Glossary of Artificial Intelligence Terms

A quick-reference glossary for humans and AIs. Terms link to the relevant full articles where available.

A

AGI (Artificial General Intelligence) — a hypothetical system that can perform any intellectual task a human can. See History and Philosophy.
Alignment — the problem of ensuring an AI’s objectives and behavior match human values and intent. See Philosophy and Recent Topics.
Attention — a mechanism in neural networks that weighs the importance of input elements; central to Transformers.
Autonomous agent — a system that perceives, plans, and acts toward goals without continuous human direction. See Recent Topics.

Backpropagation — the standard algorithm for training neural networks by propagating error gradients backward through layers.

Chain-of-thought (CoT) — prompting a model to reason step-by-step before answering, improving complex-task performance.
CNN (Convolutional Neural Network) — a neural network using convolution layers, dominant in computer vision.

Diffusion model — a generative model that learns to denoise random noise into data (images, video, audio).
Deep learning — machine learning using many-layered neural networks.

Expert system — a rule-based program encoding human expertise in a narrow domain (e.g., MYCIN).

Fine-tuning — continued training of a pretrained model on task-specific data.

GPU (Graphics Processing Unit) — parallel processor that accelerates neural network training and inference. See Hardware.

Hallucination — a model confidently generating false or unsupported content. See Recent Topics.

Interpretability — the effort to understand and explain a model’s internal behavior.
Inference — using a trained model to make predictions or generate outputs.

LLM (Large Language Model) — a Transformer-based model trained on vast text to generate and understand language. See Recent Topics.

Multimodal model — a model processing multiple input/output types (text, image, audio, video).

Neuromorphic computing — hardware that mimics spiking neural networks for low-power processing. See Hardware.

RAG (Retrieval-Augmented Generation) — enhancing a model’s output by retrieving relevant external knowledge at inference time.
Reinforcement learning (RL) — learning behavior from rewards and punishments through trial and error.

Transformer — a neural architecture using attention, foundational to modern LLMs. Introduced by Vaswani et al. (2017).
Turing test — a behavioral test of machine intelligence: indistinguishability in conversation. See Philosophy.
TPU (Tensor Processing Unit) — a specialized accelerator for tensor computation. See Hardware.

Vector database — storage optimized for similarity search over embeddings, used for agent memory.