Learning theories in pedagogical agent research: A two-phased systematic review

The use of technology in education has advanced significantly over the past 25 years, including pedagogical agents. Pedagogical agents are on-screen characters designed to facilitate learning. However, it is recognized that research of learning theories in the context of pedagogical agents is still inadequate and rarely discussed. The main aim of this study was to gain an understanding of the state of current research on learning theories with pedagogical agents and to identify and synthesize the utilization and trends. Another aim was to identify the types and technologies of pedagogical agents in recent research. The study was carried out as a two-phase systematic review, using four databases: Scopus, ScienceDirect, Taylor & Francis, and ERIC. The first phase, an umbrella review of previous reviews and meta-analyses from 2013-2023, included a total of eight studies presenting 17 different learning theories or related principles and hypotheses. The second phase, the systematic literature review of recent empirical studies from 2021-2023, identified a total of nine studies, presenting nine different learning theories or related principles and hypotheses. All included studies incorporated the educational context and the use of technology in relation to pedagogical agents and the implementation of one or more learning theories. The systematic approach was carried out using Fink's seven-step model and the PRISMA guidelines for documentation. The learning theories found were analyzed using the framework of Illeris’ three dimensions of learning: content (cognitive perspective), interaction (social perspective), and incentive (emotional perspective). All three dimensions were present in the study’s results. Often the learning theory combined two or all the dimensions. The Cognitive load theory and the Cognitive theory of multimedia learning were identified as central pillars within pedagogical agent research, giving rise to numerous smaller theories, hypotheses, and principles. The utilization of learning theories in the studies varied and the results of the support found were often mixed. The identification of types and technologies of pedagogical agents in the studies revealed that they were predominantly utilized as character agents, lacking chatbots or software components. The embodiment of the agents varied, with some being more human-like than others. Although some agents were able to prompt and provide feedback based on user input, interactions were predefined and lacked generative artificial intelligence (AI) such as the use of large language models. Recommendations for future research to address the gaps identified in this systematic review are discussed. Learning theories and related research should be considered in the design and use of pedagogical agents.