Data and AI glossary: key terms in plain English

This glossary defines, in a sentence or two, the terms used in Maketools games and on the pages of this section. Where a game trains the concept, the link leads to its guide.

Artificial intelligence

AI system

Software that, from the input it receives, produces outputs such as predictions, content, recommendations or decisions, with some degree of autonomy. It is the central concept of the EU AI Act (see Article 4 of the AI Act).

Generative AI

AI that produces new content (text, images, audio, code) from an instruction. Chat assistants are the most familiar example. Knowing when it is the right tool, and when it is not, is what AI or Not AI? is about.

Large language model (LLM)

An AI model trained on large amounts of text to predict the most likely continuation of a text. That is what lets it write, summarise and answer, and also what explains its hallucinations: it produces what sounds plausible, not what has been checked.

Token

A unit of text (a word, part of a word, a symbol) processed by a language model. Providers often bill usage by the number of tokens. In AI or Not AI?, every call to AI spends a token budget.

Writing a request

Prompt

The instruction given to a generative AI. A good prompt states the role, the task, the context, the constraints, and sometimes an example and the expected format. Practised in Prompt Quest.

Context (in a prompt)

The information that frames a request: who is speaking, to whom, for what purpose, with which data. Without it, AI answers generically. The "giving context" skill in Prompt Quest.

Constraints (in a prompt)

Limits placed on the answer: length, tone, format, sources to use, what must not be made up. The "setting constraints" skill in Prompt Quest.

Examples (few-shot prompting)

One or more samples of the expected answer, included in the prompt. They often guide better than a long instruction. The "using examples" skill in Prompt Quest.

Checking an answer

Hallucination

An AI answer that presents an invented fact as true: a source that does not exist, a made-up figure, an imaginary quote. The better written it is, the more misleading. Players learn to spot them in Hallucination Hunter.

Bias

A systematic distortion in a result: a stereotype, a sweeping generalisation, a single viewpoint presented as neutral. Bias can come from training data or from how the request is worded. The "detecting bias" skill in Hallucination Hunter.

Verification

Checking an answer against a reliable source before using it. A plausible answer that cannot be checked should be treated as "to be checked", not as true. The "knowing when to verify" skill in Hallucination Hunter and Prompt Quest.

Human oversight

A person who can understand, monitor and correct what an AI system produces, and who remains responsible for the decision. The "keeping a human accountable" and "reviewing before publishing" skills in AI or Not AI?.

Script (deterministic automation)

A program or formula that always applies the same rule: a calculation, a sort, a conversion. For a mechanical task, it is more reliable and cheaper than an AI model. The "choosing a script over a model" skill in AI or Not AI?.

Data

Data quality

How fit data is for its intended use: complete, accurate, consistent, free of duplicates. Poor quality data skews an analysis or what a model learns. The theme of Data Cleaning Rush and of the data literacy page.

Duplicate

The same record appearing more than once, sometimes with different capitalisation or spelling. It inflates totals. The "spotting duplicates" skill in Data Cleaning Rush.

Missing value

An empty cell where a value is expected. It is not a zero: you must decide whether to fill it or set the row aside. The "handling missing values" skill in Data Cleaning Rush.

Format

How a value is written: date, unit, decimal separator, capitalisation. Mixed formats in one column make calculations wrong. The "fixing formats" skill in Data Cleaning Rush.

Outlier

An impossible or highly improbable value (an age of 250). It should be flagged; a rare but legitimate value should be kept. The "spotting outliers" skill in Data Cleaning Rush.

Training data

The data an AI model learns from. Its errors and biases carry through to the model's results, as every mistake in Data Cleaning Rush shows.

Data protection

Personal data

Any information relating to an identified or identifiable person, directly (name, email) or indirectly (an identifier, a combination of attributes). It should never go into an unapproved AI tool. The "protecting personal data" skill in AI or Not AI?.

Pseudonymisation

Replacing a person's identity with a pseudonym, so that they cannot be identified without extra information kept separately. Pseudonymised data is still personal data. Maketools reports are pseudonymised by default (see player privacy).

Anonymisation

Processing that makes identifying a person irreversibly impossible. Truly anonymised data is no longer personal data under the GDPR.

k-anonymity

A rule that shows a statistic only if it covers at least k people, so that an average never reveals one individual's result. In Maketools, a report filtered by group requires at least 5 players (see player privacy).

Regulation and training

AI literacy

The skills, knowledge and understanding needed to use AI systems in an informed way and to grasp their opportunities and risks. See AI literacy.

Data literacy

The ability to read, understand, use and question data at work. See data literacy.

Deployer

Under the AI Act, a person or organisation using an AI system under its own authority in a professional context. A company that gives its staff an AI assistant is one. Article 4 applies to deployers as well as providers (see Article 4 of the AI Act).

Serious game

A game designed to teach: realistic situations, a decision, immediate feedback, measurement. See serious games at work.

SCORM

A standard for exchanging data between a learning content package and an LMS (learning management system). A Maketools SCORM 1.2 package sends the learner's score and status back to the LMS (see LMS exports).

xAPI

A standard for sending "statements" ("this learner completed this activity with this score") to an LRS, the store for learning records. Maketools sends one statement per completed game, and optionally one per answer (see LMS exports).

Going further

To put these concepts into practice, start with a game from the catalogue, no account needed, or read where to start.