Related Literature Group 4, design

Excerpts & citations

Gender

• Females chose to gender-match and to interact with a more realistic VA. Males exhibited little preference for either gender, and a greater preference than females for realistic VAs. Thus, where it is not feasible to gender-match in SSCO, the recommendation is to implement a realistic female VA.

• Young and attractive female agent positively impacts interest in learning, an older an unattractive male agent does not impact motivation.

• As research suggests that the combination of an agent’s gender and personality can play an important role in user perceptions and expectations [1, 37], employing a male agent instead may result in some significant differences in user perceptions or ratings of the agent. We encourage future work to investigate how gender and personality of a workplace productivity agent might influence user experience.

• participants prefer same-gender agents when they are asked to choose their preferred agent as presenter for a multimedia slideshow.

• We found that the designed characteristics of VAs affects some aspects of user impressions (i.e. personality and trustworthiness) of the VA, while other impressions are not affected (i.e. social ability). We also found that gender matching between the agent and the user affect user impressions. // Gender similarity --> more trustworthy

• In addition, the similarity of an agent to the learner positively influences the learner’s motivation (Bailenson, Blascovich, & Guadagno, 2008 in a study with undergraduates). For example, computer-based female agents yielded better motivational outcomes for undergraduate women if they matched the students with respect to race and gender (Rosenberg-Kima, Plant, Doerr, & Baylor, 2010). Another study, conducted with undergraduates by Rosenberg-Kima, Baylor, Plant, and Doerr (2008) revealed that a female agent rated as young, attractive and “cool” succeeded in enhancing young female students’ self-efficacy, which is believed to be a driving force behind motivation (Bandura, 1997). All these findings are theoretically supported by Bandura’s social cognitive learning theory which states that people often learn behavior and norms by imitating people whom they perceive as similar (or superior: higher in rank or status) to them and who are therefore rather accepted as social role models (Bandura, 1986). This finding is supported by another study of Gulz, Haake, and Tärning (2007) which demonstrated that participants prefer same-gender agents when they are asked to choose their preferred agent as presenter for a multimedia slideshow.

Appearance and human-likeness

• In this study, it was noted that PIAs’ human-like features could influence the manner in which participants (i.e. males or females) related to a particular PIA and participants’ preference level of a particular PIA. The result of the Fischer’s exact test conducted in this study (Table V) found no statistically significant effect between participants’ gender (male or female) and their preferred gender of PIAs. It is concluded that the gender of participants (i.e. male or female) has no role (impact) on their preferences regarding the type, features, or gender of PIA, neither on their preference level of PIA.

• On the other hand, previous studies have also shown that developing overly humanized agents results in high expectations and uncanny feelings.

• Many participants saw the human-like appearance of the VA prototype as setting the wrong expectations in terms of its capabilities, and they were disappointed when the agent’s intelligence only extended towards responding to the their emotion and prompting more self-reflection. (…) We believe incorporating human-like qualities and emotional intelligence into future agents to be worthwhile; however, intelligence should also extend into other aspects of the agent’s capabilities in order to better help users be as efficient as possible in achieving their work goals.

• A study by Go and Sundar states that revealing the identity of a chatbot as a non-human can have a positive effect: user will have less high expectations about the conversation, and will be impressed when an agent shows human-like behaviour. Furthermore, they emphasize the importance of the conversational style between a human an a computer. When the dialogue resembles that of an actual human, perceived feelings of social presence and homophily will increase, leading to more positive attitudes towards the agent (and in turn potential desired behaviour consequences).

• the presence of a representation produced more positive social interactions than not having a representation
  • human-like representations with higher realism produced more positive social interactions than representations with lower realism; however, this effect was only found when subjective measures were used. Behavioral measures did not reveal a significant difference between representations of low and high realism.
    • the difference we found may also be driven by demand characteristics. Participants interacting with an animated character (as opposed to a photograph) may suppose that the researcher is expecting a high appraisal.
  • while the presence of a face is better than no face at all, the quality of the face matters much less.
    • it is quite possible that animating highly realistic faces inherently allows for residual attributes of the faces that are negative—for example making 3D human faces may produce gestures and animations that appear unnatural or disturbing
  • while most studies have found that interface agents have positive effects on task performance, these effects are overall actually quite small.

• In addition to understanding human social behavior around computers, another extensive line of work examines humans’ responses to computers with more expressive, human-like qualities, such as faces and facial expressions. In general, these studies have found that anthropomorphic properties of computers influence users’ perceptions (e.g., [13, 70, 86]), attitude (e.g., [13]), and behavior around such systems (e.g., [48, 86, 105]). For example, Sproull et al. [86] report that people respond to a text-based interface differently than to a talking face. On the one hand, users are aroused more and present themselves more positively when interacting with a talking face. On the other hand, users are less relaxed or assured when interacting with a talking face. Another study shows that users find the interface with the anthropomorphic qualities—faces and facial expressions—more likeable and engaging, although such an interface takes the users’ effort to interpret the meaning of the human-like expressions and may even be a distraction [48, 86]. More recent studies show that human-like features with higher realism elicit more positive social interactions while having no significant impact on user task performance [105]. Furthermore, a study reveals that anthropomorphism may even elicit user objections due to users’ own biases (e.g., sexism) [70].

• Physically present agent leads to better motivational outcomes than a voice or text-box. It is important to design agents realistically, because e.g. cartoon figures have been shown to diminish the positive motivating effects in comparison to realistic figures.

• It was found also that attractive VT did not improve perceived value of exercise and perceived risks of health. This result is different from some prior studies (e.g. Shiban et al., 2015) which find that attractive virtual pedagogical agents are effective for engaging students in learning. There are two possible explanations for this observed difference. First, attractive virtual agents may catch the attention of users and make users more confident in using a VTS. However, they may also distract users from core learning materials (Moreno and Flowerday, 2006). Second, attractive agents are closer to friends rather than experts. The incongruence of expert perception may lead to contextual irrelevance which makes users take health-related information from virtual trainers lightly (Veletsianos, 2010).

• If McCloud’s (1993) framework is applied, a teacher character, representing the other to a higher extent than a learning companion, might benefit from more realism in the representation. A learning companion character, being to a higher extent conceived of as an extension of oneself, may, on the other hand, benefit from a more iconic representation.

• The findings of this experiment suggest a new design method for human-agent collaboration work. If we use an agent having a non-human-like appearance (for example, the robot-like agent), the user seemed to attribute much responsibility to the agent and not to trust the agent. We suggest using the agent having human-like appearance in human-agent collaboration work from this experiment. A question is raised about the commonly accepted beliefs about agent design for social tasks. We should consider the attribution of responsibility to construct trustworthy agents.

Behaviour and personality

Behaviour

• It is possible to program a chatbot in a way that it will interact with you via a moving and talking avatar. The paper also suggests that displaying facial expressions are beneficial in the interaction with such an agent, since displaying the avatar’s emotion can enhance perceived emotional intelligence.

• Looije et al researched the guidelines that are needed when developing a personal assistant. These guidelines were derived from interviewing, persuasive technologies and from existing guidelines for personal assistants. In their research they found that guidelines were best expressed in iCat (a personal assistant) that was able to show socially intelligent behavior compared to a non-social or text interface based iCat.

• “The results suggested that it is possible to express extroversion and confidence in the agent by changing the agent's gaze amount regardless of the agent's embodiment.

• Social fidelity is related to certain human-like behaviour and cues. This includes for example speech content (personalized language, feedback, politeness, social memory, personality) and visual cues (facial expressions, gestures, gaze, emotions).

• An expressive emotionally intelligent VA is perceived as more emotionally intelligent when expressing itself through words, tone, body language, and facial expressions.

• “Expressions are no more defined by a static representation; rather they are constituted as a succession of signals that appear dynamically. Using few expressions limits the interaction. Endowing agents with large variety of expressions ensures more naturalness in the agent’s behaviour.” Examples are given in this paper as well, see for instance figure 5 or 6.

• Communication features like nonverbal cues seem to be crucial in maintaining learning motivation in virtual learning environments, probably because they inform the observer about states, involvement, responsiveness, and understanding.

• Deictic gestures (pointing towards something, indicating directions, objects etc) have been shown to guide attention, especially when the agent is static.

• A paper by Grover et al. described an experiment in which two chatbots were compared, one of which has a face and appeared to be more emotionally intelligent. Results suggested that the emotionally expressive agent can lead participants (average age 33) to be more productive and focused during working hours. The participants also reported to feel more satisfied with their achievements.

• A study by Go and Sundar states that revealing the identity of a chatbot as a non-human can have a positive effect: user will have less high expectations about the conversation, and will be impressed when an agent shows human-like behaviour. Furthermore, they emphasize the importance of the conversational style between a human an a computer. When the dialogue resembles that of an actual human, perceived feelings of social presence and homophily will increase, leading to more positive attitudes towards the agent (and in turn potential desired behaviour consequences).

Personality

• Higher feelings of social presence can be achieved when an agent’s language usage shows a consistent personality, which can be either introvert or extravert.

• As we pointed out in Section 2, education and customer services are the task-oriented domains most reported in the literature. We found conscientiousness, damage control, thoroughness, manners, emotional intelligence, and identity in studies for both domains. However, manners and emotional intelligence have a different goal in these domains. In the education context, these characteristics are designed to encourage students, especially in a situation of failure, in which the chatbot should be comforting and sensitive. This function aligns with other domains, such as health-care. The education domain also reports needs for personality, so the chatbot can be recognized as either an instructor or a student, and proactivity, so the chatbot can motivate students to participate in the interactions. Personality is also reported in other domains in which the chatbots’ character influences the interactions, such as gaming and humorous talk. Proactivity is also consistently reported in domains in which the chatbot provides guidance, such as coaching, health, ethnography, and assessment interviews.

Language

• Virtual agents which are communicating in a personalized way (using “I” and “you”) will behave more human-like and it will therefore gain more social fidelity. It will also lead to increased feelings of social presence and better learning performance and motivation.

• The previous statement is supported by Araujo . In his paper, he showed that social presence increased when the machine shows a more intelligent interaction style.

• Elaborative feedback and polite conversation has been shown to have a positive influence on performance. Furthermore, compliments (for correct answers) can encourage intrinsic motivation by positively influencing feelings of competence, self-control, self-efficacy and curiosity.

• As Emoji use in text has been shown to strengthen the perceived affect of a message (either in a positive or negative direction) compared to the same text without accompanying emojis [32], we saw incorporating emojis into the VA prototype as an easy first step towards introducing more emotional expressiveness.

• Recent research also suggests that the psychological benefits of disclosure and reflection with an agent are similar to reflection with another human [13]. Therefore, we considered the ability for users to reflect on their feelings and sense of productivity to be a beneficial final extra feature in the VA prototype. After users reported how they were feeling during the morning dialogue, they were asked to reflect upon their feelings in an open-ended response.

• The study concludes that the effects of flattery from a computer can produce the same general effects as flattery from humans, as described in the psychology literature. These findings may suggest significant implications for the design of interactive technologies.

References