AI terms, in
plain language.

The vocabulary that comes up when someone asks who is accountable for this.

Shadow AI

AI tools and services used inside an organization without approval or oversight from IT or security. The AI version of shadow IT.

AI Ethics

The study of moral issues raised by how AI is built and used.

AI Safety

The practices and research aimed at keeping AI systems from causing unintended harm.

Explainable AI

AI whose decisions can be inspected and understood by a person, rather than taken on faith.

Algorithmic Bias

Systematic errors in an algorithm that produce unfair outcomes for some groups of people.

AI Evals

Structured tests that measure how an AI system performs against a defined set of inputs, so you catch regressions before shipping.

Terms for AI that does not just answer, but acts. This is where most of the current confusion lives.

AI Agent

An AI system that can perceive its environment, make decisions, and take actions, rather than only responding to a prompt.

Agentic Workflow

A structured sequence of tasks where AI agents make independent decisions, adapt to changing conditions, and act with minimal human intervention.

Looping

An agent loop is the repeating cycle of thought, tool call, and observation that an agent runs until it reaches an answer or hits a stop condition.

Prompt Chaining

Breaking a complex task into a series of smaller connected prompts, where the output of one becomes the input to the next.

Autonomous Systems

Systems that operate independently, without a person intervening at each step.

Robotics

The design, construction, operation, and application of physical robots.

How AI systems get attacked, and the newer attacks that only exist because AI does.

Slopsquatting

Registering a software package name that an AI hallucinated, hoping someone following an AI's suggestion installs it without checking. A twist on typosquatting: instead of relying on human typos, it relies on AI mistakes.

Data Poisoning

Deliberately feeding bad data into a model's training set to corrupt what it learns or steer how it behaves.

Adversarial Attack

Inputs crafted specifically to fool a model into getting something wrong.

AI Hallucination

When an AI states something false or misleading with complete confidence.

The foundations. Worth knowing precisely, because they get used loosely.

Generative AI

AI that creates new content such as text, images, or music, rather than only classifying or predicting.

Machine Learning

AI that learns patterns from data instead of being explicitly programmed with rules.

Deep Learning

A subset of machine learning that uses neural networks with many layers to learn complex patterns.

Neural Network

A system of algorithms loosely modeled on the structure of the human brain.

Natural Language Processing

An AI system's ability to understand, interpret, and produce human language.

Computer Vision

An AI system's ability to interpret images and video.

Tokenization

Two different meanings worth keeping straight. In data security, replacing sensitive data with non-sensitive stand-in values called tokens. In language models, splitting text into the units a model actually processes.

Chatbot

A program designed to hold a conversation with a person.

Narrow AI

AI built to do one specific task well, with no general capability beyond it.

General AI

A hypothetical system with human-level intelligence able to handle any intellectual task.

AI Singularity

A hypothetical point at which AI exceeds human intelligence and begins improving itself recursively.

The training methods behind the systems, in plain terms.

Supervised Learning

Training on labeled data, where the correct answers are supplied.

Unsupervised Learning

Training on unlabeled data, where the model finds patterns on its own.

Reinforcement Learning

Training through trial and error, with rewards for desired behavior and penalties for the rest.

Transfer Learning

Taking a model trained for one task and reusing it for a different but related one.

Active Learning

The model chooses which examples it wants labeled, prioritizing the ones it will learn most from.

Federated Learning

Training across many decentralized data sets without moving the underlying data, which keeps it private.

What organizations are actually deploying today.

Predictive Analytics

Using historical data and statistics to forecast future outcomes.

Recommendation Engine

A system that suggests content or products based on behavior and preferences.

Sentiment Analysis

Using language processing to work out the emotional tone behind a piece of text.

Vibe Coding

Describing the software you want in natural language and letting an AI assistant generate, refine, and debug the code, rather than writing the syntax yourself. Prioritizes speed and experimentation, and lets people build with little coding experience.

Terms you will hear in the discourse. Useful to recognize, and useful to be skeptical of.

AI Boom

A period of rapid growth and heavy investment in AI.

AI Winter

A period of reduced funding and fading interest in AI research. There have been several.

AI Hype

Inflated expectations and marketing that run ahead of what the technology actually does.

Token Maxing

Maximizing how many AI tokens you consume on the theory that more usage equals more productivity. It usually just means a bigger bill.

Taste

The capacity to make informed, opinionated choices, such as editorial judgement and design sensibility, to tell what is actually good rather than what is statistically average.