=Paper=
{{Paper
|id=Vol-2265/paper3
|storemode=property
|title=A More Human Side of a Chatbot. Analysing Anthropomorphism in Conversations with a Virtual Agent Depending on the Level of Elicited Agent Knowledge
|pdfUrl=https://ceur-ws.org/Vol-2265/paper3.pdf
|volume=Vol-2265
|authors=Amelia La Torre
}}
==A More Human Side of a Chatbot. Analysing Anthropomorphism in Conversations with a Virtual Agent Depending on the Level of Elicited Agent Knowledge==
https://ceur-ws.org/Vol-2265/paper3.pdf
A more human side of a chatbot. Analysing anthropo-
morphism in conversations with a virtual agent depend-
ing on the level of elicited agent knowledge
1
Amelia La Torre
1
Jagiellonian University, Cracow, Poland
amelia.latorre@gmail.com
Abstract. The aim of the research is to analyse anthropomorphism in the context of short inter-
action with a chatbot. Anthropomorphism is defined here as an individual tendency to perceive
a nonhuman object as if it was a human (e.g. attributing human characteristics to it), which can
be observed in individual reaction to an object. As chatbot is similar to a human due to its con-
struction and can be anthropomorphised particularly easy, it is interesting to investigate: (1)
how anthropomorphism manifests in the human-chatbot conversation, especially in the context
of inconsistent chatbot’s behaviour (2) if any individual differences appear in that level of con-
tact (3) can the accessible knowledge about a chatbot influence anthropomorphism in a sponta-
neous conversation with it. Thirty-two chat conversations (16 male) witch Eviebot conversa-
tional agent will be analysed with qualitative methods. Half of it are conducted after experi-
mental manipulation, including the information about Eviebot read by experimenter and short
questions consolidating the provided knowledge, and second half with no additional manipula-
tion. The analysis of conversations will put an accent to the specificity of anthropomorphism in
the presented context, concentrating on the reactions to a chatbot’s inconsistent behaviours. The
possible differences between informed and non-informed group will be investigated. The study
has an explorative character and has to help to characterise anthropomorphism of a chat-bot in a
behavioural level. Conclusions and possible implications for the future research will be
underlined.
Keywords: Anthropomorphism, Chatbot, HRI.
1 Introduction
Antropomorphism is a complex phenomenon, defined in terms of objects’ observable
similarity to a human being (e.g. Demeure, Niewiadomski & Pelachaud, 2011) or as
an individual tendency to perceive an object as if it was a human (e.g. Epley, Waytz
& Cacioppo, 2007). The second definition, common in psychological research, puts
an accent to individual differences in attributing unobservable human characteristics
to a nonhuman agent, which affects the person’s behaviour towards it. The existing
research shows that anthropomorphism influence the people emotional reactions to-
wards an object, such as liking (Siino, Chung & Hinds 2008,: Waytz, Cacioppo &
Epley, 2010), trust, or moral concern (Waytz Cacioppo & Epley 2010), but also antip-
athy (Shwartz 2003), which is manifested in behavioural level. Anthropomorphism is
connected with including an object to the own social circle, and as so it can influence
a social life of a person and the way he deals with other people. For example the stud-
ies suggest, that the contact with anthropomorphised objects can be adaptive for older
people, who are lonely or lose their partners - in those situations also a social robots
can be substitutes for people ( e.g. Ring, Barry, Totzke & Bickmore, 2013). In this
context it is interesting to investigate human interaction witch social robots and chat-
terbots, created especially to perform the social functions. Anthropomorphic reactions
towards such objects can particularly exceed the level of automatic behaviours, and be
connected with developing some level of attachment. Individual differences in an-
thropomorphic reactions towards robots and its determinants are yet to be
investigated.
Three Factor Theory Of Anthropomorphism (Epley, Waytz & Cacioppo, 2007) indi-
cates the determinants of that phenomenon, connected with the person’s specific mo-
tives (sociality and effectance motivation) and to the accessibility of information
about a human which can serve as an explanation of object’s behaviour (elicited agent
knowledge ). The anthropocentric knowledge can be easily used especially if an object
is very similar to a human or the information about an object is not available.
Human – object similarity is a factor especially significant considering anthropomor-
phism in interactions with robots and robot- like agents. The classic study by Reeves
and Nass (1996) or Nass & Moon (2000), suggests for example that people automati-
cally attribute human characteristics (e.g. sex and ethnicity) to computers and obtain
the social rules in the contact with it. Those results are explained by the objective
human-object similarities, including e.g. capability to use a natural language in com-
munication. It can be assumed, that chatbots can provoke the same automatic reaction,
not necessarily connected with anthropomorphism on a deeper level. Sustaining some
social conventions in conversation can be an effect of the objective similarity of bot’s
and human behaviour.
On the other hand anthropomorphism can be seen as an effect of the lack of infor-
mation about the object. In such situation the person can easily project invisible hu-
man characteristics into an object even if it is not objectively similar to a human
(Epley, Waytz & Cacioppo, 2007). Such projection can be connected with actual
belief that the object has the given characteristics, and be influenced by the individual
characteristics and motives. In this case individual differences in anthropomorphism
can be easily manifested. The accessibility of the information about an object can be a
factor influencing those differences.
In interactions witch a chatbot it can be yet difficult to separate a motivated kind of
reaction described in the last section from the automatic reaction to the objective simi-
larity to a human. Although it can be possible to do that analysing the reactions to a
chatbot’s inconsistent behaviours. Illogical statements not related to the subject of
conversation, which show the randomness of bot’s behaviour can remain an occasion
to test the interlocutor’s reaction. The existing data suggests, that the machines mani-
festing malfunctions are easily anthropomorphised in the situation of the lack of in-
formation (Waytz, Morewedge, Epley, Monteleone et all, 2010). In that situation the
differences between a group who has the access to the information about an object
and the group who has not could be possibly visible. The aim of the experiment de-
scribed in this paper is to investigate this hypothesis, and analyse the way anthropo-
morphism manifest itself in a conversation with an anthropomorphic chatbot.
2 Methods and procedure
The thirty-two English Philology students took part in the study (age range = [20, 25],
16 men) in exchange for 5$ (equivalents in PLN). All participants were recruited by
announcements in local universities.
Eviebot conversational agent used in the study is a chatbot construed by Existors and
available on-line. It has a form of a female avatar, capable for verbal and non-verbal
expression, who can make a conversation on any subject, by reacting to the words of
interlocutor. In the experiment Eviebot is used for a chat conversation only (the voice
turned off), and the avatar is displayed on the computer screen. The conversations’
transcription is collected for analysis, with the assurance of confidentiality and ano-
nymity of the data.
The study is conducted in university laboratory. During each session only one partici-
pant and experimenter are present in the room. The participants are randomly divided
into two samples, 16 persons (8 men) in each. Sample A is asked to make a 10-minute
random chat conversation with Eviebot conversational agent. Sample B is given the
same instruction but before the conversation is additionally introduced with the in-
formation about Eviebot and is given three paper tasks related to it. The information
(read loud by experimenter) contains the operating principles of a given bot, descrip-
tion of its visual characteristics and skills (and also some information of other bots),
as well as its functions and price of an application. After that information, Sample B
is given the paper tasks to perform (to underline the most important information about
the chatbot, to indicate which information was new for them, and to write which kind
of people could be interested in using Eviebot). The function of the tasks and the in-
formation is to make the specific knowledge about a chatbot potentially available to
the members of Sample B during the conversation.
3 Expected results and conclusions
The collected data are still processed, therefore in this section a brief description of
the process of analysis and the tested hypothesis will be presented. The collected data
are analysed with the use of qualitative data analysis software (QDA Miner). The
investigation will include content analysis, the number of reactions qualified to dis-
tinguished categories will be also considered.
First, chatbot’s behaviours during the conversation will be analysed, as it can directly
influence the interlocutor reactions. This investigation will include specifically bot’s
inconsistent behaviors during the conversation. A chatbot’s behaviors used to be simi-
lar to human’s, and provokes the reactions typical to human-human conversation
automatically. The chatbot inconsistent behavior: e.g. unlogical, random statements,
can break that reactions schema. It is then important to control this factor as possibly
moderating the results. On the other hand, the participants’ reactions on a chatbot’s
inconsistent behaviours will be investigated. The reactions will be divided into
s) reactions suggesting that a participant expect from a bot the same behavior as from
a human, b)reactions connected with understanding of a machine malfunctions. Fore-
going reactions are interpreted as related to anthropomorphism, which is more than
just an act of sustaining some convention in conversation (sustaining a schema of
human-human conversation can be for participants just a funny and comfortable way
to speak with a chatbot, the chatbot malfunctions can provoke more spontaneous reac-
tions). Then, other participants reactions connected with anthropomorphism will be
analysed: attributing human characteristics to a chatbot (emotions, personality traits,
mind attribution) and expressing emotions towards a chatbot. The differences be-
tween Sample A and Sample B, as well as between men and women will be investi-
gated in all of foregoing aspects.
To sum up, the following predictions will be examined:
1) The group informed about an Eviebot characteristics and given the tasks consol-
idating that information will react differently to chatterbot’s inconsistent behaviours
in comparison to non-informed group
2) Attributing human characteristics to Eviebot will appear in a comparable level in
informed and non-informed group
3) The differences between men and women will appear in the level and quality of
anthropomorphic reactions (Eviebot is a female avatar thus it can be more easily an-
thropomorphised by women, see e.g. Eyssel, Kuchenbrandt, Hegel & de Ruiter 2012)
The content analysis will also concentrate on investigating, which reactions are ap-
pearing after confronting chatbot’s inconsistent behavior in all of the sample.
The conclusions of the research could be used to characterize the phenomenon of
anthropomorphism of a chatbot in the level of behavioral reactions. Reacting to an
object which is very similar to a human can be analysed in two levels: the level of
automatic reactions to the its similarity and the level connected with individual expec-
tations towards an object possibly connected with individual factors. That level could
be visible when analysing the reactions to the chatbot’s inconsistent behaviors, and
then the individual differences in anthropomorphism could be then manifested. The
accessibility of information about a chatbot can additionally influence those differ-
ences. The results of the study could then suggest some implications for designing the
context of interaction with a chatbot, signalizing the possible individual differences
and specificity of anthropomorphic reactions to it.
References
1. Demeure, Virginie, Radosław Niewiadomski, and Catherine Pelachaud, ‘How Is Believa-
bility of a Virtual Agent Related to Warmth, Competence, Personification, and Embodi-
ment?’, Presence: Teleoperators and Virtual Environments, 20 (2011), 431–48
2. Epley, Nicholas, Adam Waytz, and John T. Cacioppo, ‘On Seeing Human: A Three-Factor
Theory of Anthropomorphism.’, Psychological Review, 114 (2007), 864–86
3. Eyssel, Friederike, Dieta Kuchenbrandt, Frank Hegel, and Laura de Ruiter, ‘Activating
Elicited Agent Knowledge: How Robot and User Features Shape the Perception of Social
Robots’ (IEEE, 2012), pp. 851–57
4. Nass, Clifford, and Youngme Moon, ‘Machines and Mindlessness: Social Responses to
Computers’, Journal of Social Issues, 56 (2000), 81–103 https://doi.org/10.1111/0022-
4537.00153
5. Reeves, Byron and Clifford Nass, ‘The Media Equation: How People Treat Computers,
Television, and New Media Like Real People and Places.’ (1996), New York: Cambridge
University Press.
6. Ring, Lazlo, Barbara Barry, Kathleen Totzke, and Timothy Bickmore, ‘Addressing Lone-
liness and Isolation in Older Adults: Proactive Affective Agents Provide Better Support’
(IEEE, 2013), pp. 61–66
7. Swartz, Luke. ‘Why people hate the paperclip: labels, appearance, behaviour and social re-
sponses to user interface agents.’ (2003), Master’s thesis, Stanford University.
8. Waytz, Adam, John Cacioppo, and Nicholas Epley, ‘Who Sees Human?: The Stability and
Importance of Individual Differences in Anthropomorphism’, Perspectives on Psychologi-
cal Science, 5 (2010), 219–32
9. Waytz, Adam, Carey K. Morewedge, Nicholas Epley, George Monteleone, Jia-Hong Gao,
and John T. Cacioppo, ‘Making Sense by Making Sentient: Effectance Motivation Increas-
es Anthropomorphism.’, Journal of Personality and Social Psychology, 99 (2010), 410–35