Clark and Fischer (C&F) claim that people represent social robots as depictions of other humans, and not as autonomous social entities. We argue this framework for understanding human perceptions of – and interactions with – robots is limited and limiting. Instead, an eclectic approach drawing upon psychology, social neuroscience, and human–robot interaction (HRI) will best serve empirical progress as robots' social capabilities evolve.

We agree that for some people, and in particular contexts, certain robots can be seen as representing the intentions and actions of a human principal (e.g., operator/engineer). Our central argument, however, is that such a framework for understanding HRIs is not universal and may become irrelevant as increasingly intelligent and autonomous social robots are realised.

In serving their claim, the authors draw upon Wizard-of-Oz approaches commonly used in HRI research (where a person teleoperates a robot) to categorise robots alongside ventriloquist dummies as examples of “interactive” depictions, which are a step above “staged” (e.g., puppets) and “static” (e.g., statues) depictions within their taxonomy. However, the fact that a robot “depicts” human intentions/actions in reality does not mean people perceive it as such. An overlooked feature of the Wizard-of-Oz approach is the use of Turing deception, in which people believe the robot operates autonomously (Kelley, Reference Kelley1984). We argue that under many circumstances, humans do perceive social robots as autonomous, intentional agents, even when they are not. In fact, several studies have demonstrated direct consequences on subjective experiences, behaviour and neural processing during interactions with virtual agents and robots depending on whether or not people believe an agent is human-controlled (Caruana & McArthur, Reference Caruana and McArthur2019; Caruana, Spirou, & Brock, Reference Caruana, Spirou and Brock2017; Cross, Ramsey, Liepelt, Prinz, & de C Hamilton, Reference Cross, Ramsey, Liepelt, Prinz and de C Hamilton2016; Schellen & Wykowska, Reference Schellen and Wykowska2019). These studies (and others) show that under some conditions, people represent artificial agents as human depictions, and sometimes not. Thus, it remains unclear how the depictions framework can resolve such findings if human depictions are always in play. If the “variety” or “proximity” of the human depiction shapes experiences with robots, this requires further specification.

This also highlights the important roles played by knowledge and beliefs in shaping robot representations; that is, what a person believes a robot can do, which is often quite separate from what a robot can actually do. The authors touch on the issue of pretense, and correctly state that what children know and believe about robots is unclear. We would go further to state that this holds for people of all ages, highlighting another HRI research challenge unresolved by the depictions framework. Specifically, it is reasonable to assume that most people have clearly defined, relatively invariant, top-down knowledge cues concerning puppets' or ventriloquist dummies' autonomy and sentience. The presence of strings and/or the close proximity of the ventriloquist offer bottom-up cues that activate knowledge of these agents being directly operated by their human “principal.” The same, however, cannot be said of social robots, whose relationship with their principal(s) can be far more distant, ambiguous, opaque, or complex – especially upon first encounter. Social robots are also more novel and varied than puppets or ventriloquist dummies, and continue to evolve as technologies develop, further fuelling this ambiguity. The depictions framework does not accommodate for this ambiguity in robot agency, nor the variability in the kinds of cues humans rely on to resolve it. Further, many individuals are naive about the current state of robot capabilities, or biased by representations of autonomous social robots in popular media (cf. Cross & Ramsey, Reference Cross and Ramsey2021). Indeed, our own recent research encourages the hypothesis that some children may have rather realistic ideas about the autonomy and limitations of robots (Caruana, Moffat, Blanco, & Cross, Reference Caruana, Moffat, Blanco and Cross2022). Furthermore, many contexts exist for applying socially assistive robots (e.g., education, health, and aged care) where users may be more likely to overestimate robots' autonomy, and perhaps less likely to see them as depictions of humans (e.g., young children, the elderly, and individuals with intellectual disabilities).

Rather than focusing on questions of depiction and how clearly or accurately people associate a robot with its human engineer(s), we argue that thornier challenges arise from issues related to variability in HRI across (1) individual differences (e.g., personality, knowledge, attribution styles, education, cognitive ability); (2) robot form (e.g., zoomorphic, mechanoid, humanoid, size, composition), and function (e.g., verbal, mobile, expressive); and (3) application domains. Together, this considerable variation and complexity presents deep challenges to building a robust knowledge base related to social encounters between humans and robots (Cross & Ramsey, Reference Cross and Ramsey2021), and it remains unclear how this is resolved within the depictions framework. We argue that this problem requires a multidisciplinary, eclectic approach receptive to insights gained from the previous approaches that C&F summarise and dismiss.

We would further argue that these approaches have not been fully represented in this review, and that they continue to bear fruit in explaining variability across HRIs. For instance, the review of the “trait attribution” approach loosely references Epley, Waytz, Akalis, and Cacioppo's (Reference Epley, Waytz, Akalis and Cacioppo2008) concepts of “elicited knowledge” and “effectance motivation” for explaining why humans may ascribe human-like agency/intentions to objects. However, key to this approach is the idea that significant individual differences exist in people's tendencies to anthropomorphise, which these factors attempt to explain (e.g., Neave, Jackson, Saxton, & Hönekopp, Reference Neave, Jackson, Saxton and Hönekopp2015). Another overlooked component of Epley et al.' framework concerns “sociality motivation.” This refers to one's drive to be socially connected to others, and is argued to interact with the abovementioned factors to influence anthropomorphism, while also being influenced by other contextual or dispositional factors (e.g., subjective loneliness, social isolation, anxiety, personality, culture, etc.). While we fully acknowledge that none of the extant approaches for understanding HRI offers complete explanations for the “social artifact” problem, they nonetheless offer useful frameworks for understanding how some variables shape people's interactions with robots. They also continue to inspire new empirical questions that advance our knowledge of the factors that shape HRIs. To us, it remains unclear how the depictions framework offers a solution to the inadequacies of extant approaches, or hypotheses that will help advance the field.