In virtual educational settings, the impact of learner and teacher gender on human-to-human interaction and the persistence of gender stereotypes are of critical interest. In the systematic review of studies on Pedagogical Agents by Armando et al., authors discuss the impact of gender on learners' perception, academic performance, and self-evaluation skills. Findings indicate that male and female agents can improve performance, with female agents efficiently employable to contrast the stereotype threat, e.g., in male-dominated STEM environments. On the other hand, the agents' gender evidently impacts their pedagogical roles, appearance, and interaction patterns. Virtual agents whose gender does not match social stereotypes on context and roles may be less effective in conveying their messages e.g., older and elegant agents are perceived as experts; female agents are more successful in establishing positive relationships with learners. Androgynous systems as a potential solution require further investigation, as they may hinder efforts to avoid gender stereotypes. The review emphasizes the importance of gender choice and the need for further research in this area.

In the field of green economy and, in particular, regarding waste management applications, the review by Nkwo et al. highlights the significance of thoughtful and human-centered design in developing applications that raise awareness of social issues, using the Persuasive System Design (PSD) framework. The study investigates the incorporation and implementation of behavior change strategies and evaluates their effectiveness based on user ratings. The findings reveal that task-assistance strategies are prevalent, while credibility strategies enhance persuasiveness and trust. The impact of dialogue support strategies, feedback and reminder provisions, and social support strategies leveraging social influence across various dimensions, including app focus and waste management activities, correlate with app ratings. Based on the findings, the authors provide design suggestions and guidelines leveraging social influence e.g., sustainable waste management apps, emphasizing user-friendly routines, adaptive features, automated intelligent notifications, and performance tracking.

The three original research papers in this Research Topic (i.e., Thomas et al.; Chen et al.; Wang et al.) present contrasting viewpoints on user experience with digital interactive systems. Two papers analyze user behavior, while the third examines the impact of messages conveyed through such systems.

In Thomas et al., the authors critically review existing approaches to assessing message persuasiveness in different domains. As a result of their analysis, the authors propose and validate a new scale of persuasiveness based on user ratings of items from two domains: healthy eating advice and email security messages.

The other two papers focus on monitoring literacy learners' attention status and users' attitudes toward medical Artificial Intelligence. In Chen et al., the authors introduce a method to assess disengagement among literacy learners during online classes by measuring performance discrepancy between control tests proposed during class and pre-class tests proposed at the very beginning of the class (i.e., when students' attention is expected to be optimal). The authors show a strong correlation between high attention ratings obtained through their method and good performance in post-test reading comprehension.

In Wang et al., the authors examine methods for assessing people's Knowledge, Attitude, and Behavior (KAB) regarding medical AI. In doing so, they compare a person-based approach that stratifies a population's KAB based on individual profiles with the more common variable-based approach relying on isolated self-assessments of each component. This approach highlights the emergence of subtler profiles of interaction among the three components.

Overall, these papers provide valuable insights into understanding user experience, attention, and attitudes in AI interactive systems, offering new scales, assessment methods, and approaches for further exploration.

Since AI systems are increasingly relied upon for decision-making across different domains, limitations and risks associated with certain applications of AI need to be taken into consideration. Nathan and Fourneret and Yvert aim to shed light on critical issues associated with the use of AI systems.

Nathan (2023) focus on the limitations of disembodied AI (dAI) in educational systems, which emerged particularly during the COVID-19 pandemic. Such systems have two significant limitations: they struggle to model people's embodied interactions, as they primarily rely on statistical regularities rather than capturing the nuanced nature of human behavior; and they are often black boxes, lacking transparency and predictability when applied to new domains. The emergence of multimodal learning analytics and data mining (MMLA) exacerbates the issue, as data accessibility and usage are not properly regulated. To mitigate the risks associated with dAI, Nathan proposes an alternative augmented intelligence system that effectively addresses students' needs.

On the other hand, Fourneret and Yvert highlight a more subtle risk associated with using AI systems to aid human decision-making: human desubjectivation. People's increasing reliance on AI system recommendations has led to various forms of digital normativity, where algorithms establish standards that individuals adopt as the norm in their daily lives, a phenomenon that may affect the acquisition and exercise of subjectivity, influencing critical thinking. Relying entirely on AI systems for decision-making promotes human comfort but discourages individuals from challenging or refusing system suggestions due to their perceived infallibility. To address the risk of desubjectivation, Fourneret and Yvert highlight the importance of an Ethics-by-design methodology, involving ways to protect the subjective thinking process during the project's ideation phase rather than at implementation. They emphasize the importance of involving philosophers and ethicists in the development of new technologies and emphasize the need to educate future generations about the risks of silent acceptance of AI governmentality (see also Franzoni, 2023).

One of the significant challenges lies in the transparency and explainability of AI systems. Generalist AI systems often employ sophisticated algorithms and deep neural networks, making it difficult to understand and explain their decision-making processes (see Adadi and Berrada, 2018; Balasubramaniam et al., 2023). The lack of transparency and interpretability raises concerns about discrimination and unfair or unjust outcomes.

As AI systems take on increasingly autonomous decision-making roles, traditional models of responsibility may not adequately capture the new unique challenges posed. Establishing clear frameworks for assigning responsibility and addressing questions of negligence, oversight, and the potential for unintended consequences is essential to ensure accountability for the decisions made by AI systems, capable of making autonomous decisions across various domains without human intervention. Balancing the autonomy of AI systems with human judgment and intervention is necessary to prevent undue reliance on AI decisions and preserve human agency and accountability (see Beckers, 2022; Cavalcante Siebert and Lupetti, 2023).

AI systems can inadvertently perpetuate biases present in the data they are trained on, leading to discriminatory outcomes (see Dwork, 2012; Mehrabi, 2021). Decision-making AI must be designed to recognize and mitigate biases, ensuring fairness in the decision-making process across diverse populations. This issue requires developing techniques that identify and address biases and allow designers to be conscious of their biases and limits. AI systems can have significant societal impacts, influencing resource allocation, access to services, and opportunities. Ensuring these systems are designed and deployed to benefit all individuals is critical to avoid exacerbating existing inequalities (Datta, 2023).

Generalist AI systems rely on vast amounts of data, often including sensitive personal information. Protecting individual privacy and ensuring robust data security measures become paramount to prevent misuse or unauthorized access to personal information. Balancing the benefits of artificial intelligence with privacy considerations is an ongoing challenge, as a huge number of entities are massively and continuously collecting data, virtually beyond the control of individuals (see Song et al., 2022; USA White House Executive Office, 2023).