AI Literacy Glossary

The ethical standards that govern how students and educators use AI tools in academic work — ensuring that AI assists learning without replacing original thought, honest authorship, or intellectual ownership.

An AI-powered educational approach that continuously adjusts the difficulty, pace, and content of instruction based on each individual student's performance, learning style, and progress data.

An AI system that can perceive its environment, make decisions, plan multi-step actions, and execute tasks autonomously to achieve a specified goal — operating with minimal human intervention.

A set of step-by-step rules or instructions that tells a computer how to solve a problem or make a decision. Machine learning algorithms learn to improve their performance through experience with data.

A hypothetical type of AI that would possess human-level intelligence across virtually all cognitive tasks — capable of learning, reasoning, and applying knowledge across domains the same way a human can. AGI does not yet exist.

The simulation of human intelligence processes by machines, especially computer systems — including learning, reasoning, problem-solving, perception, and language understanding.

The use of technology — including AI — to perform tasks with minimal or no human intervention, enabling repetitive processes to be completed faster, more consistently, and at scale.

A standardized test or evaluation used to measure an AI model's performance across specific tasks, allowing researchers and users to compare models against each other and against human performance levels.

Systematic and repeatable errors in AI outputs that create unfair, inaccurate, or discriminatory outcomes — typically resulting from biased training data, flawed model design, or unrepresentative datasets.

A software application designed to simulate conversation with human users — ranging from simple rule-based systems to sophisticated AI models capable of nuanced, context-aware dialogue.

A field of AI that enables computers to derive meaningful information from digital images, videos, and other visual inputs — allowing machines to identify objects, people, scenes, and actions.

The maximum amount of text (measured in tokens) that an AI language model can process and "remember" in a single interaction. Information outside this window is not accessible to the model.

The complex and evolving legal questions around ownership and intellectual property rights for AI-generated content — including who owns AI outputs, whether AI training on copyrighted data is legal, and how educators should handle AI-generated materials.

The use of AI to tailor educational content, pace, and resources to individual student needs, learning styles, background knowledge, and performance data — moving away from one-size-fits-all instruction.

A personalized version of ChatGPT configured with specific instructions, knowledge, and behaviors for a defined purpose — allowing educators to create subject-specific AI tutors, classroom assistants, or specialized tools without coding.

The legal and ethical obligation to protect personally identifiable information (PII) when using AI tools — particularly relevant for educators under FERPA, which prohibits sharing student data with unauthorized third parties including AI platforms.

AI-generated synthetic media — typically video, audio, or images — that realistically depicts a real person saying or doing something they never actually said or did, created using deep learning techniques.

A subset of machine learning that uses artificial neural networks with many layers (hence "deep") to learn complex patterns from large datasets — enabling breakthroughs in image recognition, language processing, and generative AI.

A type of generative AI model that creates new images or data by learning to gradually remove "noise" from random pixels — the technology behind many AI image generators like DALL·E, Midjourney, and Stable Diffusion.

Numerical representations of words, sentences, images, or other data in a high-dimensional mathematical space — allowing AI models to capture meaning and relationships, so that similar concepts are represented by nearby numbers.

The principles, frameworks, and practices that guide the development and use of AI in ways that are fair, transparent, accountable, safe, and beneficial to individuals and society — including in educational settings.

The degree to which humans can understand and trace how an AI model arrived at a specific output or decision — addressing the "black box" problem where AI reasoning is opaque even to its creators.

A prompting technique where you provide an AI model with a small number of examples (typically 2-5) within the prompt itself to demonstrate the format, style, or task you want — improving output quality without retraining the model.

The process of taking a pre-trained AI foundation model and further training it on a smaller, domain-specific dataset — adapting it to perform better on particular tasks, styles, or subject areas.

A large-scale AI model trained on broad, diverse datasets that serves as the base from which many specialized applications are built — including GPT-4 (ChatGPT), Gemini, Claude, and other AI tools educators use.

AI systems capable of generating new, original content — including text, images, audio, video, and code — based on patterns learned from training data, in response to user prompts.

Safety constraints and content filters built into AI systems to prevent them from generating harmful, dangerous, or inappropriate outputs — the mechanisms that cause AI to decline certain requests.

When an AI model generates false, nonsensical, or unverified information but presents it as fact with the same confident tone as accurate information — a structural artifact of how language models predict text.

An AI design approach where human judgment, review, or approval is required at key decision points — ensuring that critical decisions are not made autonomously by AI without human oversight.

The process of using a trained AI model to generate outputs — predictions, responses, or content — based on new inputs. When you type a question into ChatGPT and receive an answer, you're triggering inference.

AI tools and systems specifically designed or adapted for educational contexts — including AI tutors, automated feedback systems, content generators, and assessment tools built for or used by educators.

The professional practice of refining AI outputs through multiple rounds of follow-up prompts — progressively improving the quality, accuracy, and specificity of AI-generated content rather than accepting first-generation outputs.

A type of AI model trained on vast amounts of text data to understand and generate human-like language — the technology underlying ChatGPT, Claude, Gemini, and most conversational AI tools.

Digital platforms (such as Google Classroom, Canvas, or Schoology) that increasingly incorporate AI features to automate grading, personalize learning pathways, flag at-risk students, and generate insights from student performance data.

A subset of AI that enables systems to learn and improve from experience without being explicitly programmed — using statistical patterns in data to make predictions and decisions.

The ability to critically evaluate, analyze, and understand AI-generated content — including recognizing synthetic media, identifying AI-authored text, questioning AI accuracy, and understanding how AI systems can reflect bias or misinformation.

AI systems capable of processing and generating multiple types of data — combining text, images, audio, and video in a single model. GPT-4o and Gemini Ultra are examples of multimodal AI.

A branch of AI that enables computers to understand, interpret, and generate human language — the foundational technology behind translation services, voice assistants, chatbots, and language models.

A computing system inspired by biological neural networks in animal brains — consisting of interconnected nodes (neurons) that process information in layers, enabling pattern recognition and complex learning.

Open source AI models (like Meta's Llama) make their code and weights publicly available, allowing anyone to inspect, modify, or deploy them. Closed source models (like GPT-4 or Claude) are proprietary, accessible only through APIs or products.

The text, image, audio, video, or other content that an AI model generates in response to a prompt or input — the end product of an AI inference process.

The internal numerical variables that an AI model learns and adjusts during training — determining how it processes input and generates output. GPT-4 has an estimated 1.8 trillion parameters.

The use of AI to generate individualized, specific, and actionable feedback on student work — tailored to each student's level, learning goals, and specific errors rather than generic comments.

The input, instruction, or question given to an AI system to direct its output — ranging from simple one-line questions to complex structured instructions with role context, parameters, and format specifications.

The practice of designing, structuring, and refining prompts to consistently elicit high-quality, accurate, and useful outputs from AI models — a professional skill involving context-setting, parameter specification, and format direction.

Any data that can be used to identify a specific individual — including names, student ID numbers, addresses, dates of birth, or academic records. Federal law (FERPA) prohibits sharing student PII with unauthorized parties, including AI tools without proper data agreements.

A technique that combines an AI language model with a real-time search or retrieval system — allowing the model to pull in current, specific information from a knowledge base before generating a response, reducing hallucination.

A training technique used to align AI models with human preferences — human raters evaluate AI outputs and the model is trained to produce responses humans prefer, making it more helpful, honest, and safe.

The extent to which AI-generated content, imagery, and examples reflect the diversity of all communities — including race, ethnicity, culture, gender, ability, and lived experience. AI often defaults to dominant cultural representations if not explicitly directed otherwise.

Artificially generated data created by AI models that mimics the statistical patterns of real-world data — used to train AI models, fill data gaps, or protect privacy when real data is unavailable or sensitive.

Hidden background instructions given to an AI model before the user interaction begins — used to set the AI's persona, behavioral rules, and constraints. System prompts are how Custom GPTs and specialized AI tools are configured.

The interdisciplinary field focused on ensuring that AI systems behave safely, reliably, and in alignment with human values — including preventing unintended harmful behaviors, misuse, and catastrophic failures as AI systems become more powerful.

AI technology that converts spoken language into written text — powering voice assistants, dictation software, real-time captioning, and accessibility tools that support students with disabilities and language learners.

A parameter that controls the randomness and creativity of AI outputs — lower temperature produces more predictable, focused responses while higher temperature produces more varied, creative, and unexpected outputs.

AI systems that generate images from text descriptions (prompts) — including Midjourney, DALL·E, Adobe Firefly, and Stable Diffusion. Used by educators to create custom classroom illustrations, visual aids, and instructional materials.

The basic unit of text processed by a language model — tokens can be a whole word, part of a word, punctuation, or a single character. An AI model's context window and pricing are typically measured in tokens.

The large dataset used to train an AI model — teaching it patterns, facts, language, and relationships. The quality, diversity, and size of training data fundamentally determines an AI model's capabilities and limitations.

The neural network architecture introduced in 2017 that revolutionized AI — using "attention mechanisms" to process all words in a text simultaneously and understand context across long distances. The "T" in ChatGPT stands for Transformer.

Techniques for embedding invisible or visible markers in AI-generated content to identify it as synthetic — enabling detection tools to distinguish AI-authored text, images, and video from human-created content.

An AI's ability to perform a task it has never been explicitly trained on and given no examples for — relying entirely on its broad pre-training knowledge to generalize to novel situations.