Imagine a world where special education teachers have an invaluable assistant—a tireless digital companion capable of drafting IEPs, offering tailored lesson plans, and even communicating with parents. This isn’t science fiction; it’s the reality we’re moving toward with the growing presence of artificial intelligence in special education. But the science, so far, is imperfect, and users must guard against potential pitfalls.  

You may already know that AI can write a meal plan or create a vacation itinerary. AI tools like Chat GPT and Claude, as well as school-focused services like Magic School, can also assist with the tasks like: 

• Drafting IEP goals 
• Summarizing data for progress reports 
• Drafting emails to parents 
• Differentiating lessons and modifying assignments 
• Providing feedback to students 
• Identifying behavioral antecedents and interventions 
• Translating meeting deliberations 

But, as attorneys, we also have to warn you of the significant risks of using AI tools. Sometimes the information AI gives you is incorrect, incomplete, or biased. AI tools “learn” from collecting and processing large amounts of data, recognizing patterns, and making predictions. If the data was skewed, outdated, or limited, the output will similarly be flawed.  

The second concern when using AI tools for special education is that the output is often generic, while the IDEA and Section 504 are based on individualization. If you ask Chat GPT to write an IEP goal for a 3rd grader with a specific learning disability, you might get something like:  

Goal: By the end of the IEP period, when given a reading passage at the 3rd-grade level, the student will demonstrate improved reading comprehension skills by accurately summarizing the main idea and identifying key details with 80% accuracy on three consecutive assessments. 

Baseline: The student currently reads at a 2nd-grade level, struggles with summarizing and identifying key details in grade-level texts, and achieves 60% accuracy on similar assessments. 

Measurement and Evaluation: 

Progress towards this goal will be monitored through regular assessments of the student’s reading comprehension skills, including but not limited to teacher-made quizzes, standardized tests, and classroom observations. Data will be collected and analyzed at least once every grading period to determine the student’s performance. 

Support and Accommodations: 

To support the student in achieving this goal, the following accommodations and strategies will be implemented as outlined in the IEP: 

• The use of graphic organizers to help structure and visualize information. 
• Preferential seating to minimize distractions during reading activities. 

• Small group instruction to provide targeted support and practice in reading comprehension. 
• Extended time for assignments and assessments when necessary. 

Not bad, right? But, it assumed the specific learning disability was in reading comprehension, which may not be accurate for your student. The goal is set at grade level, which may or may not be achievable and ambitious for your student. And it made up the baseline, which may or may not be accurate for your student. The measurement is “at least once every grading period”; but once is likely inadequate and the three measurement methods are not going to provide comparable data points to demonstrate progress. The supports and accommodations are helpful ideas, but they may or may not be effective or appropriate for your student. If a teacher cut and pasted this information into an IEP, it would likely not be an appropriate goal for a particular student.  

These shortfalls bring us to both a tip and a caution. The tip first: the output will be more on target when you provide more detailed input. For instance, if you tell Chat GPT the area of student’s specific learning disability, the student’s current skills and weaknesses, and the TEKS you want to focus on, the output will be closer to what you need. But here’s the caution: putting identifiable student data into Chat GPT is a disclosure that will likely be considered a FERPA violation. So entering too much information, student names, parent names, or other PII could also lead to trouble. At some point, your IEP software may be able to use student evaluations and progress monitoring data to draft goals and interventions in a closed and confidential system, but for now, existing tools do not protect student privacy.  

As you venture into the world of AI, remember that it is a helpful assistant but lacks the knowledge and judgment of human special educators. It is great for brainstorming ideas (give me 10 accommodations for a student who has trouble staying in their seat, give me 20 activities to reinforce math facts) and can help with many other tasks (it wrote the introduction to this blog post!). But also use caution; double check for accuracy and sources, don’t input confidential information, remember it may take multiple iterations and significant editing to get the product you want, and be sure to check your district’s policies and guidelines.   

To learn more, look for Rebecca’s session at the GCASE Fall Forum October 18-20, 2023, and Kendra and Rebecca’s session at the TCASE Great Ideas Conference February 5-7, 2024.