ArticlePDF Available

Voice-Based Agents as Personified Things: Assimilation and Accommodation as Equilibration of Doubt

Authors:

Abstract and Figures

We aim to investigate the nature of doubt regarding voice-based agents by referring to Piaget’s ontological object–subject classification “thing” and “person,” its associated equilibration processes, and influential factors of the situation, the user, and the agent. In two online surveys, we asked 853 and 435 participants, ranging from 17 to 65 years of age, to assess Alexa and the Google Assistant. We discovered that only some people viewed voice-based agents as mere things, whereas the majority classified them into personified things. However, their classification is fragile and depends basically on the imputation of subject-like attributes of agency and mind to the voice-based agents, increased by a dyadic using situation, previous regular interactions, a younger age, and an introverted personality of the user. We discuss these results in a broader context.
Content may be subject to copyright.
Voice-Based Agents as Personied Things:
Assimilation and Accommodation as Equilibration
of Doubt
Katrin Etzrodt1 and Sven Engesser1
1 Science and Technology Communication, Technical University Dresden, Dresden, Germany
Abstract
We aim to investigate the nature of doubt regarding voice-based agents by referring to
Piaget’s ontological object–subject classication “thing and “person,” its associated equil-
ibration processes, and inuential factors of the situation, the user, and the agent. In two
online surveys, we asked 853 and 435 participants, ranging from 17 to 65 years of age,
to assess Alexa and the Google Assistant. We discovered that only some people viewed
voice-based agents as mere things, whereas the majority classied them into personied
things. However, their classication is fragile and depends basically on the imputation of
subject-like attributes of agency and mind to the voice-based agents, increased by a dyadic
using situation, previous regular interactions, a younger age, and an introverted personality
of the user. We discuss these results in a broader context.
Keywords: articial agents, classication, ontology, social actor, hybrids
Introduction
When it comes to being social, people are built to make the conservative error:
When in doubt, treat it as human. (Reeves & Nass, 1996, p. 22)
In 2011, Apple launched Siri, the rst commercialized voice-based agent (VBA). More
VBAs have followed rapidly and have entered the habitats of people (Newman, 2018), the
Human-Machine Communication
Volume 2, 2021
https://doi.org/10.30658/hmc.2.3
CONTAC T Katrin Etzrodt • Science and Technology Communication • Technical University Dresden • katrin.etzrodt@tu-dresden.de
ISSN 2638-602X (print)/ISSN 2638-6038 (online)
www.hmcjournal.com
57
Copyright 2021 Authors. Published under a Creative Commons Attribution 4.0 International (CC BY-NC-ND 4.0) license.
58 Human-Machine Communication
most popular being the Google Assistant (launched in 2012), Microso’s Cortana (2013),
and Amazon’s Alexa (2014). However, neither users nor scientists have yet been able to
completely grasp the nature of these VBAs (Guzman, 2019). us, clarifying the funda-
mental question of ‘what is it?’ has become one of the main research issues within human-
machine communication (A. Edwards et al., 2020). In this paper, we will explore the nature
of these VBAs by investigating how they are categorized as objects, subjects, or in-between.
We aim at making the classication of commercial VBAs and the associated processes of
equilibration” (Piaget, 1970/1974) accessible for empirical research.
Both CASA and Social Presence eory argue that VBAs are what Reeves & Nass (1996,
p. 22) describe as objects of doubt. Research under the CASA paradigm (Nass et al., 1994),
postulating that computers are social actors, has provided multifaceted evidence that peo-
ple exhibit diverse social reactions toward various technological artifacts such as traditional
media, computers, avatars, or robots. Social Presence eory (Short et al., 1976) assumes
that these reactions are caused by the perception of mediated social entities as being pres-
ent (Lombard et al., 2017), or by the failure to recognize those entities as articial (K. M.
Lee, 2004), and specically as “non-human” (Latour, 2005). Although even marginal social
cues can trigger social reactions and the feeling of presence (e.g., Reeves & Nass, 1996),
anthropological similarity can further enhance them (e.g., Appel et al., 2012; Horstmann
et al., 2018). In particular, voices are notably powerful indicators of social presence (Reeves
& Nass, 1996) and being able to talk can be an indicator for being alive (Turkle, 1984/2005,
p. 48).
Although a large body of research in the past 30 years has investigated the conse-
quences of this doubt (e.g., the activation of social heuristics), we know very little about the
doubt itself. It is still unclear toward “whom” or “what” (Gunkel, 2020, p. 54) people react
when they are interacting with (articially intelligent) machines. e various terminolo-
gies used to describe them, such as “evocative objects” (Turkle, 1984/2005), “quasi-objects”
or “hybrids” (Latour, 1991/1995), “epistemic things” (Knorr-Cetina, 2001), “non-humans
(Latour, 2005), “subject objects” (Suchman, 2011), or “social things” (Guzman, 2015) illus-
trate the diculty of capturing their essence. Although denitions dier considerably, com-
mon to all terms is the reference to ambiguous entities possessing both object qualities and
subject qualities, whose unique combination creates new qualities beyond the sum of its
parts (Roßler, 2008). In the paper at hand, we aim at investigating the categorization of these
doubtful objects.
For this purpose, we will argue that the doubt arises from an irritated ontological
object–subject classication by referring to the epistemologist Piaget (1970/1974). We will
outline how people resolve the irritation through specic equilibration processes by assign-
ing the irritating object to an existing scheme or modifying these schemes gradually.
e second part of the paper is dedicated to inuences on this classication. Research on
articial agents indicates that attributes of the situation (A. Edwards et al., 2019; Purington
et al., 2017), of the users (Epley et al., 2007; E. J. Lee et al., 2000; Woods et al., 2007), and of
the agents themselves (e.g., Nass & Moon, 2000; Reeves & Nass, 1996) may be relevant.
Based on an empirical approach, we designed a scale for the classication, assessed
situational, individual, and technological inuences in an online survey, and validated the
ndings with a second sample. By investigating the degree to which people classify a VBA
as social actor, ranging between thing and person, and the identication of inuences we
Etzrodt and Engesser 59
intend to advance the concept of human–machine communication and contribute to the
CASA paradigm.
Classication of Voice-Based Agents
Voice-based agents like Siri, Alexa, and the Google Assistant are a subtype of articial social
agents, with an operating system based on articial intelligence and natural language pro-
cessing using a disembodied voice emanating from a device (e.g., smartphone, loudspeaker
box) to communicate with the users and execute their tasks. ey are still a reasonably
new technology, meaning that people lack exhaustive previous experience with this unique
conversational interface (Guzman, 2015; Krummheuer, 2015). However, there are signs
for human-machine agent interaction scripts (Gambino et al., 2020), suggesting a gradual
object–subject classication of articial agents. us, we ask to which degree people classify
VBAs into the object–subject classication (RQ1).
Voice-Based Agents as Objects of Doubt
Referring to Piaget’s studies on epistemology (1970/1974), we assume the most fundamen-
tal way of gaining knowledge about a new object is guring out if it is part of the “psycho-
morph” or the “physicomorph” scheme, which are diametrical poles of the same ontological
classication. Turkle (1984/2005, p. 34) referred to these poles as “psychological” and “phys-
ical.” Gunkel (2020, p. 54) arrived at a similar conclusion by referring to Derrida’s distinc-
tion of “who” and “what.” e psychomorph scheme (“who”) is dened by subjects, which
are living beings, equipped with capacities like thinking or feeling, and the potential of
agency (Piaget, 1970/1974, p. 48). us, it is an analog to the scheme “other persons” (Gun-
kel, 2020, p. 54), and used “to understand people and animals” (Turkle, 1984/2005, p. 34).
In contrast, the physicomorph scheme (“what”) is dened by inorganic, non-living objects,
which are suciently comprehensible in terms of precise, logical-mathematical categories,
and deterministic causal laws (Geser, 1989, p. 233). at is, it is an analog to the scheme
“things [that] are mere instruments or tools” (Gunkel, 2020, p. 54), and “used to understand
things” (Turkle, 1984/2005, p. 34). While the physicomorph scheme (hereinaer referred
to as thing theme) results from empirical experience (e.g., physical perception and move-
ment), the psychomorph scheme (hereinaer referred to as person theme) originates in the
introspective experience of a subjectivity (Piaget, 1970/1974). is theoretical approach is
related to the impossible verication of subjectivity (Gunkel, 2020; Turing, 1950) or agency
(Krummheuer, 2015), as well as to the imputation of mental states to objects (Premack &
Woodru, 1978) and the gradual assignment of personhood (Hubbard, 2011), as discussed
prior in Etzrodt (2021).
Despite the growing ability to distinguish between psychomorph and physicomorph
due to the individual’s development and its constant confrontation with the environment
(Piaget, 1970/1974; Turkle, 1984/2005), some objects remain a challenge. Between things
and persons, there exists a wide range of objects (e.g., plants, animals, or articial entities)
that can be regarded as objects of doubt. Recent empirical research conrms the doubtful
nature of various articial social agents: Guzman (2015) noted in her qualitative interviews
a constantly shiing use of the pronouns “she” and “it” when people talked about the VBA
60 Human-Machine Communication
Siri. It seemed that Guzman’s respondents were torn between the association of a person
and a thing, which she traced back to the interviewees’ focus of attention. If their “attention
turns away from Siri the voice or Siri the image to Siri the program, Siri again becomes a
thing” (Guzman, 2015, p. 195); thus, she concludes, people “recognize certain character-
istics of humans and machines within them” (p. 257). A. Edwards (2018) found a similar
inconsistency, when she asked her participants to choose two out of the three entities (social
robot, human, and primate), that had more in common in relation to the third. Although,
participants combined humans and primates in opposition to social robots, based on robots
being articial and non-living things, some coupled robots and humans in reference to
their (assumed) resemblance in embodiment, intellect, and behavior, and the capability to
interact socially through talking and understanding.
Voice-Based Agents as Objects of Equilibration
Once the classication of an object is in doubt, the irritation has to be resolved. People
need to decide if machines are “mere things . . . or someone who matters . . . or something
altogether dierent that is situated in between the one and the other” (Gunkel, 2020, p. 56).
Piaget (1970/1974) refers to this as equilibration—a balancing and self-regulating process,
which is achieved through assimilation or accommodation (Figure 1).
Equilibration of Doubt
Raising of Doubt Irritation
Assimilation
Subjectification Objectification
Accommodation
Modification Hybridization
FIGURE 1 Equilibration of Doubt
(Inspired by Piaget, 1970/1974 and Geser, 1989)
Assimilation is the assignment of an irritating object, such as the VBA, to an exist-
ing scheme: the thing scheme (objectication) or the person scheme (subjectication). As
a result, people overestimate either the VBAs objecthood or subjecthood. Particularly for
Siri, Guzman (2015) found that some people explicitly described it as an entity, while others
viewed it as a device. Similar associations were found by Purington and colleagues (2017)
in the user comments about Alexa (Amazon Echo). ey demonstrated that, although the
objectifying pronoun “it” was used by the majority of the authors, some favored the per-
sonifying pronoun “she.” e preference of objectication is conrmed for social robots by
A. Edwards (2018), where more than half of the respondents regarded social robots as
things in contrast to living and natural subjects such as humans or apes. erefore, we
Etzrodt and Engesser 61
formulate the hypothesis (H1): VBAs are assimilated more oen to the thing scheme than
to the person scheme.
Accommodation refers either to the modication of an existing scheme or to the cre-
ation of a new one (Piaget, 1970/1974). Regarding the modication of the object–subject
classication, the thing scheme can be modied by adding person attributes or the person
scheme can be enriched with thing attributes. Apart from that, people may build a hybrid
scheme, featuring a more or less balanced combination of attributes from things and per-
sons (hybridization). While modication draws on existing heuristics and changes them
slightly, hybridization requires the active acquisition and construction of completely new
heuristics.
Research on VBAs implies accommodation: In addition to the “spectrum from fully
human to fully machine,” Guzman (2015, p. 227) identied an “overlap in the middle” con-
cerning the ontology of Siri, as a result of a recongured “understanding of humans and
machines to the degree that we now share characteristics” (p. 257). Purington and col-
leagues (2017) found a mixed use of the pronouns “she” and “it” for Alexa in the same user
comment. Furthermore, some user reactions suggest accommodation through the simulta-
neous activation of social and non-social scripts, such as inappropriate, rude, or insensitive
social behavior toward articial agents: For example, people abuse social robots (Broad-
bent, 2017), and direct bullying and sexual harassment toward VBAs like Alexa (Cercas
Curry & Rieser, 2018).
Inuences on the Classication of Voice-Based Agents
Although social reactions toward computers are fundamentally human, exist in all groups,
and occur even in the case of weak social cues (Reeves & Nass, 1996), there is evidence that
object–subject classication varies due to factors at the levels of situation, user, and agent.
us, we ask to what extent do factors at the levels of the situation, the user, and the agent
inuence the object–subject classication of VBAs (RQ2)?
Attributes of the Situation
e ontological classication of machines can dier between various situational interac-
tions and social contexts. A. Edwards and colleagues (2019) found that a positive expec-
tancy violation during an initial interaction with a social robot reinforces the feeling of
social presence and reduces uncertainty. Leite and colleagues (2013) suggested a change in
relationship through prolonged interaction. Furthermore, personication increased when
Alexa was embedded in multi-person households such as families (Lopatovska & Williams,
2018; Purington et al., 2017). Against this backdrop, we assume that previous interactions
with VBAs aect the classication (H2) and its use in the presence of others increases the
classication’s preference for the person scheme (H3).
Attributes of the User
Age appears to be a major inuence on classication. Children are known to attribute sub-
ject status to objects in general (Piaget, 1970/1974) and articial agents (Epley et al., 2007;
62 Human-Machine Communication
Turkle, 1984/2005) more strongly than adults do. Apart from that, users’ attributes seem to
aect the perceived attributes of the agents rather than their classication. In this context,
age in general aects perceived similarities between a participant’s and a robot’s personality
traits (Woods et al., 2007). In contrast, the impact of gender appears inconsistent. Gender
neither aects the perception of a social presence for synthesized voices (K. M. Lee & Nass,
2005), nor the evaluation of attery behavior of the VBA (E. J. Lee, 2008). However, the
attribution of personality to robots may be gender-specic (Woods et al., 2007); and match-
ing genders of the user and the VBA alters social reactions (E. J. Lee et al., 2000). Besides,
personality traits of the user have been crucial for research, reecting dierences between
individuals as well as parallels in interaction in general. Although research did not nd
eects of the user’s personality on the perception of social presence, matching personalities
of user and VBA may inuence perceived characteristics—regarding the trait “extravert–
introvert” (Nass & Lee, 2001), and similarity—regarding the trait “neuroticism–emotional
stability” (Woods et al., 2007). Closely related to personality is a person’s anity for tech-
nology (Attig et al., 2017), which is particularly noticeable regarding the novelty of VBAs.
Attributes of the Agent
e VBAs conversational mode exhibits subject-likeness, involving expressed eective and
meaningful behavior and the imputation of agency and mind. However, it is still unclear to
what extent they are relevant for the object–subject classication.
Agency as expression of eective behavior. Due to the recently increased capabilities
of machines to mimic natural human behavior their genuine agency is conveyed more
strongly (Guzman, 2018). VBAs can directly answer users, communicate with and control
other smart objects in their environment (e.g., switching lights on and o), collect and pres-
ent information from the internet, activate apps, or initiate purchases. us, in the context
of VBAs, the term agency refers to an eect, originating from interaction with the envi-
ronment and other beings, which may be interpreted as behavior similar to humans (e.g.,
A. Edwards, 2018). Within the framework of social interaction theory, this eect refers
to the ability of interdependence (Simmel, 1908) and orientation (Weber, 1922) toward the
behavior of others—thus, they are aecting others and are aected by others. e most
basic indicator for interdependence between VBAs and users is the VBAs’ expression of
receptiveness to vocal commands. For instance, Alexa lights up a blue ring on the Amazon
Echo and a blue line on the Amazon Echo Show to indicate it is “listening.” To demonstrate
the ability of orientation the VBA needs to express attentiveness in the rst place (Biocca et
al., 2003). Alexa does this by additionally turning a brighter light toward the direction of the
sound source in most Echo devices. e feeling of receptivity increases the perception of
sociality—including that of computers (Burgoon et al., 1999) and agency in general (Appel
et al., 2012; Guadagno et al., 2007). However, it is unclear to what degree the VBAs classi-
cation is altered if their behavior is perceived as eective.
Etzrodt and Engesser 63
Mind as expression of meaningful behavior. e human-like voice and the use of language
are expressions of meaningful behavior, which is closely linked to consciousness (Reichertz,
2014), and a theory of mind (Epley et al., 2007; Premack & Woodru, 1978). us, the
language- and voice-based subjectivity transcends the ability to interact eectively by add-
ing meaning to its actions. Meaningful behavior is crucial for orientation in any social
interaction and based on mutual understanding, on similarity in thinking and/or feeling
(to some degree), and the assumption of intentional actions. at is, before an orienta-
tion toward the actions of others is possible, these actions must be understood based on
ones own and the anticipated other’s experiences within a shared (natural or social) world
(Schütz, 1974). In particular, the conversational mode of VBAs incorporates such specic
assumptions of similarity in thinking and feeling to a certain degree: First, the use of words of
a certain language, and that VBAs understand these words, refers to a specic shared social
world between users and VBAs. Second, by referring to themselves as “I,” VBAs suggest that
they are person-like, self-conscious entities, which operate according to the same rules as
their human users.
Furthermore, the meaning of actions is closely related to the assumption of intentional-
ity, in contrast to accidental actions or for purely physical reasons (Simmel, 1908). e sim-
plest complex of intentionality that can be attributed to an action are motives (Schütz, 1974).
According to Schütz, it is sucient to put oneself in a typical position with typical motives
and plans. Personality constitutes the origin of these typical motives, attitudes, and mindsets
of persons as subjects. It contains both the assumption about behavior that is based on typi-
cal human nature, typical decisions, reactions, or feelings and on ways in which individuals
dier from each other (Buss, 2008). erefore, a perceived personality suggests a subject
who chose to act in a specic manner, but theoretically could have responded in a dierent
way (Higgins & Scholer, 2008). As a result, if personality is attributed to a VBA its actions
may be no longer viewed as random—regardless of whether they were programmed to imi-
tate intention or whether they are behaving intentionally by nature. Research conrms the
attribution of personality traits to synthetic voices (e.g., C. Edwards et al., 2019; Ray, 1986)
and very specic personalities to the commercial VBAs Alexa and the Google Assistant
(Garcia et al., 2018). However, it is uncertain whether the quantity or quality of perceived
personality traits of the VBA alter the object–subject classication. Previous CASA research
has concentrated on the quality such as the personality trait “extraversion” in interpersonal
interactions and uncovered extraverted voices are generally perceived as more socially pres-
ent (e.g., K. M. Lee, 2004; Nass & Lee, 2001; Nass & Moon, 2000). Although extraversion
may be relevant for social reactions toward VBAs, it is unanswered if extraversion or other
personality traits or that personality traits are attributed in the rst place are relevant for
their classication.
Research Questions and Hypotheses
An overview of the relevant variables, research questions, and hypothesized inuences can
be found in the theoretical model (see Figure 2).
64 Human-Machine Communication
Subject-Object
Classification
Similar Behavior
Interdependence
(Mutual)
Attenti on
Agent
Agency
Age
Gender
User
Personality Traits
Technical Affinity
Similar Thinking
Similar Feeling
(Mutual)
Understanding
Agent
Mind
Personality Traits
Person
Thing
Situation
Situational
Interaction
Social Context
RQ1RQ2
Hybrid
(H1)
(H2, H3)
Alexa
Google Assistant
Modified
Thing
Modified
Person
FIGURE 2 Theoretical Factor Model
Method
Sample
In late 2018, we conducted an online survey with a demonstrational design (N = 853) among
all students of a large German university, recruited via the university’s student mailing list
(response rate of 2.6%) and randomly assigned to either the VBA Alexa (Amazon Echo) or
the Google Assistant (Google Home). Participants had a mean age of 24, ranging from 17
to 50 years, 52% of whom were male, 76% were undergraduates, and 23% graduates. e
sample was (on a 6-point scale) above average creative (M = 3.67, SD = 1.00), conscientious
(M = 3.46, SD = 0.86), and emotionally stable (M = 3.2, SD = 0.93), as well as moderately
agreeable (M = 3.1, SD = 0.86) and extraverted (M = 3.1, SD = 1.09). It expressed an anity
for technology above average (M = 4.01, SD = 1.20, 6-point scale).
Most of the participants already knew the names Alexa (96%), and the Google Assis-
tant (71%) in general, and 84% knew their assigned VBA. Although one third owned the
assigned Google Assistant (33%), only some possessed Alexa (7%). e primary sources of
knowing Alexa were indirect ones: advertisement (43%) and contact through other people
(24%), which however, were also important for Google Assistant (17% advertisement, 12%
other people). Other indirect sources were non-ctive media (Alexa: 16%, Google Assis-
tant: 7%) and rarely ctive media (Alexa: 6%, Google Assistant: 1%). If the participants had
used their assigned VBA prior to the survey’s demonstrational interaction, most of them
had used it primarily alone (60%) and moderately frequent (M = 3.3, SD = 1.45, 5-point
scale).
According to its average age the student sample belonged to a cohort widely labelled as
Generation Z, which diers from the previous cohort (commonly labeled as Millennials)
Etzrodt and Engesser 65
in political, economic, and social factors (Dimock, 2019; Singh, 2014). Regarding these dif-
ferences of generations and the classications development during aging, the study was
repeated with an older sample (Millennials)—all employees of the same university, recruited
via the university’s sta mailing list (response rate of 6.2%)—to validate the ndings and
specify cross-generational eects. e sta sample (N = 435) was, on average, 10 years older
(with a mean age of 33, from 18 to 65 years), 65% were graduates, 26% were undergradu-
ates, and 3% had a doctoral degree. Participants were slightly more conscientious (M = 3.71,
SD = 0.78) and had less anity for technology (M = 3.96, SD = 1.20), however, their prior
experiences with the assigned VBA resembled those of the student sample.
Procedure
Although both assistants’ German voices were female, they diered in their characters,
the way they were advertised, and their manufacturing companies’ image. To distinguish
between possible variations caused by the mentioned dierences and general classications
of VBAs, one group of participants assessed Alexa, another the Google Assistant. To get
impressions as close as possible to the true perception of the two VBAs, we chose a demon-
strational design by simulating interactions with pre-recorded videos with the original
answers of the voice-based loudspeaker variants of the Google Assistant or Alexa to pre-
dened questions in German (Table A1 in the study’s OSF repository). Before the simulated
interaction, participants reported their previous experiences with various VBAs (including
the assigned VBA) and typical usage situations. During the simulated interaction, they acti-
vated four videos of the VBA’s answer one aer the other by clicking on the question. Aer
the interaction, they classied the VBA and assessed perceived attributes.
Questions for the VBAs were selected that had been advertised previously by Amazon
or Google as preferable interaction features (e.g., in commercials or on the website), and
that had the potential to exhibit personality characteristics of the VBA. If a VBA provided
multiple answers to the same question, we randomly selected one. Because people form
their impressions within the rst few seconds of contact with a voice (Ray, 1986), the video
sequence for each simulated interaction had a duration between 7 and 17 seconds. e vid-
eos can be obtained from this study’s OSF repository.
Measures
Object–Subject Classication. To examine the object–subject classication, we drew on
the diametrical relation of the thing scheme and the person scheme described above and
asked participants: “What would you say, is Alexa [or the Google Assistant] rather like
a thing (object) to you or rather like a person (subject)?” Because “person” refers to the
status “personhood,” it can be assumed with Hubbard (2011) that it is a gradual assign-
ment, whose highest degree is represented by the term “person.” To address the continuum
between the schemes we used a 100-point scale, consisting of the two poles “thing (object)”
and “person (subject).
As discussed in Etzrodt (2021), the broad-scale allowed intuitive answers—independent
of the participant’s ability to verbalize the classication (Turkle, 1984/2005, p. 48). It also
allowed to detect minor forms of accommodation and to distinguish between modication
66 Human-Machine Communication
and hybridization (see Figure 3). As described by the author, classication as the result
of assimilation into the thing or the person scheme refers to the absence of any previous
accommodation, indicated on the scale by the ratings of 1 or 101. us, VBAs are added
to the existing scheme (thing or person), but the scheme itself does neither get in conict
with the other nor does it change. Classication as a result of accommodation depends
on pre-existing structures (Piaget, 1974, p. 34), referring to a change or rearmation of
the categories’ borders (Turkle, 2005). Hence, Etzrodt (2021) concludes, the accommodated
classication is measured by a weak or strong merging of the thing and the person scheme,
implied if people were distancing from one of the poles on the scale. A weak merging is
represented by ratings close to one of the schemes (2–33 and 67–100), indicating the mod-
ication of a dominant scheme by implementing elements of the other. A strong merging
is represented by ratings near the scale’s center (34–66), indicating a hybridization with a
more or less balanced reunion of both schemes. To determine how sophisticated the equil-
ibration process was, we asked how condent participants were in their classication on a
5-point scale (Etzrodt, 2021).
1 101
Assimilation
Accommodation
2 - 33 34- 66 67- 100
Thing - Scheme
Person - Scheme
Modified
Thing - Scheme
Modified
Person - Scheme
Hybrid
Scheme
FIGURE 3 Equilibration on the Object–Subject Classication Scale (1 to 101),
Etzrodt (2021)
Attributes of the situation were measured as previous knowledge of and interactions with
the assigned VBA and the social contexts in which these interactions are usually embedded.
Previous knowledge assessed if the participants had ever heard about or knew any VBA
independent of their assigned VBA. Previous interactions dierentiate between participants
who had contact with their assigned VBA for the rst time through our survey or knew this
VBA solely from ction or non-ction media, advertising, or had seen others using it, but
had none or only minimal prior interactions, and those who had continuous previous inter-
actions through ownership. erefore, the rst group can be assumed to have had none or
only few equilibration processes before the study. In contrast, the latter group was likely to
have had undergone this process several times. In addition, we considered the frequency of
use on a 5-level rating scale from “very occasionally” to “very oen.” We assessed the social
contexts of use by asking if participants usually used the VBA in multi-person contexts
(family, friends, or acquaintances) or dyad contexts (the absence of other people).
Etzrodt and Engesser 67
Attributes of the user were assessed by the user’s personality, anity for technology, and
demographics such as age and gender. Personality was measured with the “Big Five Inven-
tory” short scale using the original 5-level rating scale (Rammstedt et al., 2013). Principal
component analysis (PCA) conrmed the factors agreeableness, conscientiousness, emo-
tional stability, extraversion, culture (creativity) (χ2(45) = 1451.02, p < .001, KMO = .60,
most factor loadings and h2 > .60). “Agreeableness” exhibited the poorest performance and
was interpreted with caution. Anity for Technology was measured with nine items of the
German ATI Scale (Franke et al., 2018), indexed into one component via PCA (χ2(36) =
5375.01, p < .001, KMO = .91, Cronbachs alpha = .92, factor loadings .63–.88; h2 > .60). e
PCAs in the sta sample conrmed these factors (see OSF).
Attributes of the Agent. Agency was conceptualized as an assigned capability to act,
divided into the three dimensions: similarity to the behavior of the respondent (“Does not
behave like me—behaves like me,” 7-level rating scale), attributed attention of the respec-
tive VBA, and the feeling of interdependence. Subsequent operationalizations were realized
either for Alexa or Google Assistant. For easier reading, only the Alexa variant is provided
in the examples. e two latter dimensions were measured with ve items based on the
reduced relational communication scale by Ramirez (2009), using the original 6-level rating
scale, but formulated for a hypothetical situation (“Imagine you and Alexa are having a con-
versation . . . ”). However, the conrmatory factor analysis (CFA) indicated that these items
loaded on the dimensions attention (“How attentive is Alexa to the interaction?”, “How
strongly is Alexa involved in an interaction?”), and interdependence (“How much is Alexa
adapting to the interaction?”, “How ready is Alexa to have an interaction with you?”, “How
strongly does Alexa respond to your comments or questions?”) suggested by Biocca and
colleagues (2003). To assess convergent validity, the standardized factor loadings, average
variance extracted (AVE), and reliabilities (omega and Cronbachs alpha) were examined.
e CFA showed both an excellent model t for the dimensions (χ2 (4) = 8.987, GFI = .994,
CFI = .995, TLI = .989, RMSEA = .044) and a moderate convergent validity (AVE >.51,
omega > .68, Cronbachs alpha > .66), conrmed by the sta sample (see OSF).
Attributes of the Agent. Mind was operationalized as the VBAs similarity in thinking and
feeling (e.g., “inks like me–does not think like me,” 7-level rating scale), the VBAs abil-
ity to understand its user (“How well does Alexa understand you?”, 7-level rating scale),
and attributed personality traits. e modied Minimal Redundancy Scale based on Lang
(2008) was used to measure the VBAs personality on a semantic dierential using the origi-
nal 6-level rating scale, indexed to the Big Five factors. However, not all human personality
traits worked for the VBAs. CFA identied the items warmhearted, seless, over-accurate,
condent, self-contented, open, loving company, and inventive as insucient. Aer their
removal the model produced an excellent t (χ2(80) = 227.330, GFI = .964, CFI = .964,
TLI = .950, RMSEA = .046), a good reliability with most factor loadings larger than 0.7,
and a moderate convergent validity (AVE > 0.40, omega > 0.50), conrmed by the sta
sample (see OSF). As a result, the factor culture includes items exclusively referring to cre-
ativity (creative, imaginable, artistic) and was interpreted accordingly. In line with previous
research (Garcia et al., 2018; Guzman, 2020), participants had problems assigning emotions
to the VBA explaining why the factor emotional stability displays the poorest performance.
68 Human-Machine Communication
e factor was maintained for comparison with subsequent studies but was interpreted
cautiously.
Results
Data was analyzed using R. In particular, the VBAs’ classication and attributes showed
non-normal, mainly positively skewed and heavy-tailed distributions. us, robust
tests (packages WRS2, and robustbase) were used to control the results of common sta-
tistical tests. e stepwise robust regression found several outliers (with weights = 0 or
~= 1). Although, the signicance of the predictors was equal, the estimates diered slightly
between OLS and robust regression. us, to avoid inaccuracy, we report the estimates of
the robust regression. Reported results apply to the student sample and will be validated
by the sta sample. e supplemental tables and gures can be found at this study’s OSF
repository.
Equilibration of the Object–Subject Classication
Almost one out of three participants assimilated VBAs into an existing scheme, while more
than two out of three had accommodated their schemes (RQ1, Figure 4).
0.00
0.02
0.04
0.06
Density
Thing scheme
Modified
thing
scheme
Hybrid
scheme
Modified
person
scheme
Person scheme
0 25 50 75 100
0 25 50 75 100
Object–Subject Classification (1 = Thing, 101 = Person)
FIGURE 4 Density Plot and Box Plot‚ of the Equilibration
on the Object-Subject Scale, Student Sample
As predicted (H1), in case people did assimilate, they almost always objectied VBAs
(Table 1). In contrast, apart from two people, the VBAs were not subjectied at all. In case
people accommodated, most of them classied the VBAs into modied schemes primarily
with respect to the thing scheme (49%, Table 1), whereas only a minimal number of people
(2%) modied the person scheme. us, VBAs were mainly classied as things supple-
mented by aspects of a person. However, 17% classied the VBA into a hybridized scheme.
Etzrodt and Engesser 69
TABLE 1 Equilibration of the VBA
Student Sample Sta Sample
n%n%
Assimilation Thing scheme 263 31.0 141 32.7
Modied thing scheme 419 49.4 208 48.3
Accommodation Hybrid scheme 148 17.4 67 15.6
Modied person scheme 17 2.0 15 3.5
Assimilation Person scheme 2 0.2 0 0.0
Note: On a scale from 1 to 101, 1 = thing, 2 to 33 = modied thing scheme, 34 to 66 = hybrid
scheme, 67 to 100 = modied person scheme, 101 = person scheme
Although most of the participants were quite certain with their classication (M = 4.34,
SD = 0.89), the further they moved away from the thing scheme, the more uncertain they
became, r = –.54, t(842) = –18.56, p < .001. However, the LOESS graph (Figure 5) indicates
that the uncertainty increases if the classication moves away from any existing scheme.
Although very few people subjectied VBAs, they did it with the same level of certainty
as those who objectied them. Consequently, modication and hybridization exhibited
uncertainty.
1
2
3
4
5
0 25 50 75 100
Object–Subject Classification (1 = Thing, 101 = Person)
Certainty about the Classification
FIGURE 5 Classication Correlated With Certainty
About the Classication, Student Sample
Inuences on the Object–Subject Classication
To identify relevant inuences on the classication (RQ2) we conducted a stepwise robust
regression (SRR). e SRR indicated that more participants classied Alexa further away
from the pole thing than Google Assistant if the previous experienced situations were held
constant (Table 2, Model 2). However, the explained variance was low, and including the
VBAs’ agency eliminated this eect (Table 2, Models 4).
70 Human-Machine Communication
TABLE 2 Stepwise robust regression (M-estimators) using object–subject classication as the criterion, student sample
Model 1 (n = 849) Model 2 (n = 849) Model 3 (n = 831) Model 4 (n = 683) Model 5 (n = 535)
B [CI] β B [CI] β B [CI] β B [CI] β B [CI] β
(Intercept) 14.47 *** [13.00, 15.95] 12.01 *** [9.33, 14.69] 12.91*** [9.90, 15.94] 14.41*** [11.08, 17.76] 17.64*** [13.58, 21.70]
VBA
Alexa a) 1.53 [-0.57, 3.63] .09 2.82 *[0.34, 5.29] .16 3.44 ** [0.94, 5.93] .20 1.30 [-1.36, 3.96] .07 1.66 [-1.44, 4.77] .09
Previous Situation
Dyad b) 2.79 *[0.30, 5.28] .16 2.38 +[-0.14, 4.90] .14 1.65 [-1.00, 4.30] .09 0.27 [-2.63, 3.16] .02
Frequency of use -0.29 [-1.41, 0.83] -.02 -0.48 [-1.60, 0.65] -.03 0.06 [-1.14, 1.27] .00 0.10 [-1.28, 1.49] .01
Knows VBA in general c) 0.06 [-3.19, 3.31] -.00 -0.99 [-4.28, 2.29] -.06 0.78 [-2.78, 4.34] .04 -0.39 [-4.55, 3.77] -.02
Owns the VBA d) 4.66 ** [1.58, 7.73] .27 4.34 ** [1.23, 7.46] .25 1.47 [-1.80, 4.74] .08 0.47 [-3.26, 4.20] .03
User
Agreeableness 1.55 ** [0.46, 2.64] .09 0.93 [-0.25, 2.10] .05 0.99 [-0.36, 2.34] .06
Conscientiousness -0.17 [-1.27, 0.92] -.01 0.15 [-1.02, 1.32] .01 0.22 [-1.56, 1.12] -.01
Emotional stability -1.19 *[-2.32, -0.06] -.07 -0.39 [-1.60, 0.82] -.02 -0.48 [-1.84, 0.87] -.03
Extraversion -1.24 *[-2.36, -0.12] -.07 -1.09 [-2.28, 0.10] -.06 -0.45 [-1.81, 0.91] -.03
Culture (creativity) -0.32 [-1.40, 0.76] -.02 0.11 [-1.04, 1.26] .01 0.23 [-1.07, 1.54] .01
Anity for technology 1.42 *[0.16, 2.69] .08 0.52 [-0.84, 1.88] .03 0.19 [-1.36, 1.75] .01
Women e) 0.12 [-2.47, 2.71] .01 0.59 [-2.16, 3.34] .03 -0.64 [-3.77, 2.50] -.04
Age -1.96 *** [-3.03, -0.90] -.11 -0.61 [-1.85, 0.62] -.03 -0.25 [-1.68, 1.17] -.01
Agent: Agency
Similar behaving 5.16 *** [3.98, 6.35] .29 3.04*** [1.47, 4.61] .17
Interdependence 0.88 [-0.66, 2.42] .05 0.26 [-1.66, 2.17] .01
VBA’s attention 3.45 *** [1.95, 4.94] .20 2.84 ** [1.09, 4.59] .16
Agent: Mind
Similar thinking 4.01 *** [2.20, 5.82] .23
Similar feeling 1.95 *[0.45, 3.45] .11
VBA’s understanding 1.90 ** [0.47, 3.33] .11
Personality traits (count) 1.71 *[0.18, 3.25] .10
Agreeableness -0.50 [-1.96, 0.95] -.03
Conscientiousness -0.55 [-2.03, 0.93] -.03
Emotional stability -0.14 [-1.55, 1.28] -.01
Extraversion -0.24 [-1.67, 1.18] -.01
Culture (creative) 1.96*[0.43, 3.40] .11
Interactions
Attention * interdepend. -0.13 [-1.19, 0.92] -.01 -0.48 [-1.79, 0.84] -.03
Sim. behaving * thinking -0.82 [-2.18, 0.54] -.05
Sim. behaving * feeling 1.42 +[-0.03, 2.87] .08
Sim. thinking * feeling -1.34 +[-2.81, 0.13] -.08
R2.002 .02 .07 .22 .34
Change in R2.002 .02 .05 .15 .12
Note. + indicates p < .1 * indicates p < .05. ** indicates p < .01. *** indicates p < .001. Dichotomic variables are compared with following omitted (reference) level: a) Google Assistant,
b) Multi-Person Interaction, c) Does not know any VBAs, d) Knows interaction with VBA secondhand, e) Men and diverse
Etzrodt and Engesser 71
Attributes of the Situation. As predicted in H2, previous interactions aected the classi-
cation (Model 2, Table 2). If people owned the VBA they classied it more distanced from
the pole thing than people who presumably equilibrated their classication for the rst
time. Contrary to our assumption (H3), the absence of other people increased the distancing
from the mere thing scheme. However, the explained variance was still less than 1%, and the
eects disappeared aer including the VBAs’ agency (Table 2, Model 4).
Attributes of the User. As age increased, the classication tended toward the thing scheme
(Table 2, Model 3). In contrast, people with more anity for technology tended to distance
from the mere thing scheme (Table 2, Model 3). In line with K. M. Lee & Nass (2005) the
gender of the user was not signicant at all for the classication. However, personality traits
were. Agreeable users were more inclined to classify the VBA distanced from the thing pole,
while more emotionally stable and extraverted users exhibited objectication tendencies.
Nevertheless, the explanatory power of the model remained small and the inclusion of the
VBAs’ attributes eliminated most eects. Whereas agency negated the eects of age, anity
for technology, emotional stability, and agreeableness (Table 2, Model 4) assumed mind
negated the eects of extraversion (Table 2, Model 5).
Attributes of the Agent. Perceived agency contributed signicantly to a classication dis-
tanced from the mere thing scheme and increased the explained variance to 22% (Table
2, Model 4). e VBAs’ attentiveness (M = 3.3, SD = 1.29) and interdependence (M = 3.9,
SD = 1.23) were rated moderately high, whereas their behavior was not perceived as very
similar (M = 2.1, SD = 1.64). However, only increasing attentiveness and similar behavior
increased a distanced classication from the pole thing. e eects held when mind attri-
butes were added but weakened substantially.
Perceived mind increased the explained variance to 34% (Table 2, Model 5). Although
the similarity of mind was rated low, the VBAs’ thinking (M = 2.12, SD = 1.61) was assumed
to be more similar than their feeling (M = 1.60, SD = 1.61). Consistently, distancing of the
thing scheme was predicted to a higher degree by similarity in thinking than in feeling (Table
2, Model 5). In contrast, the VBAs’ ability to understand the user was rated moderately high
(M = 3.64, SD = 1.67) and also aected the distancing from the thing scheme to a simi-
lar amount. Participants assigned on average 16 out of 25 personality items to the VBAs
(M = 16.5, SD = 10.2). ey were perceived as very conscientious (M = 4.9, SD = 0.86),
emotionally stable (M = 4.7, SD = 0.97), agreeable (M = 4.6, SD = 1.00), and extraverted
(M = 4.5, SD = 1.09) but less creative (M = 3.0, SD = 1.33). Alexa was perceived as slightly
more conscientious (M = 5.1, SD = 0.77) than Google Assistant (M = 4.7, SD = 0.91),
t(644.34) = 5.803, p < .001. However, only the number of personality items and the VBAs’
creativity predicted a classication distanced from the mere thing scheme (Table 2, Model 5).
Mediation Analyses. Based on the indications of previous studies (e.g., Nass & Lee, 2001;
Woods et al., 2007), and because the VBAs’ attributes eliminated the eects of the situations’
and the users’ attributes, we investigated whether mediation eects were involved. We focus
on mediated variables that had a signicant impact on the classication in Models 1 to 3,
veried by the second sample (see below), and on mediating variables which were signi-
cantly aecting these. Results are reported in Table A2 (OSF).
72 Human-Machine Communication
Alexa’s classication increasingly distanced from the pole thing, due to a higher per-
ceived capability of understanding the user compared to the Google Assistant. However,
the explained variance is very low, thus the two agents do not dier very much from each
other in this respect. Regarding the eect of the situation, both ownership and the previous
use of the VBAs in the absence of others were mediated by agent attributes. Whereas own-
ership increased the VBAs’ similarity in behaving and thinking and its attributed attentive-
ness, previous use in the absence of other people heightened all agency and mind attributes
of the VBA (except similar feeling), encouraging a classication distanced from the thing
scheme. User attributes exhibit mediation eects for age and extraversion, both increasing
a classication toward the pole thing. Whereas increasing age lowered the attributed atten-
tion, number of personality items in general, and the VBAs’ creativity, more extraverted
users perceived less similarity in feeling and fewer personality items. Eects of matching
the respondents and VBAs personality indicated by previous ndings (Nass & Lee, 2001;
Woods et al., 2007) could not be conrmed.
Based on the results of the stepwise robust regression and the mediation analysis, the
theoretical model was adapted (see Figure A1 in the OSF).
Validation of the Results by the Sta Sample
In this section, we will focus on the most important commonalities and dierences in rela-
tion to the student sample. e sta sample conrmed the VBAs’ classication (Table 1), the
amount of condence (M = 4.45, SD = 0.84) and its relation to the classication, r = –.48,
t(429) = –11.38, p < .001. e inuences on the classication were partly conrmed (Table
A3 in the OSF). Whereas, the impacts of age and extraversion (Model 3), the VBAs’ similar
behaving (Model 4) and thinking, its creativity, and the interaction of similar thinking and
feeling (Model 5) were conrmed, the inuences of Alexa (vs. Google Assistant), ownership
and previous use in the absence of other people on the classication were not (Model 2).
However, the direction of the ownership’s and dyad’s eect remained and—consistent with
the student sample—mediating eects of the VBAs’ attributes occurred, although some
involved attributes diered (Table A4 in the OSF). e widespread impact of the previous
dyadic use on the VBAs’ attention, similar thinking and feeling, and creativity was strength-
ened; the impacts of the users’ age through decreased attributed attention to the VBA and
of the users’ extraversion through decreased perceived similarity in feeling were conrmed.
Discussion
In this paper we analyzed how people classify their counterpart when they interact with
voice-based agents and how attributes of the situation, the user, and the agent inuence this
classication. By referring to Piaget (1970/1974), we introduced an empirically measurable
gradual ontological object–subject classication of VBAs, based on a 100-point scale rang-
ing from thing to person, enabling the identication of the degree to which articial agents
are objects of doubt (Geser, 1989; Reeves & Nass, 1996; Turkle, 1984/2005). Using the VBA
as an example, we have demonstrated the potential of this scale, providing the basis for sys-
tematic investigations into how people classify various machines and how the classication
is aected.
Etzrodt and Engesser 73
Consistent with previous research (e.g., A. Edwards, 2018; Guzman, 2015), the major-
ity of our participants could not denitely classify VBAs. e object–subject classication
ranged between the two schemes thing (“what”) and person (“who”), as predicted by Geser
(1989) and Gunkel (2020). at is, most people were indeed in doubt. How people compen-
sated for this doubt was analyzed through the concept of equilibration (Piaget, 1970/1974):
Whereas some people had assimilated VBAs by objectifying, almost none had subjectied
them. However, most people had accommodated their classication by modifying the thing
scheme, and some even hybridized it, but still with a bias toward the thing scheme. at
is, people rearmed their lines between things and persons (Turkle, 1984/2005, p. 34) by
embedding VBA in the world of things, partly enriched with aspects of the person scheme.
Hence, VBAs are personied things.
By understanding that people classify VBAs as personied things, counterintuitive
ndings (Gambino et al., 2020) can be interpreted more precisely. e thing aspect in the
classication things and personied things, on the one hand, emphasized the VBAs arti-
ciality (see Guzman, 2015), thus encourages the reference “it” (see Purington et al., 2017),
or the separation of machines from humans and apes (see A. Edwards, 2018). e person
aspect of personied things, on the other hand, emphasized the VBAs personhood, causing
the simultaneous use of the pronouns “it” and “she” (Guzman, 2015; Purington et al., 2017)
and the assignment of social machines to (living) beings (see A. Edwards, 2018). Whereas
the aspect of personhood may cause social reactions toward articial agents (see Appel et
al., 2012; Horstmann et al., 2018), their dominant nature as things is likely to be responsible
for rather timid, and normatively undesirable social reactions, such as insults or discourtesy
(see Cercas Curry & Rieser, 2018; C. Edwards et al., 2019).
However, the classication of personied things was characterized by high uncertainty.
e further the participants moved away from the established thing or person scheme, the
less condent they became about their classication. at is, classifying by assimilation is
the easy way, it is the passive (habituated) assignment of an object to an existing scheme.
Whereas accommodation—as an active cognitive process of rearming the boundaries of
the schemes—is fraught with far more doubts. As a result, this classication of personied
things is fragile and unstable and may change signicantly in time—especially regarding
further developments of the agents’ abilities and their societal embedding. Further longitu-
dinal research will be required.
How much VBAs were classied as personied things was—in accordance with pre-
vious research (e.g., A. Edwards, 2018; Moon et al., 2016; Nass & Brave, 2005)—mainly
aected by the perception of the agent’s attributes of agency (behaves similarly, is attentive)
and mind (thinks and feels similar, understands, has a vast personality, especially a creative
one). However, for the same eect on the classication, it required a much lower degree of
mind than agency. Hence, “personied things” expands “social things” (Guzman, 2015) by
implying abilities associated with personhood—even if their degree is low (Hubbard, 2011).
However, attributes of the situation and the user did indirectly alter the classica-
tion by aecting the VBAs’ assumed agency and mind. In accordance with prior research
(A. Edwards et al., 2019; Leite et al., 2013), previous regular interactions positively aected
the perception of VBAs as personied things. In contrast to previous studies (e.g., Pur-
ington et al., 2017), the main use of VBAs in the dyad situation (i.e., in the absence of oth-
ers) caused a stronger classication toward personied things, through increased perceived
74 Human-Machine Communication
agency and mind. One explanation for this may be that the dyad fosters a more intense
experience and lacks the “blueprint” of a human being. As a result, the users can engage
more eectively with the VBA but have only themselves for comparison to project sub-
jectivity. e eect found by Purington et al. may be caused by dierent dynamics. us,
we encourage to further explore the dynamics of dyadic and multi-person environments
regarding the classication.
Corresponding with Piaget (1970/1974), age was a crucial factor for subjectication by
altering the VBAs agency and mind (e.g., Woods et al., 2007): e younger the participants,
the more attentive, creative, and substantial in personality the VBAs were perceived to be.
An essential factor in this respect is the continually increasing experience with objects of the
environment, which gradually complement the classication and thus extend the options
for comparison of included objects in the scheme.
Rather than an eect of matching personalities of the user and the VBA (Nass & Lee,
2001), the ndings were more ambiguous. Although VBAs were seen as very extraverted
across subgroups, introvert people subjectied them more oen, through increased percep-
tion of similarity in feeling and personality items in general. at is, introverts perceived a
richer personality and emotionality in VBAs. Again, this phenomenon can be related to the
above-mentioned projection of the users’ experienced own subjectivity (Piaget, 1970/1974).
As introverts may experience themselves as rich in emotion and personality, even though
they are—compared to extroverts—less inclined to express these qualities to others, they
likely attribute the same intensity of feelings and personality to the weak expressions of
the VBAs.
As with all research, this study has several limitations, of which its hypothetical interac-
tion with its pre-dened questions and answers is one. Additional studies need to validate
the VBAs classication in real interactions and with dierent contents. A second limitation
is the paper’s focus on VBAs. Comparing various evocative technologies may allow a more
distinct classication and may reveal dierences in inuencing attributes. ird, the clas-
sication of an object is the result of a dynamic equilibration process (Piaget, 1970/1974).
Hence, we solely examined a snapshot of the process, not the process itself. However, it is
unclear whether the classication resulted from initial or multiple equilibration processes.
Most of the participants had at least some experience with the VBA, indicating potential
prior equilibrations. Further research on the process itself, concerning initial and repeated
equilibrations, inuences, and related classications, is needed.
To conclude, human-machine communication in the context of voice-based agents
indicates that people communicate with personied things dened by a moderate agency
and a basic mind. e more VBAs behave, think, feel similar, are perceived to be attentive
to, and understand their users, and are at least to some extent creative, the more they are
classied as personied things. Although this requires a moderate agency, a rather limited
mind is sucient. Depending on the opportunities for comparison on dierent levels, the
classication is more or less hybrid. Comparisons take place on an individual (age, person-
ality) and situational level (dyad). Age relates to the amount of experience with compara-
ble evocative objects, users’ personality relates to the comparison with their own behavior
and mind, and the dyadic situation relates to the comparison with another human subject.
However, the classication of VBAs as personied things is still fragile and they remain
objects of doubt.
Etzrodt and Engesser 75
Author Biographies
Katrin Etzrodt (MA) is a research assistant and PhD student at the Chair of Science and
Technology Communication at the Technical University Dresden. From 2017 to 2020 she
was granted a scholarship of the Program for the Promotion of Early-Career Female Scien-
tists of TU Dresden.
https://orcid.org/0000-0001-6515-9985
Sven Engesser (PhD) is a Professor of Science and Technology Communication at the
Technical University Dresden.
https://orcid.org/0000-0003-1638-7548
Acknowledgments
e authors would like to thank the reviewers and the editorial team, whose suggestions
have considerably improved the paper. Additionally, the authors would like to thank Lisa
Weidmüller for her input and feedback at dierent stages of development. Ms. Etzrodt
gratefully acknowledges the support and generosity of the Scholarship Program for the Pro-
motion of Early-Career Female Scientists of TU Dresden, without which the present paper
would not have been possible.
e supplemental material can be obtained from the study’s Open Science Framework
(OSF) repository: https://doi.org/10.17605/OSF.IO/5UNS2
References
Appel, J., von der Pütten, A., Krämer, N. C., & Gratch, J. (2012). Does humanity matter?
Analyzing the importance of social cues and perceived agency of a computer system
for the emergence of social reactions during human-computer interaction. Advances in
Human-Computer Interaction. https://doi.org/10.1155/2012/324694
Attig, C., Wessel, D., & Franke, T. (2017, July 9–14). Assessing personality dierences in
human-technology interaction: An overview of key self-report scales to predict success-
ful interaction. Proceedings of HCI International, Vancouver, BC, Canada. https://doi.
org/10.1007/978-3-319-58750-9_3
Biocca, F., Harms, C., & Burgoon, J. (2003). Towards a more robust theory and measure
of social presence: Review and suggested criteria. Presence, 12, 456–480. https://doi.
org/10.1162/105474603322761270
Broadbent, E. (2017). Interactions with robots: e truths we reveal about ourselves.
Annual Review of Psychology, 68(1), 627–652. https://doi.org/10.1146/annurev-
psych-010416-043958
Burgoon, J. K., Bonito, J. A., Bengtsson, B., Ramirez, Jr., A., Dunbar, N. E., & Miczo, N.
(1999). Testing the interactivity model: Communication processes, partner assess-
ments, and the quality of collaborative work. Journal of Management Information Sys-
tems, 13(6), 33–56. https://doi.org/10.1080/07421222.1999.11518255
76 Human-Machine Communication
Buss, D. M. (2008). Human nature and individual dierences. Evolution of human person-
ality. In O. P. John, R. W. Robins, & L. A. Pervin (Eds.), Handbook of personality. eory
and research (3rd ed., pp. 29–60). e Guilford Press.
Cercas Curry, A., & Rieser, V. (2018, June). #MeToo Alexa: How conversational systems
respond to sexual harassment. Proceedings of ACL Workshop on Ethics in Natural Lan-
guage Processing. https://doi.org/10.18653/v1/W18-0802
Dimock, M. (2019). Dening generations: Where Millennials end and Generation Z begins.
Pew Research Center. http://tony-silva.com/esle/miscstudent/downloadpagearticles/
defgenerations-pew.pdf
Edwards, A. (2018). Animals, humans, and machines: Interactive implications of ontolog-
ical classication. In A. L. Guzman (Ed.), Human-machine communication: Rethinking
communication, technology, and ourselves (pp. 29–49). Peter Lang.
Edwards, A., Edwards, C., Westerman, D., & Spence, P. R. (2019). Initial expectations, inter-
actions, and beyond with social robots. Computers in Human Behavior, 90, 308–314.
https://doi.org/10.1016/j.chb.2018.08.042
Edwards, A., Westerman, D., Edwards, C., & Spence, P. R. (2020). Communication is . . .
transhuman. In A. Tyma & A. Edwards (Eds.), Communication is … Perspectives on
theory (pp. 49–66). Cognella Academic Publishing.
Edwards, C., Edwards, A., Stoll, B., Lin, X., & Massey, N. (2019). Evaluations of an arti-
cial intelligence instructor’s voice: Social identity theory in human-robot interactions.
Computers in Human Behavior, 90, 357–362. https://doi.org/10.1016/j.chb.2018.08.027
Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor the-
ory of anthropomorphism. Psychological Review, 114(4), 864–886. https://doi.
org/10.1037/0033-295X.114.4.864
Etzrodt, K. (2021). e ontological classication of conversational agents. An adaptation
of Piaget’s equilibration theory. In A. Følstad, T. Araujo, S. Papadopoulos, E. L.-C. Law,
E. Luger, M. Goodwin, & P. B. Brandtzaeg (Eds.), CONVERSATIONS 2020—4th Inter-
national Workshop on Chatbot Research. Lecture Notes in Computer Science. Springer.
Franke, T., Attig, C., & Wessel, D. (2018). A personal resource for technology interaction:
Development and validation of the anity for technology interaction (ATI) scale. Inter-
national Journal of Human–Computer Interaction, 35(6), 456–467. https://doi.org/10.10
80/10447318.2018.1456150
Gambino, A., Fox, J., & Ratan, R. A. (2020). Building a stronger CASA: Extending the com-
puters are social actors paradigm. Human-Machine Communication, 1, 71–85. https://
doi.org/10.30658/hmc.1.5
Garcia, D. M. P., Lopez, S. S., & Donis, H. (2018, July). Voice activated virtual assistants
personality perceptions and desires: Comparing personality evaluation frameworks. Pro-
ceedings of International BCS Human Computer Interaction Conference, Swindon
United Kingdom. https://doi.org/10.14236/ewic/HCI2018.40
Geser, H. (1989). Der PC als Interaktionspartner [e PC as interaction partner]. Zeitschri
für Soziologie, 18(3), 230–243. https://doi.org/10.1515/zfsoz-1989-0305
Guadagno, R. E., Blascovich, J., Bailenson, J. N., & McCall, C. (2007). Virtual humans and
persuasion: e eects of agency and behavioral realism. Media Psychology, 10(1),1–22.
Gunkel, D. J. (2020). An introduction to communication and articial intelligence. Wiley.
Etzrodt and Engesser 77
Guzman, A. L. (2015). Imagining the voice in the machine: e ontology of digital social
agents [PhD esis]. University of Illinois at Chicago.
Guzman, A. L. (2018). What is human-machine communication, anyway? In A. L. Guzman
(Ed.), Human-machine communication: Rethinking communication, technology, and
ourselves (pp. 1–28). Peter Lang. https://doi.org/10.3726/b14399
Guzman, A. L. (2019). Voices in and of the machine: Source orientation toward mobile
virtual assistants. Computers in Human Behavior, 90, 343–350. https://doi.org/10.1016/j.
chb.2018.08.009
Guzman, A. L. (2020). Ontological boundaries between humans and computers and the
implications for Human-Machine Communication. Human-Machine Communication,
1, 37–54. https://doi.org/10.30658/hmc.1.3
Higgins, E. T., & Scholer, A. A. (2008). When is personality revealed? A motivated cognition
approach. In O. P. John, R. W. Robins, & L. A. Pervin (Eds.), Handbook of Personality.
eory and Research (3rd ed., pp. 128–207). e Guilford Press.
Horstmann, A. C., Bock, N., Linhuber, E., Szczuka, J. M., Straßmann, C., & Krämer, N. C.
(2018). Do a robot’s social skills and its objection discourage interactants from switching
the robot o? PLoS ONE, 13(7): e0201581. https://doi.org/10.1371/journal.pone.0201581
Hubbard, F. P. (2011). “Do androids dream?”: Personhood and intelligent artifacts. Temple
Law Review, 83, 405–474. https://ssrn.com/abstract=1725983
Knorr-Cetina, K. D. (2001). Objectual practice. In T. R. Schatzki (Ed.), e practice turn in
contemporary theory (pp. 175–188). Routledge, London.
Krummheuer, A. (2015). Technical agency in practice: e enactment of artefacts as
conversation partners, actants and opponents. PsychNology Journal, 13, 179–202.
http://www.psychnology.org/File/PNJ13%282-3%29/PSYCHNOLOGY_JOURNAL_13_2_
KRUMMHEUER.pdf
Lang, D. S. (2008). Soziale Kompetenz und Persönlichkeit. Zusammenhänge zwischen sozia-
ler Kompetenz und den Big Five der Persönlichkeit bei jungen Erwachsenen [Social com-
petence and personality. Correlations between social competence and the Big Five
of personality in young adults] [PhD esis]. Universität Koblenz-Landau. https://
kola.opus.hbz-nrw.de/opus45-kola/frontdoor/deliver/index/docId/264/file/Soziale
KompetenzundPersoenlichkeit.pdf
Latour, B. (1995). Wir sind nie modern gewesen. Versuch einer symmetrischen Anthropologie
[We have never been modern]. Suhrkamp. (Original work published 1991.)
Latour, B. (2005). Reassembling the social. An introduction to Actor-Network-eory (ANT).
Oxford.
Lee, E. J. (2008). Flattery may get computers somewhere, sometimes: e moderating
role of output modality, computer gender, and user gender. International Journal of
Human-Computer Studies, 66(11), 789–800. https://doi.org/10.1016/j.ijhcs.2008.07.009
Lee, E. J., Nass, C., & Brave, S. (2000, April). Can computer-generated speech have gender?
An experimental test of gender stereotype. Proceedings of CHI on Human Factors in
Computing Systems, the Hague the Netherlands. https://doi.org/10.1145/633292.633461
Lee, K. M. (2004). Presence, explicated. Communication eory, 14(1), 27–50. https://doi.
org/10.1111/j.1468-2885.2004.tb00302.x
78 Human-Machine Communication
Lee, K. M., & Nass, C. (2005). Social-psychological origins of feelings of presence: Creating
social presence with machine-generated voices. Media Psychology, 7(1), 31–45. https://
doi.org/10.1207/S1532785XMEP0701_2
Leite, I., Martinho, C., & Paiva, A. (2013). Social robots for long-term interaction: A survey.
International Journal of Social Robotics 5, 291–308. https://doi.org/10.1007/s12369-013-
0178-y
Lombard, M., Lee, S., Sun, W., Xu, K., & Yang, H. (2017). Presence theory. In P. Rössler,
C. A. Honer, & L. van Zoonen (Eds.), e international encyclopedia of media eects
(pp. 1–13). John Wiley & Sons Inc. https://doi.org/10.1002/9781118783764.wbieme0087
Lopatovska, I., & Williams, H. (2018, March 11–15). Personication of the Amazon Alexa:
BFF or a mindless companion? Proceedings of the ACM CHIIR Conference, New
Brunswick. https://doi.org/10.1145/3176349.3176868
Moon, Y., Kim, K. J., & Shin, D.-H. (2016, July 17–22). Voices of the Internet of ings: An
exploration of multiple voice eects in smart homes. In P. M. Norbert Streitz (Ed.), Pro-
ceedings of the DAPI Conference held as part of HCI International, Toronto, Canada.
https://doi.org/10.1007/978-3-319-39862-4
Nass, C., & Brave, S. (2005). Wired for speech—How voice activates and advances the
human-computer relationship. MIT Press.
Nass, C., & Lee, K.-M. (2001). Does computer-generated speech manifest personality?
Experimental test of recognition, similarity-attraction, and consistency-attraction. Jour-
nal of Experimental Psychology: Applied, 7(3), https://doi.org/10.1037/1076-898X.7.3.171
Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses to computers.
Journal of Social Issues, 56(1), 88–103. https://doi.org/10.1111/0022-4537.00153
Nass, C., Steuer, J., & Tauber, E. R. (1994). Computers are social actors. Proceedings of CHI
Conference on Human Factors in Computing Systems, Boston Massachusetts USA.
https://doi.org/10.1145/191666.191703
Newman, N. (2018). Digital news project. e future of the voice and the implications of news.
Reuters Institute for the Study of Journalism & University of Oxford. https://reuters
institute.politics.ox.ac.uk/our-research/future-voice-and-implications-news
Piaget, J. (1974). Abriß der genetischen Epistemologie [e principles of genetic epistemol-
ogy] (F. Kubli, Trans.). Walter-Verlag AG Olten. (Original work published 1970.)
Premack, D., & Woodru, G. (1978). Does the chimpanzee have a theory of mind? Behav-
ioral and Brain Sciences, 1(4), 515–526. https://doi.org/10.1017/S0140525X00076512
Purington, A., Ta, J. G., Sannon, S., Bazarova, N. N., & Taylor, S. H. (2017, May). “Alexa is
my new BFF”: Social roles, user satisfaction, and personication of the Amazon Echo.
CHI’17 Extended Abstracts. https://doi.org/10.1145/3027063.3053246
Ramirez Jr., A. (2009). e eect of interactivity on initial interactions: e inuence of
information seeking role on computer-mediated interaction. Western Journal of Com-
munication, 73(3), 300–325. https://doi.org/10.1080/10570310903082040
Rammstedt, B., Kemper, C. J., Klein, M. C., Beierlein, C., & Kovaleva, A. (2013). A short
scale for assessing the big ve dimensions of personality: 10 item big ve inventory (BFI-
10). Methoden, Daten, Analysen, 7(2), 233–249. https://doi.org/10.12758/mda.2013.013
Ray, G. B. (1986). Vocally cued personality prototypes: An implicit personal-
ity theory approach. Communication Monographs, 53(3), 266–276. https://doi.
org/10.1080/03637758609376141
Etzrodt and Engesser 79
Reeves, B., & Nass, C. I. (1996). e media equation: How people treat computers, television,
and new media like real people and places. Cambridge University Press.
Reichertz, J. (2014). Von Menschen und Dingen. Wer handelt hier eigentlich [Of people and
things. Who is actually acting]? In A. Poferl & N. Schröer, (Eds.), Wer oder was handelt?
Zum Subjektverständnis der hermeneutischen Wissenssoziologie, 95–120. Springer Fach-
medien. https://doi.org/10.1007/978-3-658-02521-2
Roßler, G. (2008). Kleine Galerie neuer Dingbegrie: Hybriden, Quasi-Objekte, Grenzob-
jekte, epistemische Dinge [Small gallery of new terms: hybrids, quasi-objects, boundary
objects, epistemic things]. In G. Knef, M. Schröer, & E. Schüttpelz (Eds.), Bruno Latours
Kollektive (pp. 76–107). Suhrkamp.
Schütz, A. (1974). Der sinnhae Auau der sozialen Welt: Eine Einleitung in die verstehende
Soziologie [e phenomenology of the social world: An introduction into verstehende
sociology]. Suhrkamp, Frankfurt.
Short, J., Williams, E., & Christie, B. (1976). e social psychology of telecommunications.
Wile y.
Simmel, G. (1908). Soziologie Untersuchungen über die Formen der Vergesellschaung
[Inquiries into the construction of social forms]. Duncker & Humbolt.
Singh, A. (2014). Challenges and issues of Generation Z. In Journal of Business and Man-
agement, 16(7), 59–63. https://pdfs.semanticscholar.org/b109/73a5c6d11f37542adc34455
bb0c7fbcbbb9d.pdf
Suchman, L. (2011). Subject objects. Feminist eory, 12(2), 119–145. https://doi.
org/10.1177/1464700111404205
Turing, A. M. (1950). Computing machinery and intelligence. Mind, 59(236), 433–460.
https://www.jstor.org/stable/2251299
Turkle, S. (2005). e second self (Twentieth anniversary edition). e MIT Press. (First
edition published by Simon & Schuster, Inc., New York, 1984). https://doi.org/10.7551/
mitpress/6115.001.0001
Weber, M. (1922). Wirtscha und Gesellscha [Economy and society]. Mohr, Tübingen.
Woods, S., Dautenhahn, K., Kaouri, C., te Boekhorst, R., & Koay, K. L. (2007). Is this robot
like me? Links between human and robot personality traits. Interaction Studies, 8(2),
281–305. https://doi.org/10.1075/is.8.2.06woo
... This indicates that some kind of ambiguity was perceived nonetheless-especially compared to the gender perception for the distinctly gendered voices. In accordance to Piaget (1997), it could be interpreted as an evoked equilibration process due to the uncertainty in gender ascription: Therefore-similar to other ambiguous objects (Etzrodt & Engesser, 2021)-when confronted with ambiguity, people most of the time use the less exhausting strategy of accommodating the voice by modifying an existing category stemming from ...
... Besides the perceived topic gender, age appeared to be the only influential factor, indicating that older people perceived the voice as more ambiguous, whereas younger people tended toward a more male assessment on average. A reason for this might be availability heuristics as described in the theory section: At increasing age, people have had more chances to encounter voices with acoustic parameters that do not fit into the prevailing genderism which might have led to the accommodation (Etzrodt & Engesser, 2021) of their gender scheme, enabling them to classify ambiguity. ...
... Hence, it is plausible that VBAs' application as taskfulfilling assistants in everyday life and their artificiality cause this more instrumental bias. This strengthens previous reflections on the emergence of novel heuristics regarding artificial agents (e.g., Etzrodt & Engesser, 2021;Gambino et al., 2020;Guzman, 2020). If VBAs now have their own heuristics as this indicates, traditional gender stereotypes might not be as relevant for their classification anymore, causing the lack of stereotype effects. ...
Article
Full-text available
Recently emerging synthetic acoustically gender-ambiguous voices could contribute to dissolving the still prevailing genderism. Yet, are we indeed perceiving these voices as “unassignable”? Or are we trying to assimilate them into existing genders? To investigate the perceived ambiguity, we conducted an explorative 3 (male, female, ambiguous voice) × 3 (male, female, ambiguous topic) experiment. We found that, although participants perceived the gender-ambiguous voice as ambiguous, they used a profoundly wide range of the scale, indicating tendencies toward a gender. We uncovered a mild dissolve of gender roles. Neither the listener’s gender nor the personal gender stereotypes impacted the perception. However, the perceived topic gender indicated the perceived voice gender, and younger people tended to perceive a more male-like gender.
... Thus, they inhabit crucial primary cues (Lombard and Xu, 2021) that have a high potential to evoke social responses. Previous studies found various (social) reactions to and relationships with different technologies, including computers in general (e.g., Turkle, 1984Turkle, /2005Reeves and Nass, 1996), digital agents (e.g., Sundar, 2008;Skjuve et al., 2021), talking computers (e.g., Burgoon et al., 1999;Nass and Brave, 2005), social robots (e.g., Edwards et al., 2019;Laban et al., 2021), and commercial VPAs (e.g., Etzrodt and Engesser, 2021;Guzman, 2015;Shani et al., 2021;Wienrich et al., 2021;Chung et al., 2021). ...
... However, the situations in which more than one person is involved with the artificial agent are becoming increasingly important, especially when technologies integrate into the users' everyday environment. Recent studies on robots (e.g., Diederich et al., 2019;Thompson and Gillan, 2010;Fortunati et al., 2020) and commercial VPAs (Etzrodt and Engesser, 2021;Lopatovska and Williams, 2018;Porcheron et al., 2018;Purington et al., 2017;Raveh et al., 2019) indicate that the social situation regarding how many people interact with the agent alters the interaction with and the perception of the agent as well as how people might relate to it. However, their results remain vague and are primarily collateral findings. ...
... In contrast, Etzrodt and Engesser (2021) uncovered that people who had primarily prior dyadic interactions with a VPA rated the VPA's self-similarity higher than those with prior multi-person interactions, which in turn moved the classification towards subjectivity. The authors suggested that the lower self-similarity in triads may originate from the presence of the second person, emphasizing the difference between the first person and the VPA in terms of subjectivity by serving as a blueprint for subjects and subject-like behavior. ...
Article
As commercial voice-based personal agents (VPAs) like Google Assistant increasingly penetrate people’s private social habitats, sometimes involving more than one user, these social situations are gaining importance for how people define the human-machine relationship (HMR). The paper contributes to the understanding of the situation’s impact on HMR on a theoretical and methodological level. First, Georg Simmel’s theory on the “Quantitative determination of the group” is applied to the HMR. A 2x1 between-subjects quasi-experiment (N = 100) contrasted the defined HMR in dyadic social situations (one human interacting with the Google Assistant) to the defined HMR in triadic social situations (two humans interacting with the Google Assistant). Second, the method of central tendency analysis was extended by the more robust and informative comparison of distributions and quantiles using the two-sample Kolmogorov–Smirnov test and the shift function. The results show that the triadic situation, compared to the dyadic one, led to a more confounded categorization of the VPA’s subjecthood in terms of self-similarity, while simultaneously strengthening a definition of the relationship that resembled those of a business relation through lowered intimacy and feedback, mainly grounded in a more realistic definition of the agent’s inability to understand affects. In contrast to Simmel’s inter-human theory the relationship’s dimension of reciprocity and commitment remained unaffected by the situation. The paper discusses how these effects and non-effects of the triad could be explained by drawing on Simmel as well as peculiarities of HMR and methodology. Finally, it offers preliminary hypotheses about the situation’s implications for the HMR and outlines avenues for future research. Free download until 09/30/2022 at: https://authors.elsevier.com/a/1fZVe3pfaRpm5N
... A second important question in the debate about gender-neutral, ambiguous, or androgynous VBA voices is whether people actually form different stereotypes of them and react to them differently than acoustically male or female VBAs. On this second question, the ambiguous voice was situated between the male and female voices on perceived instrumentality (a masculine stereotype) and expressiveness (a feminine stereotype), which "strengthens previous reflections on the emergence of novel heuristics regarding artificial agents (e.g., Etzrodt & Engesser, 2021;Gambino et al., 2020;Guzman, 2020)" (p. 64). ...
... Continuing with a provocative and valuable interpretation of the finding that respondent gender was not influential to perceptions of VBA gender, Mooshammer and Eztrodt find additional support for the operation of a VBA-specific gender heuristic in the lack of impact of the theoretically indicated over-exclusion of the ambiguous voice from the participants' own gender. If the VBA was not perceived as a gendered person, but as a gendered 'personified thing' (Etzrodt & Engesser, 2021) or 'social thing' (Guzman, 2015), the VBA is already part of an outgroup, independent of its gender. " (p. ...
Article
Full-text available
In this introduction to the fifth volume of the journal Human-Machine Communication, we present and discuss the five articles focusing on gender and human-machine communication. In this essay, we will analyze the theme of gender, including how this notion has historically and politically been set up, and for what reasons. We will start by considering gender in in-person communication, then we will progress to consider what happens to gender when it is mediated by the most important ICTs that preceded HMC: the telephone, mobile phone, and computer-mediated communication (CMC). We outline the historical framework necessary to analyze the last section of the essay, which focuses on gender in HMC. In the conclusion, we will set up some final sociological and political reflections on the social meaning of these technologies for gender and specifically for women.
... Although the impression of something lies in a pre-cognitive dimension, it is essential to explore it since it "often shapes our final appraisal of that object" (de Graaf & Allouch, 2017, p. 28). These findings, which are in line with the studies carried out by Fortunati et al. (2021) and cited above, seem to point mainly to the digital world in which Alexa lives, while, for example, Etzrodt and Engesser (2021) found that VBAs were conceptualized as "personified things. " However, Etzrodt and Engesser's findings and the current study may be the result of an artifact of methodology. ...
... Respondents did not consider Alexa as a mere gadget but as the outcome of the most innovative high-tech industry, with one foot in the future. Respondents did not report any hybridization or uncertain boundaries between Alexa and humans, although an increasing amount of scientific literature reflects the blurring boundaries between humans and machines (Etzrodt & Engesser, 2021;Weidmüller, 2022). Decidedly, these first three categories accounted for 75.1% of the words that form the core of the social representations of Alexa. ...
Article
Full-text available
Mainly, the scholarly debate on Alexa has focused on sexist/anti-woman gender representations in the everyday life of many families, on a cluster of themes such as privacy, insecurity, and trust, and on the world of education and health. This paper takes another stance and explores via online survey methodology how university student respondents in two countries (the United States, n = 333; and Italy, n = 322) perceive Alexa’s image and gender, what they expect from this voice-based assistant, and how they would like Alexa to be. Results of a free association exercise showed that Alexa’s image was scarcely embodied or explicitly gendered. Rather, Alexa was associated with a distinct category of being—the VBA, virtual assistant, or digital helper—with which one talks, and which possesses praiseworthy technical and social traits. Expectations of Alexa and desires regarding Alexa’s ideal performance are presented and compared across the two country samples.
... On the one hand, there has been a proliferation of HMC as an object of investigation. Digital interlocutors, such as Artificial Intelligence (e.g., Gunkel 2020; Guzman and Lewis 2020; Schäfer and Wessler 2020; Sundar and Lee 2022), avatars (e.g., Banks and Bowman 2016), chatbots (e.g., Araujo 2018;Edwards et al. 2014;Brandtzaeg and Følstad 2017;Gehl and Bakardjieva 2017), voice-based assistants (e.g., Etzrodt and Engesser 2021;Humphry and Chesher 2021;Natale and Cooke 2021), and social robots (e.g., Hepp 2020; Fortunati 2018; Peter and Kühne 2018) are on the rise. As a result, we are witnessing a profound change, in which communication through technologies is extended by communication with technologies (cf. ...
Article
COVID-19 has fundamentally changed the way people connect, collaborate, and socialize. With the ongoing pandemic amplifying people’s feelings of loneliness, voice assistants are growing as a pandemic-era staple of supporting people’s well-being and mitigating feelings of disconnectedness. Combining the uses and gratification approach and theory of anthropomorphism, this study examined social attraction and social presence as drivers for people to anthropomorphize voice assistants during the COVID-19 pandemic. Further, this study investigated whether loneliness, social disconnection, and attachment can moderate the effect of social attraction and social presence on the anthropomorphism of voice assistants. Drawing on survey data from 458 US voice assistant users, the results indicated that social attraction and social presence positively affect peoples’ anthropomorphism toward voice assistants. Moreover, the moderating effects of loneliness and social disconnection were examined and found positive impacts on the effect of social presence on anthropomorphism. The findings have implications for theorizing the anthropomorphism of digital media when face-to-face communication is less available. This study is also helpful for voice assistants’ developers and brands to design these smart devices appealing to customers and fostering a more customized and more robust user-technology interaction.
Article
Full-text available
When intelligent voice-based assistants (VBAs) present news, they simultaneously act as interlocutors and intermediaries, enabling direct and mediated communication. Hence, this study discusses and investigates empirically how interlocutor and intermediary predictors affect an assessment that is relevant for both: trustworthiness. We conducted a secondary analysis using data from two online surveys in which participants ( N = 1288) had seven quasi-interactions with either Alexa or Google Assistant and calculated hierarchical regression analyses. Results show that (1) interlocutor and intermediary predictors influence people’s trustworthiness assessments when VBAs act as news presenters, and (2) that different trustworthiness dimensions are affected differently: The intermediary predictors (information credibility; company reputation) were more important for the cognition-based trustworthiness dimensions integrity and competence. In contrast, intermediary and interlocutor predictors (ontological classification; source attribution) were almost equally important for the affect-based trustworthiness dimension benevolence.
Article
Full-text available
Communication scholars are increasingly concerned with interactions between humans and communicative agents. These agents, however, are considerably different from digital or social media: They are designed and perceived as life-like communication partners (i.e., as “communicative subjects”), which in turn poses distinct challenges for their empirical study. Hence, in this paper, we document, discuss, and evaluate potentials and pitfalls that typically arise for communication scholars when investigating simulated or non-simulated interactions between humans and chatbots, voice assistants, or social robots. In this paper, we focus on experiments (including pre-recorded stimuli, vignettes and the “Wizard of Oz”-technique) and field studies. Overall, this paper aims to provide guidance and support for communication scholars who want to empirically study human-machine communication. To this end, we not only compile potential challenges, but also recommend specific strategies and approaches. In addition, our reflections on current methodological challenges serve as a starting point for discussions in communication science on how meaning-making between humans and machines can be investigated in the best way possible, as illustrated in the concluding section.
Article
Understanding players' mental models are crucial for game designers who wish to successfully integrate player-AI interactions into their game. However, game designers face the difficult challenge of anticipating how players model these AI agents during gameplay and how they may change their mental models with experience. In this work, we conduct a qualitative study to examine how a pair of players develop mental models of an adversarial AI player during gameplay in the multiplayer drawing game iNNk. We conducted ten gameplay sessions in which two players (n = 20, 10 pairs) worked together to defeat an AI player. As a result of our analysis, we uncovered two dominant dimensions that describe players' mental model development (i.e., focus and style). The first dimension describes the focus of development which refers to what players pay attention to for the development of their mental model (i.e., top-down vs. bottom-up focus). The second dimension describes the differences in the style of development, which refers to how players integrate new information into their mental model (i.e., systematic vs. reactive style). In our preliminary framework, we further note how players process a change when a discrepancy occurs, which we observed occur through comparisons (i.e., compare to other systems, compare to gameplay, compare to self). We offer these results as a preliminary framework for player mental model development to help game designers anticipate how different players may model adversarial AI players during gameplay.
Article
Full-text available
Digital Voice Assistants (DVAs) have become a ubiquitous technology in today’s home and childhood environments. Inspired by (Bernstein and Crowley, J Learn Sci 17:225–247, 2008) original study (n = 60, age 4–7 years) on how children’s ontological conceptualizations of life and technology were systematically associated with their real-world exposure to robotic entities, the current study explored this association for children in their middle childhood (n = 143, age 7–11 years) and with different levels of DVA-exposure. We analyzed correlational survey data from 143 parent–child dyads who were recruited on ‘Amazon Mechanical Turk’ (MTurk). Children’s ontological conceptualization patterns of life and technology were measured by asking them to conceptualize nine prototypical organically living and technological entities (e.g., humans, cats, smartphones, DVAs) with respect to their biology, intelligence, and psychology. Their ontological conceptualization patterns were then associated with their DVA-exposure and additional control variables (e.g., children’s technological affinity, demographic/individual characteristics). Compared to biology and psychology, intelligence was a less differentiating factor for children to differentiate between organically living and technological entities. This differentiation pattern became more pronounced with technological affinity. There was some evidence that children with higher DVA-exposure differentiated more rigorously between organically living and technological entities on the basis of psychology. To the best of our knowledge, this is the first study exploring children’s real-world exposure to DVAs and how it is associated with their conceptual understandings of life and technology. Findings suggest although psychological conceptualizations of technology may become more pronounced with DVA-exposure, it is far from clear such tendencies blur ontological boundaries between life and technology from children’s perspective.
Chapter
Full-text available
This paper focuses on the attributed nature of the voice-based agents Alexa and Google Assistant in conversational contexts. Using Piaget’s equilibration theory, enhanced by Hubbard’s concept of personhood the paper considers how people categorize voice-based agents along a thing–person spectrum and whether this categorization reflects assimilation or accommodation of these technologies. The results of two studies (a hypothetical conversation with the agent via an online-survey, N = 1288, and a real conversation with the agent, N = 105) are indicating a modified classification towards personified things, which is reinforced by younger age and a higher quality of interaction. Implications, limitations, and further research regarding a more detailed classification of conversational agents are discussed.
Article
Full-text available
The computers are social actors framework (CASA), derived from the media equation, explains how people communicate with media and machines demonstrating social potential. Many studies have challenged CASA, yet it has not been revised. We argue that CASA needs to be expanded because people have changed, technologies have changed, and the way people interact with technologies has changed. We discuss the implications of these changes and propose an extension of CASA. Whereas CASA suggests humans mindlessly apply human-human social scripts to interactions with media agents, we argue that humans may develop and apply human-media social scripts to these interactions. Our extension explains previous dissonant findings and expands scholarship regarding human-machine communication, human-computer interaction, human-robot interaction, human-agent interaction, artificial intelligence, and computer-mediated communication.
Article
Full-text available
In human-machine communication, people interact with a communication partner that is of a different ontological nature from themselves. This study examines how people conceptualize ontological differences between humans and computers and the implications of these differences for human-machine communication. Findings based on data from qualitative interviews with 73 U.S. adults regarding disembodied artificial intelligence (AI) technologies (voice-based AI assistants, automated-writing software) show that people differentiate between humans and computers based on origin of being, degree of autonomy, status as tool/tool-user, level of intelligence, emotional capabilities, and inherent flaws. In addition, these ontological boundaries are becoming increasingly blurred as technologies emulate more human-like qualities, such as emotion. This study also demonstrates how people’s conceptualizations of the human-computer divide inform aspects of their interactions with communicative technologies.
Article
Full-text available
Building on the notion that people respond to media as if they were real, switching off a robot which exhibits lifelike behavior implies an interesting situation. In an experimental lab study with a 2x2 between-subjects-design (N = 85), people were given the choice to switch off a robot with which they had just interacted. The style of the interaction was either social (mimicking human behavior) or functional (displaying machinelike behavior). Additionally, the robot either voiced an objection against being switched off or it remained silent. Results show that participants rather let the robot stay switched on when the robot objected. After the functional interaction, people evaluated the robot as less likeable, which in turn led to a reduced stress experience after the switching off situation. Furthermore, individuals hesitated longest when they had experienced a functional interaction in combination with an objecting robot. This unexpected result might be due to the fact that the impression people had formed based on the task-focused behavior of the robot conflicted with the emotional nature of the objection.
Chapter
This chapter considers common perceptions about the fundamental natures of animals, humans, and machines by exploring individuals’ perceptions of the similarities and differences among those groups of agents. Specifically, this chapter focuses on understanding how a cross-section of U.S. American adults used ontological classification to understand and construct the differences between humans, animals, and machines and to explore one of the ways in which such classification may matter: responses to the destruction of the social robot hitchBOT. A sizeable majority of participants classified humans with anthropomorphic animals, with fewer identifying humans with robots, or as fundamentally distinct from both. Thematic analysis relates participants’ reasoning to cultural constructions of the inherent nature of each entity and to the nature and level of their concern for hitchBOT.
Chapter
This edited volume introduces readers to the growing area of Human-Machine Communication (HMC) research within the communication discipline. As defined within this chapter, HMC is the creation of meaning between humans and machines, with technology theorized as a communicator, a subject with which people communicate, instead of a channel through which humans interact with one another. HMC research focuses on the process of communication between human and machine and the implications of encounters between people and technology for individuals, society, and humanity. HMC envelopes communication research within Human-Computer Interaction (HCI), Human-Robot Interaction (HRI), and Human-Agent Interaction (HAI) while at the same time is inclusive of philosophical, critical/cultural, and related approaches regarding the integration of social technologies into everyday spaces. This introduction outlines the fundamental aspects of HMC by answering key questions regarding
Article
This study employs the Computers are Social Actors (CASA) paradigm to extend the predictions of Social Identity Theory (SIT) to human-robot interaction (HRI) in the context of instructional communication. SIT posits that individuals gain a sense of personal worth from the groups with which they identify. Previous research has demonstrated that age group identification is meaningful to individuals’ self-concepts. Results demonstrated that higher age identified students rated the older A.I. voice instructor (representing an out-group member) higher for credibility and social presence and reported more motivation to learn than those students with low age identification. Implications are discussed for SIT and design features of computerized voices.
Article
Research regarding source orientation has demonstrated that when interacting with computers, people direct their communication toward and react toward the technology itself. Users perceive technology to be a source in human-machine communication (HMC). This study provides a new dimension to those findings with regard to source orientation with voice-based, mobile virtual assistants enabled by artificial intelligence (AI). In qualitative interviews regarding their conceptualizations of mobile conversational agents (Apple's Siri, Google Voice, Samsung S-Voice) and their perceptions of interactions with these specific technologies, some participants describe the agent they can hear but not see as a voice in the mobile phone (assistant as distinct entity) while others perceive the technology that they command to be the voice of the phone (assistant as the device). Therefore, congruent with existing research, users of mobile assistants orient toward a technology, instead of thinking they are interacting with a human, but, in contrast to existing research, attend to different technologies. When technologies possess a disembodied voice and are designed with various social cues and degrees of intelligence, the locus and nature of the digital interlocutor is not uniform in people's minds. Link for free access until Jan. 7, 2019: https://authors.elsevier.com/a/1Y4hh2f~UW4lja