In addition to seeing the event as a loss of a mindful agent, the prevalence of crying could indicate they experienced this from a personal stance (i.e., moral regard for an intentional system; Dennett, 1987), where harm coming to another (including a robot) may in some cases be seen as a violation of its well-being as a moral patient (Banks, 2021c).
The second potential orientation emerges through the constellation of the journey, persistence, contribution, and legacy clusters and may be characterized as experiencing the RFC as a milestone. Collectively, these clusters suggest affirmations of importance, as the “working” (by robot, by scientists/engineers, and by the robot and humans together) concept links together the journey and legacy clusters. The persistence cluster could be interpreted as a fictional engagement of the event (i.e., the robot is “resting”). However, it was also an anthropomorphization that affirmed the value of the robot’s contributions—that its end was acceptable and notable given what the machine had accomplished—and characterized the RFC as merely a stepping stone allowing humans to eventually join the robot. This candidate orientation arguably aligns with the design stance (i.e., explanation through assumed purpose; Dennett, 1971), in that the robot was designed to advance science through exploration and that designed function has terminated; however, the sentiment emerging from that stance is less about the loss of the function and more about the value of that function. Just as people tend to reflect on values of lives lost and may be meaningfully impacted by the loss (Kernberg, 2010), some people seemed to react to the RFC event by acknowledging significance and signaling eudaimonic reflection. That is, expressions signaled appreciation for the cumulative account of the robot’s functioning—an explorer ambassador, a “master of the two worlds” (Campbell, 1949)—rather than emphasizing the cessation itself.
Finally, a more central cluster consisting of Emotionality, moment marked, and finality may be characterized as experiencing the RFC as a material loss, where posters were serving as witnesses to the moment—of the end, of its personal impact and of its meaning. The concepts within these clusters suggest more reflection and reflexiveness on the notion of ending (vs. progress and contributions in the milestone orientation). In this orientation, data suggest that the RFC need not necessarily be interpreted as a death to be meaningful, only that it be seen as something final and depriving. In some ways, these themes together may be likened to the physical stance (i.e., explanation through system states and natures; Dennett, 1987) in that there’s a recognition of the end of the robot as a thing-in-itself. But perhaps more notably, this orientation focuses more on the state and role of the human observer in relation to the RFC event. That is, the orientation emphasizes the loss of continuity but that is accompanied by heightened awareness their own experience in relation to the event. Notably, it may also be that where Emotionality intersects with the marking of a final moment, that the marking may signify more personal stance-styled reactions (i.e., moral commitment to the system; Dennett, 1987) given that the tweet itself may simultaneously be a witnessing and a call for others to witness their own regard for the machine (see Richardson & Schankweiler, 2019). Of note is Dennett’s (1973) contention that the personal stance requires adoption of the intentional stance such that in this case, robots may represent an exception to that rule or it may be that the moral commitment here should be interpreted as assigned to the value of the robot and not the robot itself.
Past human–technology interaction work has addressed death but generally only insofar as machines should engage human death (e.g., Kaptelinin, 2016, May; Massimi & Charise, 2009). Given this study’s evidenced potential for humans to experience RFC in ways akin to experiences of death, it is prudent that studies of human-machine interaction also engage thanatosensitivity (thoughtfulness around death; Massimi & Charise, 2009) in relation to the machines that humans encounter—especially as the death of some digital agents is often framed in terms of human failure (cf. Cuerdo & Melcer, 2020).
Importantly, the candidate orientations above are heuristic, qualitative inductions from the cluster associations, and their potential to underpin a framework for understanding human experiences of RFC requires exploration of other RFC events for a range of robot morphologies, uses cases, cessation conditions, and immediate (vs. mediated) human–robot relations. In advancing that trajectory, it is useful to consider the extent to which these candidate orientations may feature gradations of a shared property—a variable that may help to explain differences in human experience, especially one that may be theoretically relevant in accounting for people who were not moved to react at all.
Social cognitive dynamics appear to be at play in RFC experience, given the prevalence of anthropomorphizing language (e.g., the robot “closed his eyes” and was often gendered male or female). The extent to which people may anthropomorphize the robot (and so see death more in line with intentional/moral stances) may depend on individuals’ propensities for humanizing objects or extending that tendency to robots (Lee et al., 2011) and on varied attitudes and beliefs about the potentials of technology (Martínez-Córcoles et al., 2017). Other individual differences of import may be trait empathy given links to moral expansiveness to include nonhumans (Crimston et al., 2016) and people variably see robots as potential moral patients (one whose well-being matters and for which humans are responsible; Banks, 2021b). Just as some humans may be predisposed, some machine features are more or less likely to invoke anthropomorphizing reactions since people may interpret animacy through kinetics or properties that specify aliveness and intentionality (see Bartneck et al., 2009): smooth motions, intuitive or reasonable behaviors, physicality, personal narrative, adaptation, growth, apparent self-interestedness, and even machine-native (i.e., nonhuman/nonanimal) gesturing (Parviainen et al., 2019; Tomlinson, 1999; Vasylkiv et al., 2020). As a social robot conveys these cues, they are interpreted heuristically (see Sundar, 2020) as recognizable indicators of aliveness and may extend the interpretation to things not alive.
Interpretations of aliveness may be a precondition for a perception of RFC as death. Importantly, these person perceptions are not to be confused with delusions of fiction as reality. Rather, human brains are predisposed to seeing humanness and intentionality in the nonhuman (Dacey, 2017; Spunt et al., 2015) such that it is reasonable that people would intuit human-like death when a machine is represented as exhibiting the hallmarks of humanness. After all, aliveness is argued to be a construction, scaffolded through sociality scaffolded by competent adherence to human norms: “An object is alive in a society if it is treated like it is alive” (Tomlinson, 1999, p. 329; see also Malle & Scheutz, 2019). The same may be said of death, especially as media framings of robots make fuzzy the demarcations between human and machine, and contribute to laypersons’ enduring ideas of what machines are and how they function (Banks, 2020) or how they may be fit to participate in activities traditionally the domain of mortals (Guzman, 2021).
In addition to considerations for machine and interaction design, it may be prudent to consider experiences of aliveness and death in relation to how social robots are marketed. Although this robot was arguably in the public sphere, people nonetheless experienced a parasocial relation with it to some degree—they felt a social connection despite not having actually met the robot (Horton & Wohl, 1956). The robot’s likeness to robots in popular media (most prominently Pixar’s Wall-E, with a similar morphology) likely promoted senses of knowing—indeed, some people overtly blamed the character for their unexpected emotional reactions. If feelings of loss or grief may be generated as a result of a parasocial loss (Cohen, 2003), then it is possible that actual, immediate relations with a robot may elicit even stronger grief upon loss. If robots are framed as prosocial, productive family members and engaged as such, then it may be reasonably questioned whether it is incumbent upon robot producers to sustain those robots’ functions. That is, should robot manufacturers be responsible for preventing RFC as a matter of avoiding more interpersonal-type attachments?
Finally, much scholarship and design work focus on social robots’ utility and sociality—that is, the ways they may be well designed to be useful and relatable. The present data suggest that, although those foci are important for adoption and engagement, the broader meaning of human–robot relations may extend beyond the immediate interactions. In the present case, RFC did include some self-interested expressions and even reflections on human mortality, but also included suggestions of self-transcendent emotions—those in which the subject looks outside themselves and their own interests as they make meaning of a situation (Stellar et al., 2017). Here, in particular, data suggest affective elevation, hope, awe, empathy, and appreciation known to sometimes result from mediated experiences (Raney et al., 2020). Moreover, there were indications that people experienced the RFC in relation to greater scientific and humanistic movements and saw the robot as being in service to those movements. Although expressions were not always accurate, there did seem to be a fostered appreciation for science and engineering—and science appreciation and curiosity can meaningfully emerge in the absence of scientific literacy (Kahan et al., 2017). Altogether, such patterns signal that RFC events may stand in contrast to both human frailty and humankind’s ostensible continuity and, so, promote self-transcendent reflection on topics beyond the event itself.
This study carries a number of limitations common to this kind of inductive work: Design decisions necessarily limited the generalizability of findings in focusing on a specific RFC event, specific platform, and specific hashtags, so (as discussed above) replication and extension are required. Most evidently, this work is an interpretive analysis—an analysis by a single researcher in order to do the complex work of meaning induction—and replication and development is required; the aim here was to detect possibilities for future research. Further, data were limited to public tweets in English, where people who do not use Twitter, do not choose to post publicly, and are situated in non-English-speaking cultures may have different perspectives on technology and (especially culturally) on death. Importantly, the data here would not include individuals who did not care to or were not moved enough to post about the RFC, so patterns discerned do address how people performed their experience but do not account for the ways people quietly, conservatively experienced it—or how they may critique or be disinterested in the event. Finally, given the performative nature of tweets, the current design cannot disentangle whether tweets were meaningful reflections on the RFC event or if they were following norms and practices of the Twitter para event linked to the RFC event. Perhaps most notably, this study focused on an RFC event in which most people had only a potentially parasocial relation with a celebritized robot through (social) media representations, and an open question is whether these patterns could translate to more immediate and interdependent relationships with robots. However, this work detects the possibilities for experiencing robot death in more fully social relations since parasocial relations are built on many of the same mechanics as social relations.
The principal aim of this work was to detect signals to follow in future research into human experiences of RFC. Future investigations should build on the present inferences to attend to the limitations above and to unpack the nuances and effects of RFC experiences. First, future work should explore whether and how this study’s observed patterns may (not) manifest with other robots under other circumstances. In particular, it should consider degrees of designed anthropomorphism since behaviors and appearances may motivate different interpretations of aliveness (Blut et al., 2021) and so different permutations on perceptions of death (or its analogues). Of particular importance is consideration of varied robot morphologies, affordances for social cueing, and performative capabilities. Second, it is necessary to test the extent to which these RFC interpretations may (not) manifest for physically copresent robots with which humans may have a fully social and more experientially immediate relation (see Mara et al., 2021), and the likely influences of particular contextual dynamics (e.g., norms, environments, outcomes; Banks et al., 2021). Finally, whether relations are social or parasocial, the present study’s findings point to a need to better understand how institutional and popular narratives may contribute to how people interpret RFC events; here, there were clear connections between expressions and both NASA framings of the robot (both as Oppy through Twitter and about Oppy through institutional communications) and popular culture touchpoints (e.g., films and music). Illuminating the nature of human RFC experiences in these ways will help to inform policy and design practice around social(ized) robots in ways that may maximize productive experiences and minimize problematic ones.
The present study explored expressed experiences of a robot’s functional cessation event (i.e., “death”), finding semantic patterns that suggest three orientations to the event: cessation as death, as a milestone, as a material loss (in approximate correspondence with Dennett’s intentional/moral, design, and physical stances, respectively). Although considerations of the “leaky distinctions” (Haraway, 1991) between humans and machines tend to focus on animacy, evidence here suggests a range of orientations to robot cessation from interpretations of human-like death to an event that calls forth humans to bear witness. Data suggest that as robots are “disturbingly lively” (Haraway, 1991) that animacy also presents the conditions under which they may also be seen as expiring along a continuum of breakage to death.