What Is Generative AI?

AI, machine learning, and generative AI

Artificial intelligence is an umbrella term for computer systems designed to perform tasks associated with capabilities such as perception, prediction, language, planning, or decision support. Machine learning is one way of building such systems: instead of encoding every rule directly, developers train a model to identify patterns in data.

Generative AI produces new content—such as text, images, audio, video, or code—from a user’s instructions and other inputs. Large language models (LLMs) are generative models trained to predict and produce sequences of language. Their ability to generate plausible text does not mean that they understand a subject, hold beliefs, or consult a reliable database of facts.¹

Model and product are not the same

A model is the underlying system. A product may combine one or more models with search, uploaded files, calculators, code execution, memory, safety rules, and an interface. Two products using a related model can therefore behave differently, and the same product can change after an update.

When evaluating an educational use, ask about the complete product and workflow:

• What model and additional tools does it use?
• Can it retrieve current or supplied sources?
• What happens to prompts, files, and outputs?
• Does the provider use them for training or product improvement?
• Can users inspect sources, export work, and delete data?
• Is the product approved and accessible for the intended learners?

How an LLM produces a response

An LLM generates text incrementally from patterns learned during training and from the current context. The response depends on the prompt, supplied material, system instructions, model version, settings, and sometimes external tools. Small changes can produce different results.

This explains both the flexibility and the central risk: a model can produce a clear, relevant answer even when the answer is unsupported or wrong. Fluency is not evidence.

What generative AI can contribute

Depending on the tool and task, it can help educators or students:

• generate alternatives and questions;
• transform supplied material into another structure or reading level;
• draft explanations, activities, examples, code, or media;
• simulate a perspective for practice;
• critique work against explicit criteria; or
• support repeated, low-stakes practice.

These are possible functions, not guaranteed benefits. The educational value depends on what the learner does, what the model replaces, the quality of the interaction, and the review process.

What it cannot establish by itself

Model output is not sufficient evidence that:

• a factual claim is true or current;
• a citation exists and supports the claim;
• a teaching activity is aligned, inclusive, or feasible;
• a student understands, authored, or misused a piece of work;
• a grade or other consequential decision is fair; or
• using the tool improves learning.

Continue with Limitations and risks, then use When to use generative AI before applying a tool to a real task.

References & Footnotes

• National Institute of Standards and Technology. (2024). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. https://doi.org/10.6028/NIST.AI.600-1 
• UNESCO. (2023). Guidance for generative AI in education and research. https://unesdoc.unesco.org/ark:/48223/pf0000386693