Title: Pedagogical Conversational Agent PCA

Project duration: 01.05.2024 – 30.10.2025

Research area: early childhood education/development, Social-emotional learning, AI

A pedagogical conversational agent (PCA) is a digital system that aims to support and scaffold students’ learning through flexible, realistic, and intelligent dialogue. It can serve as a “digital teaching assistant” that enhances the learning process for students and alleviates the instructional burden for educators. For this project, we wish to validate a PCA capable of delivering a social-emotional training to 3- to 5-year-old children. The PCA, represented as an animated monkey named “Uku”, will leverage natural language processing and artificial intelligence techniques to guide children through exercises in perspective-taking, emotion regulation, empathy, and prosocial behavior.
In the first phase, human experts in clinical-developmental psychology will deliver the training to children via the agent functioning as a digital puppet. A micro-language model underlying the PCA will store and continuously learn from the content of these human-led interactions. By teaching our PCA with an aggregated dataset of real conversations involving experts and children, we aim to overcome the rigidness and lack of contextualization that plague existing PCAs in the educational technology space. In a second phase, the PCA will use the language model’s framework to facilitate the training on its own. Ultimately, we aim to evaluate our PCA’s capability to match or exceed human-led social-emotional learning programs.

Problems and Aims

The global education sector faces a critical challenge as the shortage of teachers is leading to large classroom sizes resulting in insufficient individualized attention for students. The educator’s ability to meet the student’s unique learning needs is thereby hampered which ultimately affects the academic and educational outcomes on the student’s side. To address this issue the ScaDS.AI Dresden/Leipzig, in collaboration with the Humboldt Science Center for Child Development (HumanKind) at Leipzig University is developing an advanced Pedagogical Conversational Agent (PCA). This Agent aims to support educators by engaging students in intelligent, contextualized dialogue. It thereby reduces the instructional burden of educators and enhancing the individual student’s learning experience.

Practical Example

In a typical classroom setting, teachers often struggle to provide personal and individualized attention to each of their students due to the previously mentioned classroom sizes. By serving as a digital teaching assistant, our Pedagogical Conversational Agent (PCA) addresses this challenge. We envision the agent to be able to interact with students on an individual basis, small groups or even with the entire class. As a first step, for the purpose of validating our system,  the PCA – embodied as a digital puppet named ‘Uku’ will be trained to deliver social-emotional trainings (SEL) so that it can mediate and facilitate conversation or discussion, provide feedback and guide students through the envisioned activities. This capability is not confined to just in-person settings, but rather can also operate seamlessly over video conferencing platforms. Thereby ensuring consistent support in both physical and virtual learning environments.

Technology

The PCA system leverages advanced natural language processing (NLP) and Artificial Intelligence (AI) so that it can learn how to facilitate meaningful, contextualized conversation with students. The hybrid system will allow educators to control the agent as a digital puppet in real-time, providing a personalized touch while the PCA learns from said interactions to enhance its autonomous capabilities. By aggregating data from real, personal educator-student dialogues, the PCA can dynamically update its knowledge base. This ensures that it remains contextually relevant and effective in delivering educational content. This ongoing learning process enables the PCA to support and even replicate human-led instruction, aiming to match (or exceed) human interventions in terms of student satisfaction and task completion rates.

Outlook

The potential impact of the Pedagogical Conversational Agent extends beyond immediate educational support. With the global Education Market projected to exceed 7 trillion USD by 2025, and a significant portion allocated to educational technology, the PCA is poised to play a substantial role in the future of education and social-emotional learning. The system could be adopted by schools, cultural centers, and healthcare institutions, providing personalized educational experiences and procedural guidance. Furthermore, as the technology evolves, we envision a possible expansion of the application to other areas such as consumer engagement and procedural support in various industries. The ultimate goal is to advance the PCA from a hybrid, educator-assisted system to a fully autonomous digital assistant, providing scalable and high-quality educational support independent of location.

By addressing the teacher shortage and enhancing student engagement through innovative AI technology, the PCA project aims to not only transform the educational landscape, ensuring each and every student receiving the attention and support they need to succeed academically. It also provids teachers with comprehensive support and relief. We foresee the latter to enable educators to focus more on the individual needs of their students and better fulfill the pedagogical aspects of their roles, hence, empowering their practice.

Team

Lead

Eric Roldan Roa

Team members

• Prof. Dr. Sayan Mukherjee
• Prof. Dr. Tina Malti
• Ph. D. Tyler Colasante
• Doris Kristina Raave
• Juan Carlos Ramos Martínez
• Hongtao Li
• Johanna Eller
• Kuzma Kagadeev
• Sina Gibhardt
• Carolina Santos
• Aleyna Gezginci