Special education in the United States is underfunded and pervasive, with many school districts struggling to hire qualified and motivated professionals.
Amid these long-standing challenges, there is growing interest in leveraging artificial intelligence tools to fill some of the gaps that school districts currently face and reduce labor costs.
More than 7 million children receive federally funded entitlements under the Individuals with Disabilities Education Act. The law ensures that students receive instruction tailored to their physical and psychological needs, and provides a legal process for families to negotiate support. Special education involves a variety of professionals, including rehabilitation specialists, speech therapists, and classroom instructional assistants. However, these professionals are in short supply despite the proven need for their services.
As an associate professor of special education who works with AI, I understand the potential and pitfalls of AI. While AI systems can reduce administrative burden, provide expert guidance, and help busy professionals manage cases, they can also raise ethical challenges, from machine bias to broader questions about trust in automated systems. It also risks exacerbating existing problems with how special education services are delivered.
But rather than waiting for the perfect solution, some in the field are choosing to test AI tools.
Faster IEPs, but how individualized are they?
AI is already shaping special education planning, workforce preparation, and assessment.
One example is the Individualized Education Program (IEP), which is the primary vehicle for guiding which services a child can receive. An IEP uses a variety of assessments and other data to describe a child’s strengths, determine needs, and set measurable goals. Every part of this process relies on trained professionals.
However, with ongoing staffing shortages, school districts often struggle to complete assessments, update plans, and integrate parent input. Most school districts develop IEPs using software that requires practitioners to choose from generalized, rote answers or options, increasing the level of standardization and potentially failing to meet a child’s true individual needs.
Preliminary research shows that large-scale language models such as ChatGPT are adept at leveraging multiple data sources, including information from students and families, to generate key special education documents, such as IEPs. Chatbots that can quickly create IEPs could help special education practitioners better meet the needs of individual children and their families. Some special education professional organizations are encouraging educators to use AI in documents such as lesson plans.
Disability training and diagnosis
AI systems may also support professional training and development. In my own workforce development work, I combine several AI applications with virtual reality to allow practitioners to rehearse instructional routines before engaging directly with children. Here, AI can serve as a practical extension to existing training models, providing repeated practice and structured support in ways that are difficult to maintain with limited human resources.
Some school districts have begun using AI for assessments, which may include academic, cognitive, and medical assessments. AI applications that combine automatic speech recognition and language processing are currently being employed in computer-mediated oral reading assessments to score tests of student reading comprehension.
Practitioners often struggle to understand the amount of data that schools collect. AI-driven machine learning tools can also help here by identifying patterns that are not immediately recognizable to educators when making assessments and instructional decisions. Such support could be particularly useful in diagnosing disorders such as autism and learning disabilities, where masking, variable presentation, and incomplete medical histories can make interpretation difficult. My ongoing research shows that current AI can make predictions based on data that may be available in some districts.
Privacy and trust concerns
There are serious ethical and practical issues with these AI-assisted interventions, from risks to student privacy to machine bias to deeper questions of family trust. Some rely on the question of whether AI systems can provide services that are truly compliant with existing laws.
The Individuals with Disabilities Education Act requires nondiscriminatory methods of assessing disabilities to avoid inappropriately identifying students for services or failing to provide services to eligible students. Additionally, the Family Educational Rights and Privacy Act clearly protects student data privacy and parents’ rights to access and retain their children’s data.
What happens if an AI system uses biased data or techniques to generate recommendations for a child? What happens if a child’s data is misused or leaked by the AI system? By using an AI system to perform some of the functions described above, families are placed in a position where they are expected to trust not only the school district and its special education personnel, but also commercially available AI systems whose inner workings are largely inscrutable.
These ethical issues are not unique to special education. Many have been nurtured in other fields and addressed by early adopters. For example, automatic speech recognition (ASR) systems have struggled to accurately assess accented English, but many vendors are now training their systems to accommodate specific ethnic or regional accents.
However, ongoing research work suggests that some ASR systems have limited ability to accommodate speech differences associated with impairment, account for classroom noise, and distinguish between different speech sounds. These issues may be resolved by technological improvements in the future, but for now they are consequential.
embedding bias
At first glance, machine learning models may appear to improve traditional clinical decision-making. But because AI models must be trained on existing data, their decisions can continue to reflect long-held biases about how to identify failures.
In fact, research shows that AI systems are routinely hampered by biases within both the training data and system design. AI models can also introduce new biases by missing subtle information revealed during in-person assessments or by overrepresenting group characteristics in the training data.
Advocates may argue that such concerns are addressed by safeguards already built into federal law. Families have considerable latitude in what they agree to and can choose alternatives as long as they know they can direct the IEP process.
Similarly, using AI tools to construct IEPs and lessons may seem like a clear improvement over undeveloped or token plans. But true personalization requires feeding protected data into large language models, potentially violating privacy regulations. And while AI applications can easily create better-looking IEPs and other documents, that doesn’t necessarily mean better services.
fill the gap
Indeed, it is not yet clear whether AI can provide a standard of care comparable to the high-quality traditional care afforded to children with disabilities under federal law.
In 2017, the Supreme Court rejected the idea that the Individuals with Disabilities Education Act would provide only minor “de minimis” advances for students, undermining one of the key rationales for pursuing AI: that AI can meet minimum standards of care and practice. And because AI has not actually been empirically evaluated at scale, it has not been proven that it satisfies the low bar of merely improving a flawed status quo.
However, this does not change the reality that resources are limited. For better or worse, AI is already being used to bridge the gap between what the law requires and what systems actually deliver.
