AI Glossary

Agenten-Orchestrierung

Agents and automation

Agent orchestration is the coordination of one or more AI agents, tools and process steps, deciding which component handles which task, in what order, and how results are combined and checked.

Agentische KI; Agentic AI

Agents and automation

Agentic AI describes AI systems with a degree of autonomy: they can break down goals, decide on steps, use tools and act with limited human direction.

KI-Einführung; KI-Adoption

Adoption and implementation

AI adoption is the process by which an organisation and its people take up AI and integrate it into everyday work in a way that is sustained, useful and responsible.

KI-Agent

Agents and automation

An AI agent is a system that uses an AI model to pursue a goal by planning steps, using tools such as search, databases or other software, and adjusting its actions based on the results.

KI-Governance

Responsible AI and governance

AI governance is the set of structures, roles, policies and processes an organisation uses to decide how AI is selected, developed, used and monitored responsibly.

KI-Kompetenz

Responsible AI and governance

AI literacy is the knowledge, skills and understanding people need to use AI systems appropriately, assess their outputs critically and understand their opportunities, limits and risks.

KI-Reifegrad

Adoption and implementation

AI maturity describes how advanced an organisation's capabilities are for using AI effectively and responsibly, across areas such as strategy, governance, data, technology, skills and culture.

KI-Risikobewertung

Responsible AI and governance

An AI risk assessment is a structured review of what could go wrong with an AI system, who could be harmed, how likely and serious that would be, and which measures reduce the risks.

KI-Anwendungsfall; KI-Use-Case

Adoption and implementation

An AI use case is a specific, defined situation in which AI is applied to a task or problem, describing who uses it, for what purpose, with which data and what outcome is expected.

KI-Workflow; KI-gestützter Arbeitsablauf

Agents and automation

An AI workflow is a defined sequence of steps in which AI performs one or more tasks, often combined with traditional software steps and human review.

Programmierschnittstelle; API

Architecture and integration

An API (Application Programming Interface) is a defined way for one software system to request data or services from another.

Künstliche Intelligenz (KI)

AI foundations

Artificial intelligence is the broad field of building computer systems that perform tasks we usually associate with human thinking, such as recognising patterns, understanding language, making predictions or generating content.

Prüfpfad; Audit-Trail; Protokollierung

Responsible AI and governance

An audit trail is a secure, time-stamped record of what happened in a system: who did what, when, with which inputs and outputs, and which decisions were made.

Authentifizierung

Architecture and integration

Authentication is the process of verifying who a user or system is, for example through a password, a smartcard, a one-time code or an electronic ID.

Autorisierung; Berechtigung; Zugriffssteuerung

Architecture and integration

Authorisation is the process of deciding what an authenticated user or system is allowed to access or do.

Backend

Architecture and integration

The backend is the part of a software system that runs on servers and is not directly visible to users, handling data storage, business logic, security checks and connections to other systems.

Verzerrung; Bias

Responsible AI and governance

In AI, bias means systematic patterns in a system's outputs that unfairly favour or disadvantage certain people or groups, or that consistently distort results.

Veränderungsmanagement; Change-Management

Adoption and implementation

Change management is the structured approach to helping people and organisations move from current ways of working to new ones, including communication, involvement, training, support and addressing concerns.

Cloud-KI; KI aus der Cloud

Architecture and integration

Cloud AI means using AI models and services that run on a provider's infrastructure and are accessed over the internet, usually through an API or web application.

Kontextfenster

AI foundations

The context window is the maximum amount of text, measured in tokens, that a language model can consider at one time, including the instructions, any documents provided and the conversation so far.

Datenklassifizierung; Schutzbedarfsfeststellung

Responsible AI and governance

Data classification is the practice of labelling information according to its sensitivity and protection needs, for example public, internal, confidential or restricted, so that appropriate rules apply.

Datenpipeline

Architecture and integration

A data pipeline is an automated sequence of steps that collects data from sources, cleans and transforms it, and delivers it to where it is needed, such as a database, analytics tool or AI system.

Datenreife; Datenbereitschaft (no single established term)

Adoption and implementation

Data readiness describes how prepared an organisation's data is for a specific AI use: whether the right data exists, is accessible, is of sufficient quality and structure, is legally usable, and is maintained.

Chunking; Aufteilung in Textabschnitte

Knowledge systems and RAG

Document chunking is the process of splitting long documents into smaller sections before they are turned into embeddings and stored for retrieval.

Embeddings; Vektorrepräsentationen

Knowledge systems and RAG

Embeddings are lists of numbers that represent the meaning of a piece of text, an image or other content, created by an AI model.

KI-Verordnung (Verordnung (EU) 2024/1689)

Responsible AI and governance

The EU AI Act is the European Union's regulation on artificial intelligence, which sets rules for AI systems based on their level of risk.

Erklärbarkeit

Responsible AI and governance

Explainability is the degree to which people can understand why an AI system produced a particular output or decision, in terms that are meaningful to them.

Feinabstimmung; Fine-Tuning

AI foundations

Fine-tuning is the process of further training an existing model on a smaller, specialised dataset so that it behaves differently, for example using specific terminology, a particular writing style or a fixed output format.

Basismodell; in the EU AI Act: KI-Modell mit allgemeinem Verwendungszweck (general-purpose AI model)

AI foundations

A foundation model is a large AI model trained on broad data so that it can be adapted to many different tasks, rather than being built for one purpose.

Frontend; Benutzeroberfläche

Architecture and integration

The frontend is the part of a software system that users see and interact with, such as a website, app or chat window.

Datenschutz-Grundverordnung (DSGVO)

Responsible AI and governance

The General Data Protection Regulation (in German, Datenschutz-Grundverordnung, DSGVO) is the EU law that governs how personal data is collected, used, stored and shared.

Generative KI

AI foundations

Generative AI refers to AI systems that create new content, such as text, images, audio, code or video, based on patterns learned from large amounts of existing data.

Guardrails; Schutzmechanismen; Leitplanken

Agents and automation

Guardrails are technical and procedural controls that limit what an AI system can receive, produce or do, such as filtering harmful content, blocking personal data, restricting topics or requiring confirmation for certain actions.

Halluzination; Konfabulation

AI foundations

A hallucination is when an AI model produces information that sounds plausible but is false, unsupported or invented, such as a non-existent source, a wrong date or a fabricated quote.

Handschriftenerkennung

Knowledge systems and RAG

HTR is technology that converts images of handwritten documents into machine-readable text.

Menschliche Aufsicht

Agents and automation

Human oversight is the ability of people to monitor, understand, intervene in and, if necessary, stop an AI system, both in individual cases and at the level of the whole system.

Mensch-KI-Zusammenarbeit

Adoption and implementation

Human-AI collaboration describes how people and AI systems divide and share work, so that each contributes what they do best and the combined result is better than either alone.

Human-in-the-Loop; menschliche Beteiligung im Entscheidungsprozess

Agents and automation

Human-in-the-loop describes a system design in which a person reviews, approves, corrects or completes AI outputs at defined points before they take effect.

Stichwortsuche; Volltextsuche

Knowledge systems and RAG

Keyword search finds content that contains the words in a query, often with techniques that account for word forms and how important a word is.

Wissensdatenbank; Wissensbasis

Knowledge systems and RAG

A knowledge base is an organised collection of information, such as articles, policies, FAQs and procedures, that people or systems can search and use.

Wissensgraph

Knowledge systems and RAG

A knowledge graph represents information as a network of entities, such as people, places, organisations and objects, and the relationships between them.

Großes Sprachmodell

AI foundations

A large language model is an AI model trained on very large amounts of text to predict and generate language.

Maschinelles Lernen

AI foundations

Machine learning is a way of building AI in which a system learns patterns from examples instead of following rules written by hand.

Metadaten

Knowledge systems and RAG

Metadata is data that describes other data, such as a document's title, author, date, version, language, status, access level or subject.

Modellevaluierung; Modellbewertung

Responsible AI and governance

Model evaluation is the systematic testing of an AI model or system to measure how well it performs on the tasks it is meant to do, using defined test cases, metrics and quality criteria.

Modell-Hosting; Modellbereitstellung

Architecture and integration

Model hosting describes where and how an AI model is run so that applications can use it, including the servers, operating responsibilities and the interface through which requests are sent.

Inferenz (Modellinferenz)

AI foundations

Inference is the stage when a trained AI model is used to produce an output, such as an answer, a classification or a prediction, from new input.

Modellüberwachung; Modell-Monitoring

Adoption and implementation

Model monitoring is the ongoing tracking of an AI system's performance, behaviour and usage after it goes live, to detect problems such as declining accuracy, changing input data, errors or misuse.

Multi-Agenten-System

Agents and automation

A multi-agent system uses several AI agents, each with a specific role, that work together on a task, for example one gathering information, one drafting and one checking.

On-Premises-KI; lokal betriebene KI

Architecture and integration

On-premises AI means running AI models on hardware that the organisation owns or directly controls, such as its own data centre or a dedicated government computing centre.

Open-Weight-Modell; Modell mit offenen Gewichten

Architecture and integration

An open-weight model is an AI model whose trained parameters, called weights, are published so that others can download, run and often adapt it.

Optische Zeichenerkennung; Texterkennung

Knowledge systems and RAG

OCR is technology that converts images of printed or typed text, such as scans and photos, into machine-readable text.

Pilotprojekt; Pilotphase

Adoption and implementation

A pilot is a limited real-world deployment of a solution with actual users, data and processes, designed to test whether it works in practice and to learn before wider rollout.

Datenschutz durch Technikgestaltung (Privacy by Design)

Responsible AI and governance

Privacy by design means building data protection into a system from the start, through its architecture, defaults and processes, rather than adding it at the end.

Prompt; Eingabeaufforderung

AI foundations

A prompt is the input given to a generative AI model to tell it what to do, such as a question, instruction, example or document.

Prompt-Engineering

AI foundations

Prompt engineering is the practice of designing and testing prompts so that an AI model produces more useful, reliable and consistent outputs.

Machbarkeitsnachweis; Proof of Concept (PoC)

Adoption and implementation

A proof of concept is a small, time-limited experiment to test whether an idea is technically possible and promising, usually with sample data and outside normal operations.

Abruf; Retrieval

Knowledge systems and RAG

Retrieval is the step in which a system searches its sources and selects the information most relevant to a user's question.

Retrieval-Augmented Generation; abrufgestützte Generierung

Knowledge systems and RAG

RAG is a technique that allows an AI system to retrieve relevant information from external sources and provide that information to a language model when generating a response.

Semantische Suche

Knowledge systems and RAG

Semantic search finds content based on meaning rather than exact word matches, typically by comparing embeddings.

Erfolgskennzahlen; KPIs

Adoption and implementation

Success metrics are the measures an organisation uses to decide whether an AI initiative is achieving its intended outcomes, such as time saved, quality, user satisfaction, error rates or service improvements.

Systemintegration

Architecture and integration

System integration is the work of connecting an AI tool with the existing systems, data and processes an organisation already uses, so that information flows between them reliably and securely.

Technische Machbarkeit

Adoption and implementation

Technical feasibility is the assessment of whether a proposed AI solution can realistically be built and operated with the available technology, data, systems, skills, budget and constraints.

Tokens

AI foundations

Tokens are the small units of text that a language model reads and writes, often whole words, parts of words or punctuation marks.

Tool-Aufruf; Funktionsaufruf

Agents and automation

Tool calling is a capability that lets an AI model request the use of external functions, such as searching a database, checking a calendar or performing a calculation, by producing a structured request.

Transformer (Transformer-Architektur)

AI foundations

A transformer is a type of neural network architecture, introduced in 2017, that underlies most modern large language models.

Transparenz

Responsible AI and governance

Transparency in AI means being open about when, where and how AI is used, what it is designed to do, what data it relies on, and what its limitations are.

Vektordatenbank

Knowledge systems and RAG

A vector database is a database designed to store embeddings and quickly find the ones most similar to a given query.

Workflow-Automatisierung; Prozessautomatisierung

Agents and automation

Workflow automation uses software to carry out repetitive process steps according to predefined rules, such as moving data between systems, sending notifications or creating tasks.