Skip to content
LIRGOoperational intelligence

Glossary

The words you’ll come across on this site and in the papers. Two or three sentences each, with a made-up example from everyday work.

Agent
An artificial intelligence you hand a whole task to, and which chooses for itself the steps to finish it. It might read a customer’s email, check the stock and prepare the work order. It can stop wherever you need it to, so that a person makes the decision.
AI Act
The European law on artificial intelligence, Regulation (EU) 2024/1689, which has applied since 2 August 2026, with some exceptions: the rules for high-risk uses, for instance, start on 2 December 2027 or 2 August 2028. The riskier the use, the more rules it has to follow. If you put a chatbot on your workshop’s website, for instance, people writing to it must be told they’re talking to a machine, unless that’s obvious. The law asks this of whoever supplies the chatbot.
Algorithm
A set of precise rules that, followed in order, lead to a result. One example is the sum that tells you how much of an item to reorder, based on what went out over the last three months. It isn’t necessarily artificial intelligence: a formula in an Excel sheet is an algorithm too.
API
The door one piece of software opens to another, under precise rules, to hand over data or carry out a task. Your management software might have an API that says how many of an item you’ve got. The agent asks there, and nobody has to copy anything out.
Artificial intelligence (AI)
A machine that works out, from what it receives, how to reach a result: a forecast, a text, a recommendation, a decision. That’s the European law’s definition, in plain words. Given ‘same as last time but smaller’, for instance, it works out that a measurement is missing.
Automation
When a machine or piece of software does a job on its own, following rules a person wrote in advance. For example: if an item falls below 100 units, an order goes to the supplier. If a person wrote all the rules, European law doesn’t count it as artificial intelligence.
Business management software
The software where a business keeps its orders, stock, invoices and accounts. It’s where an order gets entered and where the parts used come off the stock count. In the papers, the fuller versions go by the initials ERP.
Chatbot
A piece of software that answers people who write or talk to it. Older ones follow scripted replies, newer ones use a model: ChatGPT is one of these. On a print shop’s website, for instance, it might answer customers’ questions about formats and turnaround times.
Cloud
Using computers, storage and software over the internet, kept somewhere else rather than on your premises. One example is management software you open in a browser, with your data on a supplier’s servers. Before you put your data there, ask in writing where it’s kept and who can see it.
Data
A piece of recorded information you can read and compare: the date of an order, the parts rejected on a shift, the oven temperature. What an operator knows by heart isn’t data yet: it becomes data once it’s written down. Since 12 September 2025 a European law, the Data Act, has let anyone who owns, rents or leases a connected machine access the data they generate by using it.
Generative AI
Artificial intelligence that creates new content: text, images, sound, video. It might draft a reply to a customer or mock up a label. The images and videos on this site were made this way.
Machine learning
How a system learns from examples instead of following rules written by hand. From the timings of thousands of past jobs, for instance, it can estimate how long the next one will take, if it’s similar. Strictly speaking it’s one branch of artificial intelligence, though the two terms are often used as if they meant the same thing.
Model
The part of artificial intelligence that takes something in and gives a result back. It might take in the text of an email and give back the item, the quantity and the delivery date. GPT and Gemini, which you read about in the papers, are language models: built to work on language, and now able to read images too.
Model hallucination
When a model writes something false that sounds true. It might quote an item code that doesn’t exist in your stock. You can reduce it but not remove it altogether, which is why anything important gets checked against the real data.
Personal data
Information that identifies a person, not a company, or lets you trace them, even indirectly. The name of the clerk who sends you an order is personal data, and so is a van’s number plate; an item code isn’t. It’s protected by the GDPR, the European data protection regulation.
Prompt
The text you give a model that’s already been trained, to tell it what to do: a question, an instruction, some examples. For instance: ‘From this email, pull out the item, the quantity and the delivery date.’ The model doesn’t change: what changes is what you ask of it.
Sensor
A device that measures a physical property, such as speed, temperature or flow, and turns it into an electrical signal. On a bakery’s oven, for instance, it records the temperature minute by minute. That way the data’s there even when nobody’s walking past to read it.
Training
The stage in which a model reads a huge number of examples and adjusts itself little by little. A model learning to spot faulty welds, for instance, is shown thousands of photos of good and rejected welds. Before you trust a service with your data, ask in writing whether it uses that data for training: it depends on the service and the contract.

Sources

  1. European Commission, AI Act page: risk levels, chatbots, application dates (updated 19 October 2026, read on 2 November 2026). digital-strategy.ec.europa.eu
  2. European Commission, AI Act Service Desk: Article 3, definition of an AI system (consolidated text as of 27 July 2026, read on 2 November 2026). ai-act-service-desk.ec.europa.eu
  3. European Commission, AI Act Service Desk: Recital 12, rules written only by people and the capability to infer (read on 2 November 2026). ai-act-service-desk.ec.europa.eu
  4. European Commission, AI Act Service Desk: Article 50, informing people who interact with an AI system (read on 2 November 2026). ai-act-service-desk.ec.europa.eu
  5. European Commission, AI Act Service Desk: Article 113, application dates (read on 2 November 2026). ai-act-service-desk.ec.europa.eu
  6. European Commission, Data Act explained: who can access data from connected products, industrial machinery, applying from 12 September 2025 (read on 2 November 2026). digital-strategy.ec.europa.eu
  7. Garante per la protezione dei dati personali (Italian data protection authority), Cosa intendiamo per dati personali (read on 2 November 2026). www.garanteprivacy.it
  8. Garante per la protezione dei dati personali (Italian data protection authority), statement of 31 March 2023 su ChatGPT (read on 2 November 2026). www.garanteprivacy.it
  9. Google, Machine Learning Glossary: machine learning, model, training, generative AI, large language model, prompt, prompt-based learning, hallucination, artificial intelligence (read on 2 November 2026). developers.google.com
  10. NIST, AI 600-1, Generative AI Profile: confabulation, also called hallucination (read on 2 November 2026). nvlpubs.nist.gov
  11. NIST, CSRC Glossary: API (read on 2 November 2026). csrc.nist.gov
  12. NIST, CSRC Glossary: sensor, definition from SP 800-82r3 (read on 2 November 2026). csrc.nist.gov
  13. NIST, CSRC Glossary: algorithm (read on 2 November 2026). csrc.nist.gov
  14. NIST, SP 800-145: definition of cloud computing (read on 2 November 2026). csrc.nist.gov
  15. Anthropic, Building effective agents: definition of an agent, pauses for human control (read on 2 November 2026). www.anthropic.com
  16. Anthropic, Reduce hallucinations: they can be reduced, not eliminated altogether (read on 2 November 2026). platform.claude.com
  17. Anthropic Privacy Center, use of data for training in commercial products (read on 2 November 2026). privacy.claude.com
  18. Anthropic, Updates to consumer terms: use of data for training in plans for private individuals (read on 2 November 2026). www.anthropic.com
  19. IBM, What is a chatbot?: definition, rule-based chatbots and model-based chatbots (read on 2 November 2026). www.ibm.com
  20. IBM, What is ERP?: definition and activities covered (read on 2 November 2026). www.ibm.com