The Past, Presence, and Future of Intelligent Pedagogical Agents

• In Event: The Past, Present, and Future of Intelligent Pedagogical Agents: From Avatars to Human Digital Twins

Abstract

Objectives & Background: For a number of decades, pedagogical agents have played a central role in computer-based learning environments (CBLEs; Johnson, et al, 2000). Of particular importance to us is the role of pedagogical agents in open-ended learning environments (OELEs). OELEs adopt a constructivist epistemology to support the acquisition of domain knowledge along with critical thinking and problem-solving skills (Biswas, et al, 2016). Students working in these environments have a specified learning goal, e.g., construct a model of a scientific process. Pedagogical agents are frequently integrated into OELEs to provide cognitive and metacognitive support to learners, especially when they face difficulties in understanding concepts and progressing in their problem-solving tasks (Graesser, et al, 2004). In contrast to providing one-to-one tutoring and mentoring in traditional intelligent tutoring systems, the importance of the socio-cognitive theories of learning (e.g., Vygotsky, 1978) have led to considering multiple roles that agents can play in OELEs, from mentors to various forms of peers, and finally teachable agents.

Methods: A noteworthy role for pedagogical agents in OELEs has been the development of teachable agents (TAs), in particular our work on Betty’s Brain that has evolved over the last twenty-five years (Biswas, et al, 2016; Leelawong & Biswas, 2008). TAs simulate behavior of a person’s thoughts about a system or process (Biswas, et al, 2005). The goal of learning is often to simulate an expert’s reasoning processes about a domain, not the domain itself (e.g., an environment that represents a simulation of a scientific process). Learning empirical facts is important, but learning to think critically and develop reasoning and problem solving skills based on those facts more truly reflect expert thinking and behaviors. Therefore, we designed TAs to simulate forms of thought, e.g., developing and reasoning with causal maps of scientific processes, to help students structure their thinking (Schwartz, et al, 2009). Our previous research has demonstrated that that students learn better when they are asked to teach, and the process of organizing their knowledge structures, developing explicit reasoning mechanisms, and the need for monitoring, checking, and correcting their evolving solutions as they interact with their TAs has led students to develop strong metacognitive and self-regulation skills (Azevedo, et al, 2010; Chase, et al, 2009; Kinnebrew, et al, 2014; Matsuda, et al, 2020).

Scholarly Significance: With advances in computing, artificial intelligence and machine learning technologies, and multimodal sensing and analysis methods, there are several promising directions for intelligent pedagogical agents such as the ability to holistically understand learner behaviors and performance. Challenges include extension of personalization and adaptivity longitudinally so agents can play multiple roles in student learning from mentors to companions to peers, and to even more expressive forms of teachable agents. Natural questions that arise are: (1) can we extend agent technologies to provide life-long partnership and companionship to learners; and (2) what role intelligent pedagogical agents can play in environments, such as classrooms, where groups of students learn together, sometimes individually and sometimes in small groups in the presence of a human teacher.

Author

• Gautam Biswas, Vanderbilt University