Multi-Agent Systems for IEP Development
Dr. Ling Zhang
About Ling
Dr. Ling Zhang is an assistant professor at the University of Wyoming.
Her research centers on designing, implementing, and evaluating technology-enhanced personalized learning for students with and without disabilities in diverse educational settings. Currently, she is part of a team that investigates the design and development of various large language models (LLMs) based multi-agent systems that have the potential to facilitate PL experiences for students with and without disabilities as well as personalized professional learning experiences for educators.
The problem highlighted in this brief
This brief highlights how AI systems, such as multi-agent systems, can support pre-service and in-service teachers in developing Individualized Education Programs (IEPs).
Why does this topic matter to teacher preparation?
Individualized Education Programs (IEPs) are legally binding documents that ensure students with disabilities ensure students with disabilities have access to a free appropriate public education. However, developing an IEP requires a significant amount of time, which can be overwhelming for special education teachers. Multi-agent systems provide special education teachers, especially those in their early stage, reliable and collaborative support in developing IEPs.
About This Brief
Artificial intelligence (AI) can support pre-service and in-service teachers with the complex process of IEP development (Goldman et al., 2024; Mosher et al., 2024). As a rapidly evolving subfield of AI, multi-agent systems present a promising tool to support such a complex task by distributing responsibilities to multiple LLM-based agents with distinct roles and specialized functions (Guo et al., 2024). To that end, a multi-agent system, wherein multiple LLM-based agents are equipped with specialized skills (e.g., data analysis) or knowledge bases (e.g., local databases containing content standards or evidence-based practices), can be utilized to support educators in developing different components of an IEP document. This brief will introduce a multi-agent system, CoIEP, that enhances AI-human collaboration in developing IEPs. This system was designed and developed by Dr. Ling Zhang in collaboration with her colleagues, including Drs. Haidee Jackson, Sohyun Yang, Xuqin Qian, Jennifer Diliberto, Richard Carter, and Jihong Zhang, from multiple institutions.
Research and Practice Context
Using Multi-Agent Systems for Developing IEP Goals
Integrating AI to reduce teacher workload has continually intrigued researchers and educators. However, researchers, practitioners, and policy-makers are concerned about the challenges of generative AI systems such as ChatGPT and Gemini. These involve hallucinations, bias, and weakness in task specificity.
The following are key insights shared by Dr. Ling Zhang on this research. The interview focused on six questions about multi-agent systems for IEP development in teacher preparation and recommendations for teacher educators to incorporate these ideas.
Conversation with Ling
Q1: Can you briefly introduce what multi-agent systems are?
Multi-agent systems involve several AI agents working together to complete complex tasks more effectively than a single-agent system could. Each agent can specialize in a specific function, such as data analysis, calculations, or retrieving information from databases or websites. This collaborative setup enhances both the accuracy and efficiency of task performance. Multi-agent systems hold significant potential for education, particularly in supporting collaborative efforts in special education.
Dr. Zhang: “Multi-agent systems refer to advanced AI systems where multiple AI agents, supported by large language models like OpenAI’s GPT models, collaborate to perform certain parts or subtasks of a complex task. ... One of the advantages of such systems relies on those agents' capacity to perform subtasks in a more accurate and efficient way.”
Q2: What issues are you trying to address through your research and work with multi-agent systems for IEP development?
Dr. Zhang addressed two main issues through the research on multi-agent systems for IEP development: collaboration and content quality. First, multi-agent systems enable a collaborative approach that reflects the interdisciplinary efforts required in education. These systems bring together expertise from different domains, facilitating the generation of high-quality educational content.
The second issue concerns rigorous quality standards for AI-generated content, which could help reduce hallucination—an AI phenomenon in which fabricated or inaccurate information is produced. This focus is critical in special education, where reliable, high-quality resources directly impact student outcomes.
Dr. Zhang: “I was intrigued by how multi-agent systems can work collaboratively across different domains to support educational efforts, reduce workload for educators, and ensure high-quality content.”
Q3: How do multi-agent systems address those issues and support special education teachers?
Multi-agent systems support special education teachers by breaking down the complex process of IEP development into manageable tasks performed by different agent teams. A critical feature of this system is the "human-in-the-loop," where educators remain involved throughout the process by providing feedback and refining outputs. This iterative interaction helps educators develop key IEP writing and evaluation skills, fostering professional learning. Furthermore, leveraging local databases such as state education standards ensures that IEP goals align with state standards and enhances instructional quality by providing data-driven recommendations.
Dr. Zhang: “We use this sequential order of IEP development to design the workflow of the agent... It consists of 3 major teams: the Present Levels of Academic Achievement and Functional Performance (PLAAFP) writer, the PLAAFP evaluator, the IEP goal creation and evaluation team, and the individualized services and support (ISS) generator and evaluator... we also have this ‘keep human in the loop’ function, which is very important for positioning this tool as a professional learning tool.”
Q4: How would you integrate multi-agent systems into a teacher preparation program?
Dr. Zhang emphasizes that multi-agent systems can be effectively integrated into teacher preparation programs by embedding them within existing courses, such as data analysis and IEP development. Furthermore, collaboration between faculty members and special education directors at the state level ensures that these tools align with the professional learning needs of in-service teachers, particularly novice educators.
Dr. Zhang: “We are thinking about collaborating with the instructor to test out how pre-service teachers from that course [focused on data analysis and IEP development] interact with the system, and whether the system can enhance their learning outcomes in terms of IEP development.”
Q5: How can we better prepare educators for using or developing these technologies?
Dr. Zhang emphasizes the need for educators to critically evaluate the quality of AI-generated content, especially to ensure students with disabilities receive meaningful and high-quality learning experiences. Furthermore, she discusses the exciting potential of creating accessible authoring tools, enabling educators to build their multi-agent systems. This future direction aims to personalize professional development by allowing teachers to design agent systems tailored to their unique instructional needs and goals, fostering more profound engagement with the technology.
Dr. Zhang: “One thing I think, and lots of people have already talked about, is how we can provide opportunities for those educators to be more critical and reflective of the content generated by the system because we want to ensure the content is of high quality... Another thing, as you mentioned, is supporting educators' potential in developing these tools.”
Q6: What implications do you see for future research, and what are some questions we might be asking?
One key research question is how to develop automated progress monitoring tools that streamline the data collection process and generate high-quality PLAAFP statements. Additionally, she suggests exploring ways to refine SDI suggestions, assessing whether more detailed recommendations are required to support teachers effectively in lesson planning and classroom instruction. Zhang also raises the challenge of balancing the computational power needed for complex multi-agent systems with financial and operational costs, signaling an important area for future research on the scalability and sustainability of these technologies.
The manuscript of her research can be found here. If you are interested in trying out multi-agent systems developed by Dr. Zhang and her colleagues, please reach out to her via email.
References
Goldman, S. R., Taylor, J., Carreon, A., & Smith, S. J. (2024). Using AI to Support Special Education Teacher Workload. Journal of Special Education Technology, 01626434241257240.
Mosher, M., Dieker, L., & Hines, R. (2024). The Past, Present, and Future Use of Artificial Intelligence in Teacher Education. Journal of Special Education Preparation, 4(2), 6-17.
Suggested Citation
Seung, Y., & the CIDDL Team. (2024). Multi-Agent Systems for IEP Development. The Center for Innovation, Design, and Digital Learning.
This work is licensed under a Creative Commons Attribution 4.0 International License.