Key points:
The University of Wyoming is utilizing its new CoIEP initiative–a multi-agent system powered by large language models (LLMs)–to significantly enhance artificial intelligence and human collaboration in addressing challenges in (special) education.
The first complex challenge the UW College of Education will use AI-human collaboration to tackle is co-creating and co-evaluating Individualized Education Programs (IEPs) for students with disabilities in Wyoming and beyond.
Mandated by the federal Individuals with Disabilities Education Act (IDEA), IEPs are legally binding documents that outline interconnected components (e.g., annual educational goals and specially designed instruction) to ensure students with disabilities have access to the general education curriculum and high-quality learning experiences. Serving as the cornerstone of special education (Yell et al., 2016), IEPs play a critical role in tailoring educational experiences to meet the unique needs of students with disabilities.
The UW research team has developed a mid-fidelity prototype of CoIEP, specially designed to streamline the complex process of IEP development by breaking down the step-by-step process of creating core components of IEPs. Preliminary evaluations suggest that CoIEP has the potential to support educators in creating and evaluating core IEP components and providing individualized instruction for students with disabilities (Zhang et al., under review).
Developing an IEP is a complex process that requires significant time and expertise to analyze student data from various sources, identify student strengths and needs to develop goals, design evidence-based instructional practices, and address other supports (e.g., modifications, accommodations) that can support students in achieving the goals and making meaningful progress in general education curriculum. The process can be time-consuming and daunting for special education teachers, particularly pre-service and novice teachers who are newly introduced to the field, as well as experienced educators who may have a caseload of up to 50 students at a time to support but lack sufficient resources and time (Hogue & Taylor, 2020).
Furthermore, many school districts across the United States, including those in Wyoming, have faced issues related to IEP requirement violations due to the complex IEP development. These compliance issues may negatively affect the special education teacher workforce, which, in turn, impacts access to high-quality instruction for students with disabilities. To address these challenges, CoIEP leverages a team of LLM-powered agents to support educators in creating three required interconnected components of an IEP, including 1) the Present Level of Academic Achievement and Functional Performance statement, 2) IEP goals, and 3) individualized services and supports in a sequential order.
A multi-agent system is an advanced AI system in which multiple LLM-powered agents collaborate to perform sub-tasks of a complex task (Guo et al., 2024). One advantage of such systems lies in leveraging the capacity of multiple agents, instead of a single agent, to collaboratively complete specific parts of a complex task. Additionally, different agents can be equipped with distinct tools (e.g., data analysis tool, calculation tool, databases for evidence-based instructional practices) that enable them to complete the subtask more efficiently and accurately. Compared to other general-purpose AI systems (e.g., ChatGPT), this specialized, collaborative approach prevents AI agents from deviating from their specific tasks, producing more reliable outcomes and reducing the risk of hallucination (i.e., inaccurate information).
In CoIEP, different agents are designed and prompt-engineered–techniques to instruct AI agents to generate desired output–to demonstrate the step-by-step creation process and specific elements of each component of an IEP. Additionally, this system supports the keeping-humans-in-the-loop (HITL) function, allowing educators to provide feedback on each component generated by one agent before the task is passed to the next agent. These step-by-step and AI-human collaborative approaches further increase the transparency of the agent workflow and ensure the quality of generated IEP components.
The research team plans to conduct a series of pilot studies in Wyoming schools this year to test the effectiveness, usability, and user experience of the mid-fidelity prototype of CoIEP. Pre- and in-service special education teachers and directors across Wyoming and other states will be recruited to participate in these studies.
It is important to emphasize that developing IEPs involves collaborative efforts from the IEP team members (e.g., educators, families, students, school psychologists, and school administrators). Thus, CoIEP serves as a collaborative tool that supports the IEP team by co-creating and co-evaluating core IEP components. Instead of spending valuable time on paperwork, the team can focus on critically reviewing the generated content, engaging in meaningful problem-solving, and discussing effective strategies to implement the IEP.
The team can further refine the components during the IEP meeting, incorporating up-to-date student data and insights from all members to ensure a comprehensive, individualized, and high-quality IEP to best meet the student’s needs. The highly flexible workflow of CoIEP, where each agent operates independently, makes it a versatile tool that can serve different purposes based on the needs of its end users. For example, CoIEP can act as a professional learning tool by demonstrating the step-by-step process of creating IEP documents, helping educators build their expertise. It can also function as an evaluation tool to assess whether an IEP aligns with IDEA requirements, ensuring compliance and quality.
CoIEP will make a significant contribution to the research on AI in special education, particularly by providing a promising tool to reduce the excessive workloads of special education teachers in and beyond Wyoming and enhance the quality of IEPs for students with disabilities. From an economic perspective, our cost analysis suggested that CoIEP could be a cost-efficient professional learning tool compared to traditional professional development or on-site coaching, which are often high-cost and resource-intensive (Kraft et al., 2018).
More importantly, CoIEP offers a personalized approach to supporting educators in developing high-quality IEPs, thereby holding great potential for improved outcomes for SWDs. After being fully developed and deployed, educators can access CoIEP anytime, anywhere in and beyond Wyoming. Given the significant teacher shortage in special education and the associated high cost of teacher attrition (Billingsley & Bettini, 2019), CoIEP has the potential to address both historical and ongoing challenges in special education.
