The introduction of new technologies has reshaped both workplaces and learning environments, requiring the development of transversal competences such as adaptability, reflection, and perseverance. Following the COVID-19 pandemic, playful approaches have been increasingly adopted in educational settings to enhance engagement, attention, and motivation, as well as to foster awareness of specific topics. In collaboration with the Cologne GameLab, a game prototype entitled Intergalactic Growth was developed to improve understanding of the growth mindset and to promote perseverance, self-awareness, and critical behavior in the use of ChatGPT. This study has two main objectives: to address the growth mindset from an educational and neuroscientific perspective, and to collect qualitative insights on the prototype’s usability and pedagogical potential. Data were gathered through a focus group and analyzed using reflexive thematic analysis (RTA) with NVivo 14. The findings suggest that Intergalactic Growth effectively stimulates emotional engagement and reflection but requires refinement to enhance its pedagogical impact, particularly concerning feedback authenticity and the integration of ChatGPT.

Research in cognitive psychology has long shown that human judgment and decision-making are systematically influenced by cognitive biases, particularly under conditions of uncertainty (Kahneman, 2011). As artificial intelligence (AI) systems become increasingly embedded in socially relevant decision-making contexts -such as healthcare, education, communication, and governance - these cognitive dynamics do not disappear. Instead, they may be reproduced, amplified, or transformed through human-AI interaction (Parasuraman & Riley, 1997; Glikson & Woolley, 2020; Rastogi et al., 2020). Despite the rapid growth of research on cognitive biases in AI-related contexts, existing reviews remain fragmented, often focusing on single biases or specific application domains. To date, no umbrella review has systematically synthesized how cognitive biases are conceptualized, categorized, and addressed across secondary literature on human–AI interaction. This umbrella review aimed at mapping cognitive bias types, application domains, psychological outcomes, and mitigation strategies discussed in existing reviews. The categorization of cognitive biases adopted in this study follows the framework proposed by Adepoju and Adepoju (2025), which distinguishes biases across data-level, algorithmic-level, and human–interaction-level processes. From a cognitive systems perspective, these biases can be understood as emergent properties of coupled human-AI systems rather than as isolated human or algorithmic errors. The findings point to a conceptually rich but methodologically uneven field, characterized by a predominance of interaction-level analyses and a limited empirical evaluation of debiasing strategies. By offering a meta-level synthesis, this study contributes to a more psychologically informed understanding of human–AI interaction and supports the development of human-AI cognitive systems.

Research in cognitive psychology has long shown that human judgment and decision making are systematically influenced by cognitive biases, particularly under conditions of uncertainty. As artificial intelligence (AI) systems become increasingly embedded in socially relevant decision making contexts—such as healthcare, education, communication, and governance—these cognitive dynamics do not disappear. Instead, they may be reproduced, amplified, or transformed through human–AI interaction. Despite the rapid growth of research on cognitive biases in AI-related contexts, existing reviews remain fragmented and heterogeneous, often focusing on single biases or specific application domains. This article reports a structured conceptual mapping review of review-level publications examining how cognitive biases are conceptualized, categorized, and addressed in human–AI interaction. A bounded search of Scopus, Web of Science, and PubMed identified ten eligible review publications published between 2020 and 2025. Using a three-level framework, the mapping distinguishes data-level, algorithmic-level, and human-interaction-level processes. Within the retrieved corpus, the analysis reveals heterogeneous terminology, a predominance of healthcare applications and interaction-level analyses, and limited review-level synthesis of empirical evaluations of mitigation strategies. Drawing on distributed cognition and joint cognitive systems, we propose coupled human–AI cognition as a theoretical lens rather than as an empirically established conclusion. The study provides a structured conceptual map, identifies terminological and methodological gaps, and develops propositions for future empirical research.