Towards an Integrative Framework for Robot Personality Research

Within human-robot interaction (HRI), research on robot personality has largely drawn on trait theories and models, such as the Big Five and OCEAN. We argue that reliance on trait models in HRI has led to a limited understanding of robot personality as a question of stable traits that can be designed into a robot plus how humans with certain traits respond to particular robots. However, trait-based approaches exist alongside other ways of understanding personality, including approaches focusing on more dynamic constructs such as adaptations and narratives. We suggest that a deep understanding of robot personality is only possible through a cross-disciplinary effort to integrate these different approaches. We propose an Integrative Framework for Robot Personality Research (IF), wherein robot personality is defined not as a property of the robot, nor of the human perceiving the robot, but as a complex assemblage of components at the intersection of robot design and human factors. With the IF, we aim to establish a common theoretical grounding for robot personality research that incorporates personality constructs beyond traits and treats these constructs as complementary and fundamentally interdependent.

undergo over time, and iii) the stories that people construe about the robot that are enabled (but not determined) by the design features.We argue robot personality is not a property that is exclusive to the robot, nor is it an outcome of humans attributing it to the robot.Rather, robot personality emerges at the intersection of i) robot design and behaviors, ii) the task and the context of deployment of the robot, and iii) human factors such as fundamental cognitive processes, human personality, human identity, and storytelling.
It is through this complex assemblage that we experience a robot as having a distinct character that is in some way like all other robots, like some other robots, and like no other robot.While robot personality and how humans experience it are especially relevant for social robots, as they are usually deined in HRI, we do not limit our framework to just these robots.Components of all four levels of robot personality that we propose may still be relevant in interactions with all kinds of robots.
In our proposed Integrative Framework for Robot Personality, four levels of robot personality difer in terms of the relative degree to which they are open to change over time.We deine these levels as follows: • Level 1. encompasses fundamental components of robot materiality (i.e.technological and morphological embodiment) that, coupled with fundamental human cognitive processes (e.g.anthropomorphising, sociomorphing) contribute to the experience of the robot as social in some degree and capacity.
• Level 2. comprises traits, as instantiated in particular (more or less) static sets of behavioral and communication cues programmed into the robot and commonly captured under the traits theories in Social Robotics (SR) and HRI studies of robot personality.
• Level 3. concerns characteristic adaptations that we interpret as more dynamic behavioral and communication cues that result from the robot learning over time by interacting with its environment.Here, we also consider 'intentions and desires' that may be purposefully rooted in a robot and that shape the robot's behaviors over time.
• Level 4. incorporates narratives about the robot.These include the narratives that people construct about their robots and that converge to a unique robot identity over time, as well as the narratives that designers, developers, and researchers generate as a kind of 'back story' that explains the robot's existence, role and that may shape the relationship with the people interacting with it.
Our aim throughout this paper is not to criticize prior work in HRI, especially trait-based studies, nor to argue they should be abandoned; we propose viewing the diferent strands of existing work as parts of a broader robot personality research agenda.To accomplish this, we need to draw lines between those domains of personality research in which the concept of a trait does, or does not play a central role.With the Integrative Framework, we (i) establish the breadth of personality beyond trait-based models, (ii) show that considering traits in isolation does not suice for an understanding of robot personality and sociality more broadly also as a property of situated interactions and relations that develop over time, and (iii) encourage more research into robot personality dimensions that are not captured by trait-based theories.We also introduce elements of a critical perspective that draws on our expertise in science and technology studies (STS) to point out opportunities and challenges to keep developing robot personality research.
Section 2.3 explains what trait-based theories are, why they are popular in psychology, and how they came to dominate HRI research.On this foundation, we go on to argue why the SR and HRI research communities would beneit from an extended view of personality that goes beyond traits approaches.In Section 2 we introduce our theoretical motivation from psychology.Then Section 2.4 presents McAdams and Pals' framework [53] to ground our suggestions for levels of robot personality in the corresponding research on human personality.On that basis, Section 3 outlines our Integrative Framework.For each of its four levels, we discuss the respective personality constructs and related components, give examples of how existing research in HRI has already addressed some aspects at this level, and propose research themes we consider relevant for further research.Finally, Sect.3.5, we emphasize how culture shapes robot personality both in the ways we design and interpret robots.We use examples to illustrate the relationship between culture and personality on each level of the Integrative Framework.

THEORETICAL BACKGROUND 2.1 Trait-based Theories in Personality Psychology
Within personality psychology, traits are commonly understood as broad dispositions that relect łthe overall style of a person's adjustment to and engagement of the social world" [53, p.207].One advantage of understanding personality in terms of traits is that they are cross-situationally consistent [2].A variety of long-term studies on the stability of personality traits indicate that trait scores are strong predictors of behavior [54, summarise evidence for both concerning the Big Five].In trait-based approaches, personality is mainly assessed using questionnaires that are evaluated based on the factor loadings of the various items.This method aims to identify dimensions of personality that are considered quantiiable and more or less independent of each other.Among other trait-based models [62, e.g. the OCEAN-model], the Big Five is one of the most widely known and applied instantiations of the trait-based approaches [28,34].This model proposes that human personality can be accounted for by individual diferences in ive personality traits: openness, conscientiousness, agreeableness, extraversion, and neuroticism.
Since being introduced, the Big Five has received both widespread support [54] and criticism [20].Criticism notwithstanding, the Big Five can be considered highly successful.For example, one review contends that it was the ive-factor model that ended a łThirty Years' War of competing trait models" [54].Recently, McRae et al. re-emphasized the premier position of the Big Five in personality psychology [55].With the Big Five as their igurehead, traits theories remain one of the most widely used approaches in personality psychology [53].

Trait-based Research on Personality in HRI
The popularity of trait-based approaches in HRI is unsurprising considering the relative ease of capturing personality dimensions with the help of questionnaires.Robert Jr et al. identiied four domains of trait-based personality research, depending on the focus of investigation: (i) human personality as a mediator of outcomes in human-robot interaction, (ii) robot personality as a mediator of outcomes of human-robot interactions, (iii) human-robot personality complementarity, and (iv) facilitating robot personality [65].Within the latter three domains of personality research in HRI, trait models commonly appear as built-in patterns of robot behavior and communication styles that are programmed to facilitate a robot's distinctive personality and social character.
Whittaker et al. give an idea of the kinds of work being done at the intersection of robot personality and trait-based models, including who used the Big Five to develop and evaluate the perceived diferences between three robot personas [96].In terms of personality, the distinct robot personas were diferentiated by a combination of humanoid features (speech, intonation) and indirect cues (colors, gestures).The researchers' intent behind the three distinct personas was to identify whether the robots would difer on measures of proactivity and emotional impact.The authors reported that participants were able to recognize the underlying personality traits of each persona, and showed clear preferences between them.Generally, it is common for the studies that draw on traits models in HRI to evaluate how robot personality traits mediate variables such as acceptance [56], afect [61], trust [27], and performance [1] 2 .
Characteristically, trait-based studies in HRI primarily focus on the traits introversion and extraversion because these can be easily implemented in observable cues like voice pitch, talking speed, gestures, and proximity to humans and others [39, 50, 57, e.g.] as illustrated by the example above.Another reason for the popularity of trait-based studies is that these align with the general positivist3 stance that dominates HRI and SR communities.That is, trait-based studies of robot personality fall into the group of studies that can be characterized by their goal to generate unambiguous, replicable, and quantiiable results.

Limitations of Trait-based Models
Notwithstanding their advantages, trait-based approaches are limited regarding the kind of knowledge they can generate [63].Importantly, these approaches are rooted in the assumption that personality can be mapped onto an externally recognizable set of cues and behaviors.This assumption falls within what Jung refers to as the 'signaling approach' in HRI [36] that makes it diicult to study situated behaviors and interactions with robots [4,76].
Doubts about the adequacy of both trait-based approaches and the signaling paradigm as frameworks for studying social interaction are raised by studies showing that situational factors are one key to social interactions and how robots are perceived.For example, Joose et al. showed that preferences for robot personalities depend łon the context of the robot's role and the stereotype perceptions people hold for certain jobsž [35, p.2134].This leads the authors to conclude that robot behaviors need to be adapted to humans' expectations concerning the personality that is consistent with a particular task or role.This study highlights another challenge that trait-based approaches face in HRI and SR: despite expectations, the indings of trait-based studies are frequently inconsistent and cannot be generalized [64].These limitations are important to take note of because they make clear that trait theories ofer only a partial account of personality, and they question the paradigm on which trait-based studies in HRI may inadvertently be based.At the same time, in personality psychology, it is recognized that trait-based theories and models capture only one level of human personality [53].Complementary to traits, other intellectual traditions, and theories also consider more łcontextually nuanced and psychosocially constructed features of personality [...and speak] directly to how human beings respond to situated social tasks and make meaning out of their lives in culture.ž[p.205].In social robotics and HRI, calls for more process-oriented and situated studies of robot sociality have also been made [36, 70, 75, e.g.] emphasizing the importance of expanding robot personality studies beyond trait-based approaches.Before we proceed to outline our proposal for accomplishing this expansion, Sect.2.4 introduces an example of how Dan McAdams and Jennifer Pals [53] research on human personality tried to move beyond traits to a broader, deeper understanding of personality.

Fundamental Principles for an Integrative Science of Personality
McAdams and Pals argue that personality psychology has sufered in the absence of a comprehensive framework for human personality research [53].They point out that approaches to personality, including trait-based, are commonly presented as competing or mutually exclusive.In their integrative framework, these approaches to personality are complementary ś they address diferent personality dimensions that are all necessary for the study of the whole person.The motivation to integrate personality research goes back to the person-situation debate in psychology that started in the 1970s.The debate broke out over whether or not traits reliably predict behavior.The problem was that, while data indicated that traits predicted behavior well over long periods, other data showed that it was very diicult to predict a person's behavior in a particular situation, based on measured personality traits [22].The person-side argued for traits and the situation-side insisted that to know how someone would act, it was more important to understand the circumstances, because people's behavior difers strongly across situations.Today, the consensus is that both sides were right; traits relect trends, like a person's typical Understanding personality must begin at the level of human biological evolution.This establishes how every person is like all other persons.Evolution provides the ultimate context for human individuality, by providing "universal design features against which individual adaptations vary" (McAdams and Pals, 2006.p.206).In psychology, such universals are basic physical needs e.g., sleep, basic psychological needs e.g., belonging, and innate socio-cognitive mechanisms e.g., interpreting facial expressions, as well as tendencies to develop certain cultural practices and beliefs e.g., religious beliefs.
Principle 2: Dispositional traits The most stable and recognizable aspects of psychological individuality are constituted by "variations on a small set of broad dispositional traits implicated in social life" (McAdams and Pals, 2006.p.207).Dispositional traits are understood as decontextualized, generally linear, and bipolar dimensions of human individuality that are commonly known under labels such as extroversion, dominance, friendliness, agreeableness, etc.These are studied within traits theories, which consider them relatively stable over time and across situations.Refers to life stories and personal narratives that people construct to make sense of their life.These stories help to share situated action, establish identities, and integrate individuals into social life.

Principle 5: Culture
Addresses the role that culture exerts on diferent personality constructs and how they get manifested.While the efects of culture on phenotypic expression of traits are (relatively) modest, they are considerably stronger for the content and timing of characteristic adaptations.In particular, culture has a profound inluence on life stories, where it essentially provides "a menu of themes, images, and plots for the psychosocial construction of narrative identity" (McAdam and Pals, 2006.p.211).
Table 1.The Five Principles of the New Big Five ways of acting over time; at the same time, there is strong evidence that people's behavior is highly variable and that a process-oriented approach helps understand why [23].
In contrast to these parallel dimensions of personality, later eforts in psychology responded to the debate by investigating person-situation interactions [25, e.g.].We set such conceptual diferences aside because the approaches share the same essential idea that personality is made up of diferent parts or inluences which come together into the overall whole.The integrative approach of McAdams and Pals is so promising exactly because it eliminates the competition between person and situation.Speciically, they argue that traits and situation-speciic behavior simply relect diferent parts of what makes up the complete human personality [53].To be sure, there have also been other attempts to integrate conlicting results from across personality psychology, for example, that of Epstein [17,18], and the ive principles outlined by McAdams and Pals are not the only ones they might have chosen to discuss.However, Epstein does not criticize the principles themselves, as much as what elements of each level the authors choose to emphasize [18].This simply means that even McAdams and Pals' integration is not necessarily the only possible exhaustive summary of all aspects of human psychology.
In their encompassing framework, McAdams and Pals deine human personality as ł(a) an individual's unique variation on the general evolutionary design for human nature, expressed as a developing pattern of (b) dispositional traits, (c) characteristic adaptations, and (d) self-deining life narratives, complexly and diferently situated in (e) culture and social contextž [53, p. 204].Each of the ive items of this deinition represents the foundation of one of the ive principles for an integrative science of personality.The ive principles include the three levels of personality: (i) dispositional traits, (ii) characteristic adaptations, and (iii) life narratives.Each of these levels makes up one basis for human personality, whereas principles a) and e) shape, but are not part of personality themselves.The function of traits [level i)] is to sketch a general behavioral outline; the function of characteristic adaptations [level ii)] is to łill in the details of human individualityž [53, p. 212]; the function of life narratives [level iii)] is to łtell what a person's life means in time and culturež (ibid.).Table 1 provides a summary of this framework that inspired our approach.

TOWARDS AN INTEGRATIVE FRAMEWORK OF ROBOT PERSONALITY RESEARCH
Because of its multiple levels, and how it integrates stable and dynamic components of personality, McAdams and Pals' framework [53] appeared especially useful for our aim of systematizing the existing robot personality research and for suggesting directions for future research beyond trait-based approaches. 4We developed the deinitions and details of the four levels of the Integrative Framework (IF) of Robot Personality with the particularities of robot design in mind and about existing research in HRI and SR.The levels we propose should be considered on their terms, not as an attempt at direct translation from McAdams and Pals.Based on a crucial insight into diferences between human and robot personality, McAdams and Pals' Principle 1, has become Level 1 of the IF; and Principle 5, the diferentiating role of culture, has an analog in Sect.3.5, but its role is diferent in IF than in [53], as a result of the same insight that led us to deine the new Level 1 of our framework.This insight comes from a core assumption that underpins our work: unlike humans and other biological agents, robots cannot be considered a 'natural kind'; their designs are not a product of a natural evolution nor do they follow some kind of deterministic process 5 .Since robots do not evolve, the irst level of the IF is a replacement, more than an adaptation of McAdams and Pals' Principle 1.Where the principle of evolution and human nature creates a background for human personality, our Level 1 (Sect.3.1) is on a more equal footing with the other levels.Because regardless of the level, robots and their personalities are necessarily socially and culturally constructed and situated; we cannot speak about the (material) 'essence' of robots in quite the same way, as we can for humans.With the IF, we make an efort not to assume that concepts of human personality will have a direct referent in robots.In other words, when developing our framework we remained keenly aware how what makes up robot personality and how it is interpreted remains open, situation-dependent, and culturally speciic.
In what follows, we provide deinitions of the four levels of our IF, where we identify the robot-centered constructs that comprise each level, but also point selectively towards the fundamental human cognitive processes and human personality constructs that interface with robots' designs and behaviors.An exhaustive account of the human-centered factors at play is not only outside the scope of this article, but we simply do not yet know enough about it from empirical work.Table 2 provides a summary of the four levels and the diferentiating role of culture.

Level 1: Fundamental Underpinnings of Robot Personality
We propose that the irst level of the Integrative Framework of Robot Personality Research concerns the fundamental components of robot materiality.
From an engineering and design perspective, these fundamental components create afordances for diferent personality capacities; these are the robot's features, like its hardware and its morphological design, which play a role in each of the other personality levels.In addition to the robot-centred side, users contribute another fundamental building block of robot personality, namely certain fundamental socio-cognitive processes of the human mind, like sociomorphing and the tendency to anthropomorphize [see e.g.] [84].These components are essential in shaping the experience of a robot as to some degree social 6 .
We propose that one of these fundamental components of robot materiality concerns the technological embodiment of the robot7 .This includes: the hardware that a robot runs on, the kind of sensors and actuators it has.The technological embodiment ś though not immediately related to the construction of a robot's personality as instantiated at other levels of the IF ś is nevertheless crucial because it deines and constrains a robot's capacities, the way it interacts with the world, the luidity and intuitiveness (for the user) of these interactions

Personality principle Summary
Level 1: Fundamental underpinnings of robot personality Concerns the fundamental components of robot materiality (e.g., hardware, morphological design cognitive architecture) that, from engineering and design perspective, create afordances for diferent personality capacities.In addition to robot-centered fundamentals, level 1 also encompasses the fundamental socio-cognitive processes of the human mind, such as sociomorphing and the tendency to anthropomorphise.

Level 2: Traits
Concerns the variations in a small set of robot social and communicative behavior (e.g., dialogue speed, pitch, proximity to people etc.) and design features commonly captured under the trait theories in social robotics and HRI studies of robot personality.If the components of Level 1 lay ground for what a robot will be able to do, the traits as instantiated through dialogue speed, voice, gestures, proxemics, etc. refer to what the robot actually does, how, and when.

Level 3: Adaptations
Concerns dynamical changes in the robot's behaviors as a result of learning about the environment, the user and their preferences.Motivations, goals and desires rooted in a robot also contribute to adaptations by orchestrating and constraining the direction of adaptive behaviors.
Level 4: Narratives about the robot Concerns narratives about the robot.These include the narratives that people construct about their robots that converge to a unique robot identity over time.It also includes the narratives that designers, developers, and researchers generate as a kind of back story that explains the robot's existence and potential or envisioned role in the life of the user.Within the proposed framework, the constructs on this level shape how the robot is unlike any other robot.

Diferentiating role of culture
Culture afects how the constructs within each of the four levels are manifested.Cultural codes are imprinted in robot designs, deining appropriate behaviors, and expressions, and setting the context for their interpretations.Culture also inluences people's perceptions and attitudes toward robots.and, consequently, which capacity for personality, as embodied in design features, it can support [80, e.g.].For instance, a robot operating on an Arduino will difer consequentially from a robot running on high-performance processors.Similarly, a robot that can detect and respond to a user's gaze may aford a richer and more intuitive interaction than a robot that only has more limited opportunities for interaction (e.g.only a touchscreen) [97].
In addition to technological embodiment, physical and morphological embodiment is another fundamental component of robot materiality and its role in shaping robot personality, as it is commonly understood in HRI and social robotics.As demonstrated convincingly by Robert Jr et al. in the systematic review of studies evaluating aspects of robot embodiment, it is an important vehicle for communication, acceptance, and engagement [14].
The key premises about robot embodiment, as supported by empirical evidence [e.g.29,98], are thus: i) a robot having a (physical) body is signiicant for HRI, ii) robot embodiment suggests expectations both about robot functionality and sociality.Simply put: both a robot having a body and the shape this body has will elicit diferent responses in humans.Hwang et al. provide further evidence from a study that used 27 diferent shapes of robots (including visual representations and physical prototypes) to explore whether any of them aroused afective responses in humans [32].Not only did the study conclude that certain robot bodies elicited particular emotions, but also that these and the Big Five personality traits were perceived more strongly through present physical prototypes than through the images of the same robots.More recently, Dennler et al. developed an open-source database of 165 robot embodiments and assessed initial expectations that these elicit through Mechanical Turk by asking participants to describe the robots by using metaphors, to evaluate the gender expression, and to assign tasks to these robots to probe functional expectations [15].
On the other side of the coin are the socio-cognitive processes elicited by afordances of a robot's technological and morphological embodiments, especially those processes that are more cross-situationally and cross-culturally stable than those probed in studies mentioned above.Whereas assigning a task to a robot can be seen as quite a high-level, deliberative socio-cognitive process, the kinds of processes we refer to are basic, automatic mechanisms of human sociality.In HRI and SR, research into anthropomorphism and sociomorphing has produced a variety of key insights into these mechanisms.Commonly, anthropomorphism in HRI and SR refers to the tendency to attribute human characteristics to non-human agents, including robots [11,60].The core socio-cognitive mechanism behind anthropomorphism is related to how humans use existing knowledge representations to make inferences about non-human agents [60].This process is automatic, but cognitively penetrable, meaning we may intentionally revise assigning human attributes to non-human agents/objects upon deliberation on the situation [e.g.90].As a dispositional trait in humans, anthropomorphism is expressed to diferent degrees by diferent people, and in diferent situations.Building on the theoretical and empirical work in cognitive psychology and SR, Nicolas, and Wykowska have provided evidence that among the factors that contribute to a person's tendency to anthropomorphize the need for cognition (an individual's will to engage in relective processes) and need for closure (need for understanding and prediction of non-human behavior) will play an important role [60].
In addition to the challenge of diferentiating and assessing empirically the individual mechanisms that contribute to a person's tendency to anthropomorphize, other ongoing work in HRI and SR suggests that anthropomorphism may not exhaust all various cognate tendencies that people engage when interacting with non-human agents such as robots.The notion of sociomorphing complements and extends the ongoing work in HRI on anthropomorphism; sociomorphing refers to the process when humans engage a mental model of their interaction partner, in which this model can stem from the experience of human-human interactions (underlying mechanism of anthropomorphism), but it needn't necessarily [74].The motivation behind this is the recognition that a iner conceptual and methodological diferentiation is needed to account for various instances of humans attributing social capacities to (social) robots [10,11,74].Rather than being an łall-or-nothing afairž, anthropomorphism is one form that sociomorphing can take alongside other types of experienced sociality [11].The Descriptive Ontology of Asymmetric Interactions (OASIS) framework by [10,11] makes this theoretical premise available for empirical evaluation, and integrates conceptual resources to anchor forms of sociomorphing and experienced sociality.
This human/user side of the fundamental underpinnings of robot personality is crucial to keep in mind when designing robots; it is what leads us to argue that robot personality emerges at the intersection, in the interaction of robots and humans.A growing body of research in HRI and SR testiies that robotic designs constitute a łminimal form of context, modulating the efect [of the robot] at the dispositional levelž [60, p.10].For example, the closer the morphology of a robot to humans, the more likely it is to activate the human action-perception system that leads people to project human-like characteristics, including personality traits, to it [102].Similarly, a robot being able to detect human gaze plays a role in shaping intuitive interactions [97].Lorenz et al. also emphasizes movement synchrony and reciprocity in HRI as a common ground that also supports higher-level mechanisms of interaction [48].
Suggested directions of research.We agree with [60] that understanding HRI better means we also have to understand how the basic mechanisms of human cognition and psychology łinteract [with], impact or The crucial open questions for the IF concern how robot designs and human cognitive processes at this fundamental level shape perceptions of robot personality, i.e. whether they can be intentionally and speciically taken advantage of to build one or another kind of robot personality.
Considering the types of experienced sociality Damholdt et al. found manifest in situated interactions with robots of varying designs [11], we propose extending research beyond the robots designed intentionally to support and participate in social and afective interactions with humans (the so-called social robot) to include, for example, functional service robots that people also experience as in some capacity social [16, e.g.].
New directions for research emerge as people have more time to observe and experience the behavioral capabilities of robotic systems.From existing long-term studies of social robots in everyday life settings, we know that initial fascination with (social) robots decreases over time, as the novelty efect fades [37,71,91].As pointed out by Robert Jr et al., the length of interactions is an integral factor for how a robot is perceived [14].Outside of laboratory settings characterized by a limited scope and number of interactions, tracking the role of the fundamental components constitutive of robot materiality and the experience of a robot as in some form social agent remains a challenge that is nevertheless worth pursuing.
Another future direction could investigate the opportunities and issues of designing robots that are simple in the computational resources required (i.e.technological embodiment) while ofering their human counterparts a particularly engaging platform for robot personality to emerge over long-term interactions.For example, the work on the handcrafted open-source robotic platform Blossom by Suguitan and Hofman [81] is a promising step in this direction.According to its creators, design elements of Blossom (e.g., quick assembly mechanism, handcrafted appearance open to customization, tensile mechanisms, and elastic components ensuring organic movements) are conceived with the aim of low barrier-of-entry and resulting in an accessible and customizable robot (ibid.).

Level 2: Traits
The second level of the Integrative Framework concerns variations in a small set of relatively static robot social and communicative behaviors (i.e.dialogue speed, pitch, proximity to people, etc.) and design features commonly captured under the traits theories in SR and HRI studies of robot personality.
As we discussed above, the core assumption behind robot personality studies informed by trait models is that personality can be mapped onto an externally recognizable set of behaviors and communication cues.For example, Ludewig et al. distinguished two robot personalities ś extraverted and conventional ś based only on certain verbal and non-verbal characteristics and investigated whether the extroverted robot was associated with higher social acceptability [49].They hypothesized that the extroverted robot personality would be associated with the higher social acceptability of the robot.Based on a ield study with 194 participants, the authors concluded that the extroverted shopping robot received higher acceptance scores and was perceived to be more extroverted than the standard version.
Thus, the components of robot personality at this level ofer more straightforward opportunities in terms of how to design robot personalities.In other words, if technological and morphological embodiments (level 1) deine the capacities of what a robot will be able to do, the traits as instantiated through (more supericial compared to level 18 ) robots appearance features, dialogue speed, gestures, proxemics, dialogue strategies, etc. refer to what the robot does and how it does it.Today, designers working on a particular robot might have certain goal traits in mind, which teams of animators and dialogue authors work to achieve.
The challenge on this level is reliably mapping traits to certain behaviors, like speech styles.From a methodological and design perspective, this also raises the topic of user involvement in decision-making about robot design and social and communicative behaviors.Especially relevant are decisions that contribute to establishing a robot 'persona' or character that diferentiates it from other similar robots.An interesting example comes from the work of Cietto et al., who used a common hobby robotic kit and relied on participatory design methods to co-design, together with children aged 7-8 years old, an educational robot's appearance and personality [9].The open-source robotic platform Blossom already mentioned above also explores how to leverage user involvement in crafting the robot's appearance and behaviors for more sustained interactions [81].
Suggested directions of research.Within the IF, we propose SR and HRI researchers continue exploring trait models for robot and human personality, but with a renewed, acute sensitivity to the challenges embedded in this approach.For instance, the characteristics of what constitutes an 'extravert' robot in existing studies are limited.We argue that these characteristics are only one aspect of what made participants rate the 'extraverted' robot in this study as more likable and joyful to use.These limitations are exacerbated by assuming that human personality traits are always adequate for describing robot personalities.This assumption alone may lead to inconsistent results, if research participants must invent a mapping between their experience of robot personalities and pre-selected categories (e.g., the Big Five).In response to this concern, there are exceptions to the a priori use of human personality traits [5, 47, 103, e.g.] that emphasize diferences between robots and humans (and computers) and call for studying each in their own right.
One way to advance these eforts could involve deepening a budding participatory approach to researching robot personality traits.This approach would avoid naturalizing robot personality based on one particular understanding of human 'nature' in favor of developing a way to think about robot personality that begins from the robots.For example, Weiss et al. studied the adjectives that participants used to talk about companion robots qualitatively [93].That way, researchers remained open to understanding robot personality in terms of humans' experience of robots, rather than through instruments designed to measure human personality.This approach ofers a distinctive advantage in that it avoids reinscribing the limitations of personality psychology's corresponding approaches and instruments onto robot personality research.Future research could delve into how the perceived human-likeness of a robot impacts whether people ind human-speciic traits or robot-speciic traits as more appropriate in a speciic interaction context.It is important to note that our proposal does not imply that we assume that one deinitive set of robot-speciic traits should, or even could, be developed.Instead, it encourages an exploration of the possible robot-centered traits in principle, along with an examination of how robot-centered and context-centered factors determine their relevance.This approach ofers a promising avenue for addressing the limitation of solely relying on human-based traits in the interaction design.
In a similar vein, another salient question concerns the tension between evidence for the long-term stability of personality traits and the fact that this does not entail that personality traits remain unchanged across life span [53], [6,66].It remains unclear how to address the stability or change in the context of sustained interactions with socially interactive robots.Future research from a human-centered perspective might pursue how the perception of robot personality traits changes over time.At the same time, it is worthwhile to continue exploring architectures suitable to enable dynamic robot personalities.

Level 3: Adaptations
In the third level of the Integrative Framework, we propose to understand adaptations as dynamic changes in a robot's behaviors as a result of learning about the environment and the user and their preferences and responding by continuous adaptation of verbal and non-verbal behaviors and task performance.Motivations, goals, and desires rooted in a robot also contribute to adaptations by orchestrating the direction of adaptive behaviors.The aim for the development and evaluation of personality constructs at Level 3 is to support the building and maintaining of meaningful, personalized, and lasting human-robot interactions.This relects Dautenhahn's call for individualized robot companions that need to be 'socialized and personalized' to meet the emotional, social, and cognitive needs of their owners [12].Speciically, Dautenhahn draws from a developmental perspective and the model of dog-human relationships to ground her proposal for the 'bringing up' of robots.
Technically, many challenges remain to achieve such sophisticated learning and personalization, in which what needs to be learned, where to look for instances of target behaviors to learn about and how to recognize them, as well as the ideal response behavior may all be unknown.The literature in HRI is rich with diferent proposals for how to implement this kind of learning, ranging from direct mappings of human personality or task context to robot personality to machine learning approaches.Diferent models are being explored including a variety of unsupervised and reinforcement learning approaches.Diferent research groups develop cognitive architectures, often to address diferent dimensions of personalized adaptation.Ideally, users themselves are also able to reinforce to personalize robot behaviors, apparent motivations, and the personality they express.In one example, Uchida et al. integrate implicit and explicit user feedback into a single learning model [89].To keep abreast of these many developments, Kiderle et al. provide an overview of how diferent reinforcement learning approaches can be engaged for the task of supporting dynamic adaptations [38].They also discuss how neural networks can be used to realize expressive behaviors during interactions by using a data-driven method.
We mentioned above goals, motivations, and desires programmed into a robot as one component that also contributes to robot personality development at Level 3 of the IF.Goals, motivations, and desires constrain and shape the 'essence' of a robot's personality and ensure a kind of trajectory for the behavioral changes of the robot.From the user perspective, this may be perceived as the robot behaving (more) consistently than a robot without such a motivational core, and ideally in a manner that relects the user's needs and preferences.Thus, a character emerges for the robot that diferentiates it from other similar robots of the 'same kind'.The core motivations that shape a robot's personality are some of the structures that orient deep learning and other models of user desire toward making the robot respond to users in a meaningful way.Because, while model choices and optimizations present their own challenges, the basic challenge, as pointed out above, remains i) capturing relevant user behaviors and then ii) having the robot respond in a relevant and consistent manner.Rather than constraining the computational form learning about users should take, what motivations/intentions/goals ofer above all is a representation for understanding and deining the robots' basic attitude toward the user in their relationship.Concrete work on intentions and motivations in robots is being pursued by Hiroshi Ishiguro and colleagues; in Uchida et al., for example, the authors discuss the development of an autonomous dialogue robot intended to support ła symbiotic relationship with humans, where both have their own intentions and desires and infer each other's ones through dialoguež [89, p.2]. Inspired by the indings from neuroscience and cognitive science that recognize that (some) human and animal desires are instinctual and need not be explicit, they propose a cognitive architecture where desires are embedded both on the conceptual level (representations) and in the android relexive behavior.The motivation behind this dual structure is to contribute to łrational selection of behaviorž, much like in humans, and to enable the realization of the complex functions of the robot that would allow humans and robots to communicate more successfully and to learn about each other.
Other work has made eforts to design a robot's root behaviors following one or more behavioral styles of personas [51, 68, e.g.].In a study on inferring intentions from eye-gaze cues with the human-like robot Geminoid, Mutlu et al. conclude that users notice and are helped by coherent behavioral cues that suggest an underlying goal or motivation of the robot [59].These are both characteristic adaptations in the sense intended by McAdams and Pals.In another study, Tanevska et al. explored how the cognitive architecture of the humanoid robot iCub can support diferent user proiles and contribute to the low of interactions by inscribing diferent values for the robot's internal variables at the beginning [83].To test the adaptations, three user proiles were identiied based on the frequency and modality of interactions ś a highly interactive, sparsely interactive, and in-between proile.One important outcome was that even when the robot adaptation was slow-paced, it was possible to observe changes in robot behavior over time.Thus, the robot could demonstrate a capability to progressively adapt in interactions with its users.
Suggested directions of research.While there have been attempts to address how adaptability can be integrated at the level of agent architecture [83], we know little about how user-adaptive systems will play out in longterm interactions.For instance, in the case of Jibo, a personal assistant robot that learns the user's habits and preferences regarding the robot's actions [52], the robot's (limited) adaptability did not appear to support longterm engagement [31].This was also the case with Anki Vector [85], Karotz [13], and Pleo [21].One notable example is the robotic seal, Paro.Paro employs a reinforcement learning algorithm to gradually adapt to the user's preferences.However, it is not clear to what degree the relative success of Paro in its domain of application is supported by the system adaptiveness, in contrast to other features such as, for example, the haptic feedback that the Paro robot provides by the virtue of having a particular kind of material embodiment, as we discussed in 3.1.
To address the absence of adaptiveness in long-term interactions, we speculate that directing all eforts toward the pursuit of technical innovations regarding automation may in isolation not be the solution to the fundamental challenge of long-term engagement [31].In contrast, with our Integrative Framework for Robot Personality Research we propose that the automated adaptation of robots should be complemented by a user-involved design that builds on humans' needs and narratives.A complementary approach to this is to investigate how people may personalize their robots throughout interactions.This approach stems from the assumption that no design process is ever inal.Due to the dynamic and complex nature of HRI, it is impossible to predict everything about the human beings that the robot is designed for.Thus, a promising direction of research is to explore designs that enable people to implement changes and adapt their robots according to their wishes and changing needs [94, e.g.].

Level 4: Narratives about the Robot
Level four of the Integrative Framework incorporates narratives about the robot.These include the narratives that people construct about their robots and that converge to a unique robot identity over time, as well as the narratives that designers, developers, and researchers generate as a kind of 'back story' that explains the robot's existence, role and that may shape the relationship with the users.Within our Integrative Framework for Robot Personality Research, it is these narratives that shape how the robot is 'unlike any other robot'.
At the level of narratives, we distinguish those that emerge at design time and during interaction.Designed narratives may relect something like a back story for a particular robot, its role, character, and relationships with the humans around it.Such narratives can be delivered by the robot itself, or by another human, like a coworker or an experimenter.One poignant example of researchers engaging a narrative to shape HRI is the work by Jacq et al. [33].The study centered around the CoWriter activity where the aim was to enable a young participant to teach handwriting to a robot in a complex and rich interaction.Two Nao robots were used, and a narrative script was created to convince the child the robot truly needs help and beneits from the lessons.In this narrative, one of the Nao robots, called Mimi, was away on a scientiic mission, and the Nao robot called Clem was communicating with Mimi with handwritten messages just łlike humansž.Coupled with algorithmic adaptation to relect the challenges each child faced with handwriting, the learning activities designed in this way proved promising in promoting children's motivation, commitment, and overcoming low conidence.
While designed narratives can be useful to guide design choices and help steer interactions with users, they relect a weaker understanding of this level of personality, than user-driven narratives that arise during interaction.In contrast to the designed narratives, narratives that people construe about robots in situated interactions represent a stronger understanding of narratives.These narratives relect a dimension of human personality regarding how individuals make meaning of their lives and how they relate to the world [53].Thinking in terms of this stronger notion of narratives suggests a shift in emphasis from the designing of things to the designing of meaningful human-technology relations.Concerning robot personality research, we argue this calls for an extension toward the notion of 'identity' as something that is co-constructed and enacted in diferent conigurations of human, technological, and contextual factors.
Methodologically, studies of life narratives are naturalistic and can be challenging as they are incompatible with hypothesis testing [14].Rather than looking for universals or for dependent variables that reliably predict outcomes of human-robot interactions, such as the case with the studies that address personality constructs on levels 1, 2, and 3, studies that address the narratives that people construe about robots are exploratory and open-ended.For example, Syrdal et al. conducted an insightful study that attempted to provide a narrative frame for long-term (10 weeks) human-robot interaction [82].In this study, narrative framing techniques provided a narrative within which participants could interpret their experience of interacting with the robot.As part of providing an ecologically valid setting, this method also explicitly relected aspects of the culture within which the interaction takes place, including easily overlooked aspects such as the conventional layout of a domestic kitchen.In this situation, the framing took advantage of the human tendency to create narratives and leveraged this as the basis for more robust human-robot interaction.
An additional point to consider is that narratives are not only constructed by humans.Social robots are best understood as co-creators of narratives about interactions they participate in [87].In that regard, one practical concern is that an important element of designing engaging social robots may be what narratives robots themselves appear to form about the interaction.In the cases of Cog and Kismet, for example, Turkle et al. identify several narratives that the robots appear to participate in constructing [86].This includes 'the discourse on aliveness; Cog seems 'wounded' not broken, and Kismet can seem suddenly 'deaf'.From our perspective, one central aspect of such interactions is how the robots encourage such interactions by sharing in the narrative.
Suggested directions of research.In psychology, the topic of narrative identity has gained considerable attention [53].The HRI community is yet to decide which approaches to studying narrative identity are relevant and appropriate regarding gaining insights into the meaning-making process that enables interactions with and relations to social robots.In robot personality research, one research direction is to address the content themes in narratives that people construct about their robots and examine how these relate to the components of robot personality, as discussed under Levels 1-3.The above-mentioned study of the CoWriter activity with two Nao robots by Jacq et al. [33] is an excellent example of how the narrative approach could potentially also be integrated in lab studies e.g., to explore how diferent narrative scripts impact the perception of a robot's personality.At the same time, more longitudinal studies in naturalistic environments (e.g., people's homes or public spaces co-shared with robots) will be invaluable for deepening our understanding of how a unique robot identity emerges in situated interactions over time and which role it plays in the overall acceptance of robotic technologies.
The study of robot personality through the prism of narratives may be developed further by considering whether and how robots can facilitate or create their own narratives and life stories.This is diferent from the case above, however, in the sense that Cog and Kismet were designed to perpetuate the narratives that Turkle et al. discuss [86].Instead, future social robots might be designed with a view towards robots narrating their interactions and personalities more openly.One intriguing attempt in this direction comes from Winield, who outlines a proposal for an embodied computational model of storytelling for robots [99].Per Winield, if this model were built, it would open the possibility for investigation of how narratives can emerge from a robot's interactions with the world and then be shared, as stories, with others.

The Diferentiating Role of Culture
Citing Shweder and Sullivan [78], McAdams and Pals deine culture as łthe rich mix of meanings, practices, and discourses about human life that prevail in a given group or society" [53, p.211].Whereas HRI typically operationalizes culture as national culture [77, e.g.], the basis our proposal has in STS expands the IF to other notions of culture as well.For example, culture can also be understood more locally, as in the case of epistemic cultures, ways of knowing, that shape a particular community's practices [8, e.g.].The general, constructionist starting point for the IF is simple: all technologies are socially embedded, and scientiic research and engineering never exist outside of conventional meanings, practices, and discourses, i.e. culture [46,69].
In McAdams and Pals, culture is discussed in terms of how it afects the other levels of personality, each of which it afects diferently; for example, culture i) shapes how and to what extent people in diferent cultures express their dispositions (level 2), or ii) provides a menu from which people chose the narratives in terms of which they frame their own life stories (level 4) [53].Culture also exerts a particular inluence on each level of robot personality, as, in many respects, the HRI and SR communities already recognize [19, 46, e.g.]; culture inluences peoples' perceptions and attitudes toward robots [3,30]; and cultural norms are imprinted in robot designs, deining appropriate behaviors, expressions, and the context for their interpretations [43,44].The diference to McAdams and Pals is that for robots, no matter which level of personality we are referring to, it is a question of how the cultural background of a human shapes what robot personality is being constructed, at design-time and during interactions.
Concerning the stable features of robot personality (level 1), Lee et al. point out diferences across cultural backgrounds in how people interpret robots and what they expect from the look and feel of the robot [44].This captures elements of the overall design concept and embodiment of the robot, including its shape, gender, materials, and size.Lee and Sabanovic concluded that culturally-variable perceptions of robots are fundamentally related to particular norms and social dynamics, rather than being reducible to more direct factors like media exposure or religious beliefs [43].For designing robot personalities, this means that design choices, and shape, form, and character of the robot should be expected to play a diferent role for people with diferent cultural backgrounds.
Prior work also shows that the interpretation of robots' personality traits (e.g., how extroverted/introverted they are perceived to be ś level 2) is shaped by culture.For example, Weiss et al. investigated task-dependence and cultural background dependence of the socially interactive robot personality trait attribution and found that cultural background mediates how traits are attributed [95].In a living room scenario with 28 people and a robot, Woods et al. found speciic efects on the perception of robot personality based on participants' gender, age, and technological background [100].Since gender, at least, is a strongly culturally-mediated criterion, it seems useful to keep in mind ways that culture shapes perceptions of robot personality through the way it shapes users' beliefs about themselves and their relationship to the robot.
The word adaptation (level 3) ofers a curious overlap of two meanings: i) changing one's behavior (adapting) to accommodate a particular situation, and ii) McAdams and Pals' sense of adaptation, which is a question of values.When we study robot personality, these meanings are often entangled, like in one in-the-ield study, which showed that a robot was able to improve its performance on a collaborative task when it changed its behavior [i)] in response to the information it obtained about humans' cultural background, in which their values [ii)] are implicit [73].This suggests it may be important for robots to adhere to cultural norms in certain situations by dynamically adapting to social rules, to a certain extent.Doing this raises a variety of challenges, including technical ones about how to learn cultural norms efectively [72,92].Rather than designing robots for particular cultural settings, Li et al. investigated how robots could be adapted to better suit diferent sets of norms and expectations [45].In this study, the authors outline relative priorities among norms for various cultures, providing the basis for culturally sensitive adaptations to robot behavior.One example the authors give is the importance, in diferent cultures (nationalities), of the robot complying with social conventions.This compliance appears to be more important, on average, for Chinese participants than for American participants, independent of the relative strength of social norms.In another study to explore culturally sensitive adaptations, Evers et al. found diferent efects of the strength of in-group feeling for Chinese and American participants on average [19].In this study, Chinese participants, as compared with the US subjects, were more comfortable when an assistant robot was characterized as a strong in-group member.In future work, robots may be designed to be lexible in these respects, leaving it up to successive interactions with users to facilitate the robot's adaptation to its environment.
Concerning narratives (level 4), one set of studies provides insights that stand in contrast to the limited success of irst-generation social robots.A trend has emerged in Japan to hold funeral services for AIBO robot dogs, who are treated as aging relatives, and robot repair shops have come to think of themselves as łclinicsž [40].One feature of users' narratives at work here is the animism inherent in Japanese Buddhist culture, which may lead to a radically diferent set of narratives about robot identity.However, [43] suggests this is too supericial a perspective.[69] develops a deeper perspective that brings out how cultural models of social behavior, and cultural models of cognition and technology, together contribute to the developing narratives of Japan's robot culture.In this context especially, reducing culture to nationality is too supericial a level of analysis.Lee and Sabanovic, for example, point to diferences between 'tight' versus'loose' cultures instead [43], about the relative strength of social norms and forms of sanctioning deviant behavior (tight culture/high strength+sanctions; loose culture/low strength+sanctions) [26].
A inal aspect to consider on the subject of culture is a critical stance on the values we design into robots.The particular challenge this raises for social robot design is how to avoid reinscribing prejudice or reinforcing discrimination by, for example, carelessly or harmfully gendering in robot design.One issue concerning gendered physical embodiment and personality is the association of humanoid robots with a stereotypically female form with traits understood as being stereotypically feminine [101].Contrary to current practice though, humanoid social robots have the potential to dismantle gender norms by participating in the construction of a diferent narrative [67, e.g.].The relations between robot designs, personality, and ideas of how these should 'naturally' be mapped onto one another, which was discussed at the beginning of Sect.3, ofers one site for intervention.Currently, such interventions are mostly being undertaken in the realm of (performance) art, but there has been early work here that ofers a starting point for ambitious, critical, and emancipatory robots, like Sontopski creating a stereotypical-feminine-typed voice assistant that resists abusive behavior [79].

SUMMARY
In this article, we presented our theoretical proposal for how to comprehensively extend robot personality research beyond trait-based approaches for HRI and SR research.Our proposal builds on the core assumption that robot personality is not limited to a set of static design features and behavioral cues that are commonly captured under trait models, but also includes fundamental components of robot materiality and more dynamic and emergent properties at the intersection of design, interaction, user, context.Each of these dynamics plays out diferently across design time and interaction time, together creating the complex basis for robot personality that we have argued for in our work.
Inspired by McAdams and Pals' Integrative Framework of Human Personality [53], we sketched out a 4-level Integrative Framework for Robot Personality research.Table 2 provides a summary of the four levels and the diferentiating role of culture.The irst level includes the basic components of robot materiality (i.e.technological and morphological embodiment) and how these interface with the fundamental socio-cognitive properties of the human mind such as the tendency to socio-and anthropomorphize.We discussed how the components at this level establish the preconditions for a robot to be experienced as in some capacity social and create afordances and constraints for the development of robot personality construct at Levels 2-4.The second level of the IF concerns variations in a set of (more or less) static social and communicative behaviors and design features that contribute to the establishing of the personality constructs commonly captured under the trait theories in SR and HRI e.g.extraversion or introversion.At this level, we also considered user involvement in establishing a robot 'persona' that diferentiates it from other similar robots by taking advantage of techniques like letting users modify the robot's appearance.The third level of the IF captures behavioral adaptations understood as more dynamic changes as a result of learning about the environment and the user over time and adapting verbal and non-verbal behaviors and task performance in response.We also discussed motivations, goals, and desires rooted in a robot as an important component at this level orchestrating the direction of adaptive behaviors.We argued that the proposed constructs at this level address what Kerstin Dautenhahn coined as 'bringing up' robots, where the aim is to support maintaining meaningful, personalized, and lasting human-robot interactions.Level four of the IF incorporates narratives about the robot that converge to a unique robot identity.We discussed how, in a weaker sense, these narratives can include the narratives that designers and developers propose to generate a 'back story' that explains the robot's existence and sets the context for the interaction; and, in a stronger sense, steers toward narratives that users generate about the robots in situated interactions over time.We also illustrated the diferentiating role of culture in providing a host of themes, values, imaginaries, and norms, all of which inluence, in complex ways, how robot personality inds its manifestation in how constructs of robot personality are manifested at all levels of the IF.
We want to reiterate that the integrative approach put forth by McAdams and Pals is not the sole attempt in psychology to bring together diferent strands of personality research.As with many complex and multidimensional constructs, such as sociality or culture, there is often no single deinitive framework that avoids all controversies.McAdams and Pals's approach is no exception.Consequently, alternative approaches aiming to reconcile diverse personality models do exist and continue to evolve as we pointed out in 2.4.We chose to adopt McAdams and Pals' framework as it provides a sensible foundation for showcasing our eforts in delineating the breadth and depth of robot personality research within the ield of HRI.One compelling aspect of their approach is their ability to resolve the apparent conlict between individual traits and situation-speciic behaviors.In essence, it recognizes that traits and more situation-speciic behaviors are not mutually exclusive but rather represent diferent factors of what makes up the complete human personality.We also appreciate the framework integrating stable and dynamic constructs of personality.We acknowledge that relying solely on McAdams and Pals' proposal may not be the deinitive or exclusive approach, but it serves as a suitable means to achieve our speciic objectives.Furthermore, it is important to note that our proposed framework ś though grounded in ongoing empirical work in SR and HRI ś is essentially an analytical abstraction.Other scholars may identify fewer or more components at each level, or even dispute the deinition of the levels altogether.We invite further contributions to the framework development.We also recognize that practical concerns and limitations will inevitably constrain how many angles of the IF individual studies can take into account.In that regard, our framework is not intended to be used as a one-to-one blueprint for implementation in a robot ś even if a consensus is reached within the HRI and social robotics community regarding the constructs that underpin each of the levels, there will still be considerable variability in how these constructs can be instantiated in a given robot's design depending on the intended task, human factors (e.g., the tendency to anthropomorphize and sociomorph), environmental and technical constraints, and particular cultural context wherein the robot is designed and deployed.This is why we see one of the ultimate values of the proposed framework in its potential to frame seemingly distinct eforts and methods as complementary.In other words, we hope to have provided a frame of reference within which to situate, compare, and integrate robot personality research across HRI toward a cohesive, broader, and more robust response to current challenges.
of human individuality concerning a wide range of motivational, social-cognitive adaptations contextualized in time, place and social role.In comparison to dispositional traits, characteristic adaptations are more amenable to environmental and cultural inluences, and invoke dimensions of human individuality that are more closely connected to motivation and cognition.Principle 4: Life narratives [are] impacted by robotsž [p.11].We see a fruitful direction of research in further development of conceptual and methodological tools for studying i) what it is about robots that taps into basic human socio-cognitive processes, ii) which human socio-cognitive processes play what kind of role, in diferent situations or at diferent ages, for example, [58, e.g.].

Table 2 .
The four levels of the Integrated Robot Personality Framework.The levels are diferent in terms of the (relative) stability of the personality construct they encompass and in terms of how open a given dimension of personality is to being mediated by human-centered parts of the situation.Personality constructs on each level are afected by culture in diferent ways.