AI Literacy for School Psychologists: A Practical Glossary
You're in a district meeting. Someone mentions an AI tool for writing IEP goals. A colleague asks about data privacy. The vendor uses terms like "large language models" and "training data." Everyone nods along.
But here's the thing: Does anyone in that room actually understand what they're agreeing to?
Most of us became school psychologists to support students, not to decode technology. Yet AI-powered tools are entering our workflow. Without understanding what these systems actually do, we can't distinguish between real utility and marketing hype.
This glossary is a "field guide" to translate jargon into plain language and connect abstract concepts back to your daily practice.
Table of Contents
The Basics
Artificial Intelligence (AI)
Algorithm
Automation
Generative AI
Machine Learning (ML)
How It Works
Chatbot
Context Window
Embedding
Inference
Large Language Model (LLM)
Natural Language Processing (NLP)
Prompt Engineering
Tokenization
Training Data
Safety, Ethics & Privacy
Bias and Fairness
Data Privacy
Data Security
Ethical AI
FERPA and AI
Hallucination
Human-in-the-Loop
PII (Personally Identifiable Information)
Transparency
Implementation
Model Fine-Tuning
Version Control
Who This Resource Is For
School Psychologists evaluating privacy risks of new tools.
Special Educators decoding vendor pitches.
School Leaders responsible for ethical technology choices.
Anyone designing professional development on AI literacy.
The Glossary
Artificial Intelligence (AI)
Definition: Computer systems designed to perform tasks that typically require human intelligence, such as understanding language, recognizing patterns, making decisions, or generating text.
Why it matters: AI tools are increasingly used for drafting IEP goals or analyzing data. Understanding their capabilities and limits ensures you keep professional judgment central.
Example: Using an AI writing assistant to draft a psychoeducational report, then verifying all recommendations based on clinical expertise.
Algorithm
Definition: A set of step-by-step instructions that tells a computer how to solve a problem or complete a task.
Why it matters: Algorithms power screening programs and suggestion tools. These follow programmed rules, meaning they have built-in limitations and biases you must recognize.
Example: A reading screener uses an algorithm to compare student scores against benchmarks to flag those needing support.
Automation
Definition: Using technology to perform repetitive tasks without human intervention.
Why it matters: Automation handles admin tasks (scheduling, organizing), freeing you for direct service. However, clinical decision-making should never be fully automated.
Example: Setting up automated reminders for annual review meetings.
Bias and Fairness
Definition: When AI systems produce results that systematically favor or disadvantage certain groups, often reflecting biases in their training data.
Why it matters: Biased tools can misidentify students or overlook those needing support. You must critically evaluate AI recommendations for equity issues.
Example: An AI behavior system trained on suburban data misinterpreting culturally typical behaviors in urban settings as problematic.
Chatbot
Definition: An AI program designed to have text-based conversations with users.
Why it matters: Schools use these for communication. You must understand their limits in handling sensitive situations and when human intervention is required.
Example: A website chatbot answering basic evaluation questions but routing crisis keywords immediately to a counselor.
Context Window
Definition: The amount of text an AI system can "remember" or consider at one time when generating responses.
Why it matters: If you exceed the context window (e.g., a very long report), the system may "forget" earlier information, leading to inconsistent outputs.
Example: Pasting a 15-page evaluation into a tool may cause it to forget background history when summarizing recommendations at the end.
Data Privacy
Definition: The protection of personal information from unauthorized access, use, or disclosure.
Why it matters: Before using any tool, you must verify it complies with privacy regulations and doesn't store student data to train public models.
Example: Never uploading identifiable information to a free, public tool like ChatGPT.
Data Security
Definition: The technical safeguards used to protect data from breaches or hacks.
Why it matters: Privacy promises mean nothing without security. Verify platforms use encryption and access controls.
Example: Choosing a tool with end-to-end encryption and multi-factor authentication.
Embedding
Definition: A mathematical representation converting words or concepts into numbers so AI can process similarities.
Why it matters: This powers features like searching past reports for similar cases (e.g., recognizing "anxious" and "worried" are related).
Example: An AI system recognizing that a "socially withdrawn" student shares characteristics with past cases of "peer relationship difficulties."
Ethical AI
Definition: Designing and using AI in ways that align with moral principles, professional standards, and human dignity.
Why it matters: Your ethical obligations remain constant. Tools must support equity, maintain confidentiality, and strengthen (not replace) judgment.
Example: Rejecting a tool that makes diagnostic suggestions without transparent methodology.
FERPA and AI
Definition: The Family Educational Rights and Privacy Act; federal law protecting student records, which applies to AI handling of that data.
Why it matters: Compliance requires signed vendor agreements, limits on data sharing, and protecting parental rights.
Example: Verifying a vendor has signed a FERPA agreement before using their note-taking tool for IEP meetings.
Generative AI
Definition: AI systems that create new content (text, images) based on patterns learned from examples.
Why it matters: These tools can draft reports or materials, but all outputs require review for accuracy and alignment with student needs.
Example: Using generative AI to draft accommodations, then editing them based on evidence-based practices.
Hallucination
Definition: When an AI generates information that sounds plausible but is factually incorrect or fabricated.
Why it matters: AI can confidently state false legal requirements or citations. You must verify everything against reliable sources.
Example: An AI tool citing a nonexistent research study to support an intervention recommendation.
Human-in-the-Loop
Definition: An approach where humans review and make final decisions on AI outputs.
Why it matters: Essential for liability and ethics. You remain responsible for all decisions and documentation.
Example: Using AI to identify behavior patterns, but conducting your own functional analysis before finalizing a BIP.
Inference
Definition: The process where an AI applies learned patterns to make predictions on new inputs.
Why it matters: It helps you realize the tool doesn't "understand" the student; it is predicting based on probability and past training.
Example: A tool suggesting interventions based on similar cases without knowing the specific family dynamics of your student.
Large Language Model (LLM)
Definition: An AI system trained on vast amounts of text to understand and generate human-like language.
Why it matters: LLMs (like Claude or Gemini) power most writing tools. They lack clinical training and must not be used for diagnostic decisions.
Example: Using an LLM to smooth out the phrasing in a report, but not to determine the eligibility classification.
Machine Learning (ML)
Definition: AI where systems learn patterns from data rather than following explicit rules.
Why it matters: ML is used for risk prediction. Because it learns from historical data, it can perpetuate historical inequities if not monitored.
Example: An early warning system predicting dropout risk based on attendance and grade patterns.
Model Fine-Tuning
Definition: Taking a pre-trained AI and training it further on specialized data (e.g., education law).
Why it matters: "Fine-tuned" tools may be more accurate for school psychology tasks, but you must still ask what data was used.
Example: An AI writing assistant fine-tuned on IEP documents vs. a general purpose marketing writer.
Natural Language Processing (NLP)
Definition: Technology enabling computers to understand and interpret human language.
Why it matters: NLP powers transcription and theme extraction. It saves time but may miss tone or cultural context.
Example: Reviewing an NLP transcript of a parent interview to ensure emotional undertones weren't missed.
PII (Personally Identifiable Information)
Definition: Any information that can identify a specific individual (Name, DOB, ID numbers, etc.).
Why it matters: Exposing PII to unsecured AI violates FERPA. You must de-identify data before using non-approved tools.
Example: Replacing a student's name with "Student A" and removing school names before asking an AI for editing help.
Prompt Engineering
Definition: Crafting clear, specific instructions to get accurate responses from AI.
Why it matters: The quality of the output depends on the quality of your input.
Example: Instead of "Write a goal," prompting: "Draft a measurable IEP goal for a 4th grader reading at a 2nd-grade level, focusing on decoding."
Tokenization
Definition: Breaking text down into smaller chunks (tokens) for the AI to process.
Why it matters: Explains why tools have input limits (token limits). You may need to section out long reports.
Example: Breaking an 8,000-token report into two sections because the tool has a 4,000-token limit.
Training Data
Definition: The collection of information used to teach an AI system.
Why it matters: AI reflects its training. If the data lacks diversity or contains outdated practices, suggestions will be flawed.
Example: A tool trained on general education data providing irrelevant suggestions for a specialized program.
Transparency
Definition: The degree to which you can understand how an AI system reaches its conclusions.
Why it matters: You cannot rely on a "black box." You must be able to explain and defend your decisions in hearings.
Example: Choosing a tool that cites the specific data factors influencing its suggestion.
Version Control
Definition: Tracking different versions of AI systems as they update.
Why it matters: Tools change. Documenting which version you used protects your professional accountability.
Example: Noting in your records that you used "Report Assistant v2.3" in case the tool's logic changes later.
Moving Forward
You now have a shared vocabulary for AI in schools. Use this to ask better questions, spot red flags, and participate in adoption conversations with confidence.
The real work remains the same: asking hard questions, centering student welfare, and maintaining the clinical judgment that makes you effective.