Practical AI glossary: terms every business leader should know

This glossary defines AI terms every business leader encounters when evaluating AI adoption.

## Core AI concepts

Artificial intelligence (AI)

A system that learns patterns from data and uses those patterns to perform tasks, make decisions, or generate outputs.

Narrow AI

A system designed to excel at one specific task. Generative AI that writes copy. A recommendation system that suggests products. Almost everything you encounter today is narrow AI.

Machine learning (ML)

AI where the system learns patterns from data rather than being explicitly programmed. You show the system examples. It figures out the rules.

Generative AI

A system trained to generate new content—text, images, code, video, audio. You provide a prompt; it generates output. Trained on vast amounts of existing content.

Large language model (LLM)

A generative AI system trained on enormous amounts of text to understand and generate language. Powers ChatGPT, Claude, and similar systems. “Large” refers to the scale—billions of parameters.

Agentic AI

A system that can plan and take autonomous actions toward a goal without human intervention on each step. Emerging and powerful but carries more risk.

Algorithm

A step-by-step process for solving a problem. In AI, algorithms learn patterns and use those patterns to make predictions or decisions.

Parameter

In machine learning, parameters are learned values that allow the model to make predictions. More parameters usually means more capable but also slower and more expensive.

## Training and fine-tuning

Training data

The data used to teach an AI system. For language models, typically billions of words from the internet and books. Data quality matters enormously.

Training

The process of teaching an AI system to recognise patterns.

Fine-tuning

Taking a pre-trained model and updating it on a specific dataset to improve performance on a specific task. Faster and more efficient than training from scratch.

Hallucination

When an AI system generates confident-sounding outputs that are false or nonsensical. A major limitation of generative AI and a key governance concern.

Prompt engineering

Writing effective inputs to generative AI systems to get good outputs. “Write a 50-word product description emphasising waterproofing, suitable for a luxury brand” is better than “write a product description.”

## Data concepts

Data quality

How accurate, complete, and relevant your data is. Poor quality data produces poor AI systems. Data cleaning is often the majority of effort in any AI project.

Data bias

When training data has systematic patterns reflecting discrimination. Training AI on biased data perpetuates and scales that bias. A major governance concern.

Data privacy

Ensuring data isn’t misused or exposed. If you upload customer data into a generative AI system, that data may be retained. GDPR and data protection laws restrict what you can do with personal data.

Data lineage

The ability to track where data came from, how it’s been processed, and where it goes. Knowing lineage matters for understanding what could go wrong.

## Model evaluation and risk

Accuracy

How often the model gets the answer right. A model that’s 95% accurate gets the answer right 95% of the time.

Bias

When a model systematically gets certain categories of cases wrong. A hiring AI that’s accurate overall but tends to reject candidates from certain backgrounds has bias.

False positive / false negative

In classification systems, false positive means the system incorrectly predicted something as positive. False negative means it predicted something as negative when it was actually positive. Which one matters depends on context.

Robustness

How well a model works on data different from what it was trained on. Robust models work across different scenarios. Non-robust models break.

Interpretability

Whether you can understand why a model made a particular decision. “The model rejected this loan because [clear reason]” is interpretable. “The model rejected this loan because of complex parameter interactions” is not.

## Deployment concepts

Production

When a system goes live and affects real business decisions. Pilot is testing. Production is the system actually running and mattering.

Inference

Using a trained model to make predictions on new data.

Latency

How long it takes for a system to respond. For some uses (background processing), latency doesn’t matter. For others (real-time interaction), it does.

Scalability

Whether a system can handle increased volume. A model that works with 1,000 predictions per day but breaks at 100,000 isn’t scalable.

## Governance and risk

Governance

The policies and decision-making processes that manage AI risk. Who decides what systems are allowed? Who’s accountable?

Explainability

The ability to explain how an AI system works at a level stakeholders can understand.

Fairness

Ensuring AI systems don’t discriminate or unfairly advantage some groups.

Transparency

Being clear about when and how AI is being used. Telling customers “this response was generated by AI” versus not disclosing it.

Accountability

Clear responsibility for AI system outcomes.

Risk management

Identifying what could go wrong with an AI system and putting controls in place.

## Emerging concepts in 2026

Multi-modal AI

Models that work across different types of data—text, images, video, audio. More capable but also more complex.

Few-shot learning

A model learning from very few examples. Reduces the data burden.

Retrieval-augmented generation (RAG)

A generative AI system that retrieves relevant information from a database before generating a response. Reduces hallucinations and keeps information current.

Transfer learning

Taking a model trained for one task and adapting it for a different task. Faster and more efficient than training from scratch.

Agentic workflows

Systems where AI agents can plan and execute sequences of actions toward a goal. Emerging and powerful.

## Quick reference: the terms you’ll hear most often

–

AI / AI system:

Software that learns from data and performs tasks or generates content
–

Generative AI:

Systems that generate text, images, code based on prompts
–

LLM:

Large language model trained on vast text datasets
–

Machine learning:

AI that learns from data rather than being explicitly programmed
–

Hallucination:

When AI generates false but confident-sounding outputs
–

Bias:

When AI systematically gets certain categories wrong
–

Fine-tuning:

Improving an AI system by training it on specific data
–

Accuracy:

How often the model gets the answer right
–

Production:

When a system is live and affecting real business decisions
–

Inference:

Using a trained model to make predictions
–

Agentic AI:

Systems that plan and take autonomous actions
–

Prompt engineering:

Writing better inputs to get better outputs

## FAQ

What’s the difference between AI and machine learning?

A: AI is the broad field of systems that can learn and perform intelligent tasks. Machine learning is a specific approach where systems learn from data. All machine learning is AI, but not all AI is machine learning.

Do I need to understand how LLMs work to use them?

A: No. You need to understand what they can and can’t do. How they work internally is less important.

What’s the difference between accuracy and fairness?

A: A system can be accurate overall but unfair (accurate for some groups, inaccurate for others). Fairness asks whether the system treats all groups fairly.

If a system has hallucinations, should we not use it?

A: Not necessarily. It depends on the use case. For drafting copy, occasional hallucinations are acceptable if humans verify output. For medical diagnosis or loan decisions, hallucinations are unacceptable.