Adaptive Prompt Intelligence: Towards Self-Evolving Conversational Agents

Conversational Agents (CAs) are characterized by their roles within a narrative and the communication style they adopt during conversations. Within computer games, users' evaluation of the narrative is influenced by their estimation of CAs' intelligence and believability. However, the impact of CAs' roles and communication styles on users' experience remains unclear. This research investigates such influence of CAs' roles and communication styles through a crime-solving textual game. Four different CAs were developed and each of them was assigned to a role of either witness or suspect and to a communication style than can be either aggressive or cooperative. Communication styles were simulated through a Wizard of Oz method. Users' task was to interact, through real-time written exchanges, with the four CAs and then to identify the culprit, assess the certainty of their judgments, and rank the CAs based on their conversational preferences. In addition, users' experience was evaluated using perceptual measures (perceived intelligence and believability scales) and behavioral measures (including analysis of users' input length, input delay, and conversation length). The results revealed that users' evaluation of CAs' intelligence and believability was primarily influenced by CAs' roles. On the other hand, users' conversational behaviors were mainly influenced by CAs' communication styles. CAs' communication styles also significantly determined users' choice of the culprit and conversational preferences.

People with diabetes mellitus not only have to deal with physical health problems, but also with the psycho-social challenges their chronic disease brings. Currently, technological tools that support the psycho-social context of a patient have received little attention. The objective of this work is to determine the feasibility and preliminary efficacy of an automated conversational agent to deliver, to people with diabetes, personalised psycho-education on dealing with (psycho-)social distress related to their chronic illness. In a double-blinded between-subject study, 156 crowd-workers with diabetes received a social help program intervention in three sessions over three weeks. They were randomly assigned to receive support from either an interactive conversational support agent () or a self-help text from the book "Diabetes burnout" as a control condition (). Participants completed the Diabetes Distress Scale (DDS) before and after the intervention, and after the intervention, the Client Satisfaction Questionnaire (CSQ-8), Feeling of Being Heard (FBH), and System Usability Scale (SUS). Results indicate that people using the conversational agent have a larger reduction in diabetes distress (, ) than the control group (, ) and this difference is statistically significant (, ). A hypothesised mediation effect of "attitude to the social help program" was not observed. An automated conversational agent can deliver personalised psycho-education on dealing with (psycho-)social distress to people with diabetes and reduce diabetes distress more than a self-help book. This study has been preregistered with the Open Science Foundation (osf.io/yb6vg) and has been accepted by the Human Research Ethics Committee - Delft University of Technology under application number 1130. The data and analysis script are available: https://surfdrive.surf.nl/files/index.php/s/4xSEHCrAu0HsJ4P.

Discussion forums are among the most common interaction tools offered by MOOCs. Nevertheless, due to the high number of students enrolled and the relatively small number of tutors, it is virtually impossible for instructors to effectively monitor and moderate them. For this reason, teacher-guided instructional scaffolding activities may be very limited, even impossible in such environments. On the other hand, students who seek to clarify concepts may not get the attention they need, and lack of responsiveness often favors abandonment. In order to mitigate these issues, we propose in this work a multi-attribute text categorization tool able to automatically detect useful information from MOOC forum posts including intents, topics covered, sentiment polarity, level of confusion and urgency. Extracted information may be used directly by instructors for moderating and planning their interventions as well as input for conversational software agents able to engage learners in guided, constructive discussions through natural language. The results of an experiment aimed at evaluating the performance of the proposed approach on an existing dataset are also presented, as well as the description of an application scenario that exploits the extracted information within a conversation agents’ framework.

This white paper examines the necessity and viability of integrating Emotional Intelligence (EI) into Conversational AI Agents (CAs) to address the limitations of traditional, sentiment-based customer support systems. Drawing upon affective computing and organizational psychology, the study establishes a theoretical framework for computational empathy, detailing the multimodal architectural requirements for robust emotion recognition across linguistic and acoustic channels.1 Synthesis of recent case studies indicates that advanced engagement strategies, notably Reinforcement Learning (RL), deliver significant performance gains, including a reported +15.2% increase in customer satisfaction (CSAT). Crucially, the deployment of these systems must navigate profound ethical challenges, including inherent algorithmic bias stemming from cultural variability and the critical risks of privacy invasion and emotional manipulation. The paper concludes by proposing a design framework centered on transparency, accountability, and cultural sensitivity, ensuring that EICAs transition the customer service domain from operational efficiency to one of genuine emotional resonance and relational value. Keywords Emotional AI, Affective Computing, Customer Service, Empathy Modeling, Conversational Agents, Human–AI Interaction, Algorithmic Bias, Multimodal Systems.

The purpose of this scoping review was to examine the depiction of conversational agents as health professionals. We identified the professional characteristics that are used with these depictions and determined the prevalence of these characteristics among conversational agents that are used for health care. The depiction of conversational agents as health professionals has implications for both the users and the developers of these programs. For this reason, it is important to know more about these depictions and how they are implemented in practical settings. This review included scholarly literature on conversational agents that are used for health care. It focused on conversational agents designed for patients and health seekers, not health professionals or trainees. Conversational agents that address physical and/or mental health care were considered, as were programs that promote healthy behaviors. This review was conducted in accordance with JBI methodology for scoping reviews. The databases searched included MEDLINE (PubMed), Embase, CINAHL with Full Text (EBSCOhost), Scopus, Web of Science, ACM Guide to Computing Literature (Association for Computing Machinery Digital Library), and IEEE Xplore (IEEE). The main database search was conducted in June 2021, and an updated search was conducted in January 2022. Extracted data included characteristics of the report, basic characteristics of the conversational agent, and professional characteristics of the conversational agent. Extracted data were summarized using descriptive statistics. Results are presented in a narrative summary and accompanying tables. A total of 38 health-related conversational agents were identified across 41 reports. Six of these conversational agents (15.8%) had professional characteristics. Four conversational agents (10.5%) had a professional appearance in which they displayed the clothing and accessories of health professionals and appeared in professional settings. One conversational agent (2.6%) had a professional title (Dr), and 4 conversational agents (10.5%) were described as having professional roles. Professional characteristics were more common among embodied vs disembodied conversational agents. The results of this review show that the depiction of conversational agents as health professionals is not particularly common, although it does occur. More discussion is needed on the potential ethical and legal issues surrounding the depiction of conversational agents as health professionals. Future research should examine the impact of these depictions, as well as people's attitudes toward them, to better inform recommendations for practice.

Same benefits, different communication patterns: Comparing Children's reading with a conversational agent vs. a human partner

Conversational agents (CAs) powered by artificial intelligence are frequently subjected to verbal aggression by users, which can ultimately affect CAs’ business value. However, the underlying reasons for such aggression remain understudied in the literature. This study aimed to investigate the fit effect of power experience over CAs and users’ personal desire for power on verbal aggression. First, our results reveal that power experience over CAs negatively influenced users’ anticipatory guilt regarding verbal aggression toward CAs. Second, there was a fit effect between two types of power: when power experience over CAs and users’ personal desire for power were congruent, users’ anticipatory guilt was lower than when they were incongruent, leading to increased verbal aggression toward CAs. Our study contributes to the literature on verbal aggression toward CAs and can help business practitioners take appropriate measures to mitigate the potential costs incurred by users’ power experience over CAs.

A discussion forum is a valuable tool to support student learning in online contexts. However, interactions in online discussion forums are sparse, leading to other issues such as low engagement and dropping out. Recent educational studies have examined the affordances of conversational agents (CA) powered by artificial intelligence (AI) to automatically support student participation in discussion forums. However, few studies have paid attention to the safety of CAs. This study aimed to address the safety challenges of CAs constructed with educational big data to support learning. Specifically, we proposed a safety‐aware CA model, benchmarked with two state‐of‐the‐art (SOTA) models, to support high school student learning in an online algebra learning platform. We applied automatic text analysis to evaluate the safety and socio‐emotional support levels of CA‐generated and human‐generated texts. A large dataset was used to train and evaluate the CA models, which consisted of all discussion post‐reply pairs ( n = 2,097,139) by 71,918 online math learners from 2015 to 2021. Results show that while SOTA models can generate supportive texts, their safety is compromised. Meanwhile, our proposed model can effectively enhance the safety of generated texts while providing comparable support. Practitioner notes What is already known about this topic Online discussion forums have been plagued by a lack of interaction among students due to factors such as expectations to receive no response and perceptions of topic irrelevance which lead to low motivation to participate. AI‐based conversational agents can automatically support students' interactions in online discussion forums at a large scale, and their generated responses can be human‐like, contextually coherent and socio‐emotionally supportive. Unsafe discourse exchanges between students and conversational agents can be dangerous as identity attacks, aggravation and bullying behaviours embedded in discourses can disrupt students' knowledge inquiry and negatively influence student motivation and engagement. However, few educational studies have paid attention to the safety of conversational agents. What this paper adds This study proposes and synthesized strategies to build AI‐based conversational agents that automatically support online discussions with safe and supportive discourses. This study reveals the relationship between discourse safety and social support, suggesting supportive discourses can also be unsafe. This study enriches the literature on educational conversational agents by synthesizing a conceptual framework on discourse safety and social support, and by proposing concrete algorithmic strategies to improve the safety of conversational agents. Implications for practice and/or policy Researchers and practitioners can adopt strategies in this study such as generation control, open‐sourced models and public API services to evaluate students' discourse safety for early intervention or modify existing conversational agents to be safety‐aware. Practitioners can utilize the proposed conversational agent to automatically support students both safely and socio‐emotionally at a large scale. Practitioners should be cautious when examining social support with automatic analysis, as not all supportive texts are safe. While unsafe texts can provide emotional support, it does not justify their appropriateness in a learning environment.

PurposeConversational agents (chatbots, avatars and robots) are increasingly substituting human employees in service encounters. Their presence offers many potential benefits, but customers are reluctant to engage with them. A possible explanation is that conversational agents do not make optimal use of communicative behaviors that enhance relational outcomes. The purpose of this paper is to identify which human-like communicative behaviors used by conversational agents have positive effects on relational outcomes and which additional behaviors could be investigated in future research.Design/methodology/approachThis paper presents a systematic review of 61 articles that investigated the effects of communicative behaviors used by conversational agents on relational outcomes. A taxonomy is created of all behaviors investigated in these studies, and a research agenda is constructed on the basis of an analysis of their effects and a comparison with the literature on human-to-human service encounters.FindingsThe communicative behaviors can be classified along two dimensions: modality (verbal, nonverbal, appearance) and footing (similarity, responsiveness). Regarding the research agenda, it is noteworthy that some categories of behaviors show mixed results and some behaviors that are effective in human-to-human interactions have not yet been investigated in conversational agents.Practical implicationsBy identifying potentially effective communicative behaviors in conversational agents, this study assists managers in optimizing encounters between conversational agents and customers.Originality/valueThis is the first study that develops a taxonomy of communicative behaviors in conversational agents and uses it to identify avenues for future research.

This chapter describes Conversational Agents (CAs) in the context of Systems Engineering. A CA is a computer program which interacts with a user through natural language dialogue and provides some form of service. CA technology has two points of interest to systems engineers: the use of systems engineering techniques in CA research and the application of CAs in project development. CAs offer the opportunity to automate more complex applications than are feasible with conventional web interfaces. Currently such applications require a human expert in the domain to mediate between the user and the application. The CA effectively replaces the human expert. This chapter reviews the current capabilities of various CA technologies, outlines a development methodology for systems engineering practitioners interested in developing real world applications and suggests a number of directions for systems engineers who wish to participate in CA research.

We examined the potential of conversational agents (CAs) to support older adults' self-care related to chronic illness in light of lessons learned from decades of pedagogical agent research, which investigates the impact and efficacy of CAs for a wide range of learners. The role of CAs in education (i.e., pedagogical agents) has been long studied, but their potential for supporting self-care has received less attention, especially for older adults. We reviewed work on pedagogical agents and considered how it informs the design of CAs for older adults. We propose a framework for designing CAs to support older adult self-care, which organizes a review of work in this area and integration with the pedagogical agent literature. Our review of the pedagogical agent literature revealed an evolution from teaching machines to interactive, social systems that influence student motivational as well as learning outcomes. To integrate this review with work on CAs and self-care, we developed a framework that specifies how self-care goals evolve with stages of an illness, communication goals that support self-care at each stage, patient needs, and requirements for CAs to support these needs. The review identified an agenda for future research on CA functions and features that help older adults accept need for self-care, establish self-care, and sustain self-care over time. Integrating insights from the pedagogical agent literature with research on developing CAs for self-care defines an agenda for developing and evaluating CAs to help older adults manage illness.

Is cute AI more forgivable? The impact of informal language styles and relationship norms of conversational agents on service recovery

With the rise of smart devices in the 21st century, children are increasingly engaged in socially contingent interactions with conversational agents such as Alexa, Google Assistant, and Siri. Using an online parent survey, young children's verbal interactions (parasocial interactions) and emotional relationships (parasocial relationships) with conversational agents were examined in a naturalistic study. A total of 92 parents responded to the survey, 70 of whom qualified because they had a conversational agent. Exploratory and confirmatory factor analyses of emotional parasocial relationships with conversational agents in a subset of this sample (n = 58, mchild age = 5.54 years, age range 3–10, 33 females) revealed three dimensions: attachment, personification, and social realism. These dimensions are consistent with children's parasocial relationships with media characters. The relation between parasocial verbal interactions and emotional parasocial relationships with conversational agents was bidirectional. The results indicate that children develop close emotional ties with artificial beings, treating them as human-like entities with feelings and for whom they have feelings. Implications for interactions with artificial life as children's trusted social partners are considered.

Conversational agents have been established to support educational practices, solving students’ questions, recommending teaching materials, increasing the motivation of students who have little interest in certain content, among others. Although many studies are discussing the establishing of conversational agents, there is no global understanding of what has been investigated in this area. There are some previous secondary studies that have attempted to map the conversational agents used for educational purposes (i.e., pedagogical conversational agents). However, these studies are limited to include primary studies published in specific sets of conferences and journals. Therefore, an overview of the current state of the art associated with pedagogical conversational agents has not been produced yet. This research paper aims to contribute to the theme, through the presentation of planning, conduction, and results analysis of a systematic mapping on pedagogical conversational agents. We address the following questions: (i) in which areas have conversation agents been investigated?; (ii) at which education levels are conversational agents used?; (iii) which mechanisms do conversational agents use to interact with students?. We also present some discussions about research opportunities that can be explored in future works. This study aims to contribute to the topic of pedagogical conversational agents, helping researchers to have a deeper understanding of what has already been done and to guide their research to subjects not explored yet.

Conversational agents (CAs)—frequently operationalized as chatbots—are computer systems that leverage natural language processing to engage in conversations with human users. CAs are often operationalized as chatbots which are used for many applications including technical support, customer service, and digital personal assistants. Despite their widespread use, little research to date has investigated how improving the conversational skill of a CA impacts user perceptions of the agent. To elucidate this relationship, this research uses Social Presence Theory to describe how conversational skill influences perceived social presence and ultimately anthropomorphism of a chatbot. We conducted a series of studies in which 450 participants interacted with CAs exhibiting varying levels of conversational skill. We show that people perceive a more skilled CA to be more socially present and anthropomorphic than a less skilled CA. This research advances the knowledge of computer-human interface in information systems, as CA research to date has largely focused on the technical challenges rather than the behavioral questions of how users interact with CAs.