Lernen Sie die Sprache der KI
KI-Glossar
Ein umfassendes Wörterbuch für Begriffe der künstlichen Intelligenz und des maschinellen Lernens.
- Gesamte Begriffe
- 184
- Kategorien
- 6
KI-Glossar
A
- Bereitstellung
A/B Testing (ML)
A controlled experiment that compares the performance of two or more model versions by routing live traffic between them and measuring business and technical metrics. Where live experimentation is appropriate, A/B testing can help compare outcomes before broader rollout.
- Governance
Accountability (AI)
The principle that individuals and organisations involved in AI development and deployment can be held responsible for the outcomes their systems produce. The EU AI Act establishes clear accountability chains between providers, importers, distributors, and deployers.
- Grundlagen
Agentic AI
AI systems that coordinate multi-step tasks using models, tools and feedback, with a degree of autonomy defined by the application. Specify their permitted actions, stopping conditions and human approval requirements. Agentic behaviour alone does not establish reliability, safety or authority to act.
- Governance
AI Act Annex III
The list of high-risk AI application areas appended to the EU AI Act, including biometric systems, critical infrastructure management, education, employment, essential private/public services, law enforcement, migration, and administration of justice. Classification must be assessed for the specific intended use under Article 6, including its conditions and exceptions; obligations then depend on the organisation's role, such as provider or deployer.
- Governance
AI Act Article 6
The EU AI Act provision setting high-risk classification rules. It covers specified Annex I products and safety components subject to the relevant third-party conformity assessment conditions, and Annex III uses subject to the Article 6 rules and exceptions. Classification depends on the system’s intended purpose and applicable conditions, not simply its technology or industry.
- Architektur
AI Agent
An AI system that observes inputs and selects actions toward a goal, potentially using tools or multiple steps, within defined permissions. Enterprise agents need controls appropriate to their tasks, including authorised tool access, stop conditions and human review where needed. Applicable EU AI Act duties depend on the system’s intended use, classification and the organisation’s role.
- Grundlagen
AI Alignment
The research and engineering discipline of ensuring that AI systems pursue goals and exhibit behaviours that match human intentions and values. Misalignment risks range from models following instructions too literally to more speculative long-term risks discussed in AI safety literature.
- Business
AI Architect
A technical architect who designs the end-to-end AI system landscape—data pipelines, model training infrastructure, serving platforms, integration patterns, and governance tooling. AI architects ensure that individual AI solutions compose into a coherent, scalable enterprise capability.
- Governance
AI Auditability
The ability to examine an AI system’s behaviour, decisions, and data flows using relevant evidence. Logging, versioned model artefacts and documented decisions can support auditability. The evidence and retention required depend on the system, its risks and applicable obligations; a generic audit trail is not, by itself, proof of EU AI Act conformity.
- Governance
AI Bias
Systematic errors in AI systems that create unfair outcomes for certain groups, often arising from unrepresentative training data, flawed labelling, or problematic model design. Identifying and mitigating bias is a legal requirement for high-risk AI systems under the EU AI Act.
- Business
AI Business Case
A structured argument for an AI investment that quantifies expected benefits (revenue uplift, cost reduction, risk mitigation), costs (build, run, compliance), risks, and strategic alignment. A robust AI business case is the gateway to executive and board approval.
- Business
AI Center of Excellence
An internal team that provides AI expertise, shared tooling, governance guidance and support to business units. A centre of excellence can coordinate reuse and review, but its mandate should define responsibilities and decision rights. It does not replace accountable system owners or demonstrate compliance for every initiative.
- Business
AI Change Management
Preparing people and operating practices for the introduction or change of AI systems. Work can include role design, training, consultation, workflow changes, adoption support and feedback. Assess its effects with task and organisational evidence alongside technical quality; change management does not replace engineering or governance.
- Business
AI Cost Reduction
Using AI with the aim of lowering operational costs through task automation, process optimisation, predictive maintenance or resource allocation. Savings should be assessed against an agreed baseline and include implementation, operation, oversight and exception-handling costs; automation alone does not demonstrate a net cost reduction.
- Governance
AI Documentation
Comprehensive records describing an AI system's design, training data, performance characteristics, limitations, and intended use. The EU AI Act distinguishes between technical documentation (for authorities and notified bodies) and instructions for use (for deployers).
- Business
AI Due Diligence
A thorough assessment of an AI system or AI-enabled company's technical claims, data practices, compliance posture, model performance, and operational risks—typically conducted in M&A, procurement, or partnership contexts. AI due diligence is a growing practice as AI becomes a critical business asset.
- Governance
AI Ethics
The application of moral principles—fairness, accountability, transparency, non-maleficence, and human dignity—to the design, development, and deployment of AI systems. AI ethics provides the normative foundation that regulations like the EU AI Act translate into enforceable obligations.
- Business
AI for Finance (FinAI)
The application of AI to financial services, including fraud detection, credit underwriting, algorithmic trading, regulatory reporting, and customer advisory. Applicable AI Act requirements depend on the intended use and classification, alongside relevant financial-services and data-protection rules. High-risk status should be checked for the specific system rather than inferred from the financial sector alone.
- Business
AI for Healthcare
AI applications in clinical and administrative healthcare, including medical image analysis, clinical decision support, drug discovery, and patient flow optimisation. EU obligations depend on intended purpose: qualifying medical-device software may fall under the MDR or IVDR, while AI Act high-risk classification has its own conditions. Administrative use alone does not establish medical-device status or a CE-marking requirement.
- Business
AI for Legal
AI tools applied to legal work—contract analysis, due diligence, legal research, compliance monitoring, and document generation. Legal AI can dramatically reduce the cost of routine legal work but requires careful hallucination controls given the consequences of factual errors in legal contexts.
- Business
AI for Logistics
AI applications in supply chain and logistics, including route optimisation, demand forecasting, warehouse automation, and predictive shipment tracking. A business case should compare the proposed system with the existing process using relevant service, cost and reliability measures. Available data or decision volume alone does not establish ROI.
- Business
AI for Manufacturing
The application of AI, including computer vision, predictive maintenance, process optimisation, and generative design, to manufacturing operations. Potential outcomes include fewer defects, improved throughput or reduced downtime. Value must be measured against an operational baseline, including integration costs, failure cases, human work and safety constraints.
- Governance
AI Governance
The framework of policies, processes, roles, and controls that ensure AI systems are developed, deployed, and used responsibly and in compliance with applicable law. AI governance spans ethics, bias, privacy, security, and regulatory compliance across the model lifecycle.
- Business
AI Governance Committee
A cross-functional body—typically including legal, compliance, risk, IT, and business leaders—responsible for approving AI initiatives, reviewing ethical risks, and ensuring regulatory compliance. The committee is the human governance structure that operationalises AI policy.
- Governance
AI Incident Reporting
The obligation under the EU AI Act for providers of high-risk AI systems to report serious incidents—where the AI causes harm to health, safety, or fundamental rights—to national market surveillance authorities within defined timeframes.
- Governance
AI Liability Directive
A 2022 EU proposal to adapt non-contractual civil liability rules for AI-related harm, including evidence disclosure and presumptions of causation. The Commission withdrew the proposal in October 2025; it did not become an enacted AI Liability Directive. Liability for a particular AI system must be assessed under the applicable legal framework.
- Business
AI Maturity
A measure of an organisation's capability to successfully implement and scale AI initiatives, typically assessed across dimensions like data quality, talent, governance, technology, and executive sponsorship. Maturity models help organisations benchmark themselves and prioritise investments.
- Business
AI Maturity Model
A structured framework that describes progressive levels of AI capability—typically from initial experimentation through managed deployment to optimised, organisation-wide integration. Maturity models provide a common language for assessing current state and planning improvement.
- Governance
AI Notified Body
An independent conformity assessment body designated under the applicable EU framework. Whether a high-risk AI system needs a notified body depends on its classification and assessment route. Article 43 provides internal control for Annex III points 2–8 when the system is not also covered by Section A of Annex I; biometric and regulated-product routes have different conditions. A sector label alone does not determine the procedure.
- Bereitstellung
AI Orchestration
The coordination of multiple AI models, tools, APIs, and data sources within complex automated pipelines or agentic workflows. Orchestration frameworks like LangChain, LlamaIndex, and Temporal manage state, retries, and branching logic across multi-step AI processes.
- Business
AI Pilot
A time-boxed, scope-limited test of an AI solution designed to examine assumptions, measure relevant outcomes, and identify operational challenges before a wider deployment decision. Duration, permissions, success criteria, stop conditions and ownership should be set for the specific pilot; completing it does not automatically authorise production use.
- Business
AI Product Manager
A product manager specialising in AI-powered products, responsible for translating business needs into AI system requirements, managing stakeholder expectations around probabilistic outputs, and navigating data, model, and UX tradeoffs unique to AI products.
- Business
AI Readiness Assessment
A structured evaluation of an organisation's ability to adopt and scale AI, examining data quality, infrastructure, talent, processes, governance, and culture. Readiness assessments identify gaps and form the evidence base for an AI roadmap and investment case.
- Governance
AI Regulatory Sandbox
A controlled environment, established under the EU AI Act, that allows companies to develop and test innovative AI systems under regulatory supervision before full market release. Sandboxes give startups and SMEs access to regulators and reduce compliance uncertainty during development.
- Governance
AI Risk Management
A systematic process for identifying, assessing, and mitigating risks associated with AI systems throughout their lifecycle. The EU AI Act requires high-risk AI providers to implement and document a risk management system; ISO 42001 and NIST AI RMF provide complementary frameworks.
- Business
AI Roadmap
A time-sequenced plan that maps AI initiatives to business priorities, dependencies, resource requirements, and expected outcomes. A good roadmap balances quick wins that build momentum with strategic bets that transform core business processes.
- Business
AI ROI
The return on investment from an AI initiative, measured as the ratio of net benefit to total investment over a defined period. AI ROI is notoriously hard to quantify upfront but is increasingly demanded by CFOs; it spans hard savings (labour automation), soft gains (decision quality), and risk-adjusted value.
- Grundlagen
AI Safety
Research and engineering concerned with identifying and reducing potential harm associated with AI systems, including failures, misuse and interaction with people or physical processes. Safety depends on the application and operating conditions, with appropriate hazard analysis, controls and evidence. Regulatory classification or an AI compliance review alone does not establish the safety of a complete system.
- Business
AI Scaling
The process of expanding a successful AI pilot to broader organisational adoption—handling increased data volumes, additional use cases, more users, and tighter integration with business processes. Scaling typically exposes infrastructure, data governance, and change management challenges invisible at pilot scale.
- Business
AI Steering Committee
A senior leadership body responsible for setting AI strategic priorities, allocating resources, and monitoring the portfolio of AI investments. The steering committee provides executive sponsorship and escalation paths for significant AI decisions.
- Business
AI Strategy
A comprehensive plan for how an organisation will adopt, develop, and deploy AI to achieve business objectives—encompassing use-case prioritisation, build-vs-buy decisions, data infrastructure, talent, governance, and compliance. An effective AI strategy aligns technology investment with measurable business outcomes.
- Governance
AI System
As defined by the EU AI Act: a machine-based system that, for a given set of objectives, infers from the inputs it receives how to generate outputs such as predictions, content, recommendations, or decisions that can influence real or virtual environments.
- Business
AI TCO (Total Cost of Ownership)
The comprehensive cost of an AI system over its full lifecycle, including model development, data preparation, infrastructure, licensing, monitoring, compliance, maintenance, and eventual decommissioning. TCO analysis compares these costs for the expected workload and operating model; their relative importance varies by system and deployment.
- Business
AI Tech Lead
A senior technical leader who guides architecture decisions, engineering best practices, and model development for AI products. The AI tech lead bridges research and production, ensuring that promising models are transformed into reliable, maintainable, and compliant systems.
- Governance
AI Transparency
The principle that AI systems, their capabilities, limitations, and decision-making processes should be understandable to affected users, deployers, and regulators. The EU AI Act mandates transparency obligations at multiple risk tiers, from simple chatbot disclosure to full technical documentation for high-risk systems.
- Business
AI Upskilling
Structured training that builds AI literacy and practical skills, from executive awareness to hands-on work for analysts, engineers and business users. Training should reflect people’s responsibilities and the systems they use, alongside the processes, support and safeguards needed for adoption.
- Business
AI Vendor Selection
The process of evaluating and choosing AI technology vendors across capability fit, security posture, data handling, applicable compliance requirements, commercial terms and lock-in risk. Selection should include evidence for the intended use and a workable exit plan; it cannot guarantee that future replacement or integration costs will be avoided.
- Governance
AI Watermarking
Techniques that embed signals in AI-generated content to help identify its origin. Watermarking is one possible approach to synthetic-content marking. Article 50 of the EU AI Act sets machine-readable marking and detectability duties for covered providers, with conditions and exceptions; it does not prescribe a single watermarking method or guarantee reliable detection in every case.
- Business
AI-First
An organisational philosophy where AI capabilities are considered from the outset when designing products, processes, and business models, rather than applied as an afterthought. AI-first organisations embed machine intelligence into their core value proposition.
- Governance
Algorithmic Bias
Systematic and repeatable errors in algorithmic outputs that create unfair advantages or disadvantages for individuals or groups based on protected characteristics. Algorithmic bias can arise from biased training data, proxy variables, or flawed evaluation metrics.
- Grundlagen
Artificial General Intelligence (AGI)
A hypothetical level of AI capability matching or surpassing human cognitive performance across all domains, not just narrow tasks. AGI remains a research objective; current enterprise AI systems are all narrow AI. The EU AI Act does not yet regulate AGI specifically.
- Grundlagen
Artificial Intelligence (AI)
A field of computer science focused on creating systems that can perform tasks typically requiring human intelligence. This includes learning from experience, understanding language, recognizing patterns, solving problems, and making decisions.
- Techniken
Attention Mechanism
A neural network component that allows a model to dynamically focus on the most relevant parts of its input when producing each output element. Self-attention is the core innovation of the Transformer architecture and is responsible for LLMs' ability to handle long, complex contexts.
- Business
Augmented Intelligence
An approach to designing AI around support for human capabilities and decisions. People and systems receive defined responsibilities, with appropriate review and escalation. Assess effects on task quality, workload, autonomy and acceptance in the intended setting; the label does not establish improved performance or regulatory approval.
- Architektur
Autoencoder
A neural network trained to compress data into a lower-dimensional latent space and reconstruct it. Autoencoders are used for anomaly detection, dimensionality reduction, and as components in generative models like VAEs.
- Grundlagen
Autonomous Systems
Systems that perform specified tasks with varying degrees of human supervision or intervention. Capabilities and operating limits depend on the design and environment. EU AI Act high-risk classification follows Article 6 and the relevant product or intended-use criteria, including applicable exceptions; autonomy or sector membership alone does not determine that classification.
B
- Bereitstellung
Batch Inference
Running a model over a collection of inputs in a batch job rather than responding to each request immediately. It can suit report generation, bulk document classification or periodic scoring. Cost and latency depend on workload, infrastructure and scheduling; compare them with the required real-time service level.
- Architektur
Batch Normalization
A technique that normalises activations across the batch dimension during training, accelerating convergence and reducing sensitivity to hyperparameters. Though largely superseded by layer norm in transformers, batch norm remains prevalent in convolutional vision architectures.
- Architektur
BERT
Bidirectional Encoder Representations from Transformers—Google's 2018 encoder-only transformer model that set new benchmarks across NLP tasks. BERT-family models remain widely used for classification, named-entity recognition, and information retrieval in enterprise settings.
- Business
Build vs Buy (AI)
The strategic decision of whether to develop bespoke AI models and infrastructure in-house or procure commercial AI products and APIs. The decision hinges on competitive differentiation, data sensitivity, required customisation, cost, speed to value, and in-house talent availability.
C
- Bereitstellung
Canary Deployment
A progressive rollout strategy that routes a small percentage of production traffic to a new model version while the remainder continues to receive the old version. Canary deployments limit blast radius if the new model underperforms or causes regressions.
- Governance
CE Marking (AI)
The European conformity mark that high-risk AI systems must bear before entering the EU market, indicating compliance with applicable EU regulations including the AI Act. The CE mark serves as a declaration that the system meets safety, health, and environmental protection standards.
- Techniken
Chain-of-Thought (CoT)
A prompting technique that asks an LLM to generate intermediate steps before a final answer. It can help on some multi-step tasks, but usefulness depends on the model and evaluation. Generated reasoning may be incorrect or unfaithful and should not be treated as a verified account of the model’s internal process.
- Business
Chief AI Officer (CAIO)
An executive responsible for an organisation’s AI strategy, governance and coordination of AI initiatives with business priorities. The mandate should define decision rights, accountability and how value and risk will be assessed; the title alone does not establish those responsibilities or guarantee business outcomes.
- Grundlagen
Computer Vision
A field of AI that enables machines to interpret and understand visual information from the world—images, video, and sensor feeds. It underpins applications from quality-control cameras on factory floors to facial recognition and autonomous vehicle perception.
- Governance
Conformity Assessment
The formal process by which a high-risk AI system is evaluated against the requirements of the EU AI Act before being placed on the market. Depending on the risk category, assessment may be self-conducted or require a third-party notified body.
- Grundlagen
Constitutional AI
A training methodology developed by Anthropic in which an AI model is guided by a written set of principles (a "constitution") to self-critique and revise its outputs. Constitutional AI is one approach to building safer, more controllable AI systems at scale.
- Bereitstellung
Containerization (AI)
Packaging AI software and its dependencies into containers for more consistent deployment across environments. Containers help control software dependencies, but hardware, drivers, external services and data versions can still differ. Reproducibility and auditability require versioned inputs, compatible infrastructure and validation beyond the container image.
- Techniken
Context Window
The maximum amount of text (measured in tokens) an LLM can process in a single request, encompassing both the prompt and the generated output. Larger context windows—now exceeding 1 million tokens in some models—enable processing of long documents, codebases, and meeting transcripts in one pass.
- Techniken
Contrastive Learning
A self-supervised technique that trains a model by pulling representations of similar samples together and pushing dissimilar ones apart. Contrastive learning produces high-quality embeddings for images and text and underpins models like CLIP, which links vision and language.
- Architektur
Cross-Entropy Loss
The most common training objective for classification and language modelling tasks, measuring the difference between predicted probability distributions and true labels. Minimising cross-entropy loss during training drives the model to assign high probability to correct outputs.
D
- Bereitstellung
Data Drift
A change in the statistical distribution of input data fed to a deployed model, which can degrade prediction quality even if the model itself has not changed. Detecting data drift early is critical for maintaining reliable production AI systems.
- Governance
Data Governance
The policies, processes, and standards that manage data availability, usability, integrity, and security throughout its lifecycle. For high-risk AI systems, Article 10 of the EU AI Act sets data-governance and quality requirements appropriate to the intended purpose for relevant training, validation and testing datasets. Requirements differ where the system does not use model-training techniques.
- Governance
Data Lineage
The ability to track data's origin, movement, transformations, and consumption throughout a system. Data lineage is essential for AI auditability, debugging model behaviour, and demonstrating GDPR and EU AI Act compliance to regulators.
- Governance
Data Minimisation
The GDPR principle that personal-data processing should use only data relevant and necessary to its defined purposes. For AI, this requires decisions about which data to collect, retain and use. Federated learning, synthetic data and differential privacy may support particular controls, but their names alone do not demonstrate minimisation, anonymity or lawful processing.
- Business
Data Scientist
A professional who combines statistics, machine learning, and domain knowledge to extract insights and build predictive models from data. Data scientists are most effective when paired with strong data engineering and MLOps infrastructure to move from notebook experiments to production systems.
- Business
Decision Intelligence
An applied discipline that combines data science, social science, and managerial science to improve the quality of organisational decision-making through AI and analytical tools. Decision intelligence focuses on the full decision process—framing, modelling, intervention, and learning—not just prediction.
- Grundlagen
Deep Learning
A subset of machine learning based on artificial neural networks with multiple layers. Deep learning can learn complex patterns from large amounts of data and is particularly effective for image recognition, speech processing, and natural language understanding.
- Governance
Deepfake Detection
AI-powered methods for identifying media—images, video, audio—that have been synthetically generated or manipulated using generative AI. Deepfake detection is increasingly important for trust infrastructure in elections, financial services, and media.
- Techniken
Diffusion Model
A class of generative models that learn to reverse a gradual noise-addition process to produce high-quality images, audio, or video from random noise. Diffusion models power leading image generators such as Stable Diffusion and DALL-E 3.
- Business
Digital Transformation
The broad process of integrating digital technology into all areas of a business to fundamentally change how it operates and delivers value. AI is increasingly the central driver of new digital transformation initiatives as organisations move beyond basic digitisation.
- Architektur
Dropout
A regularisation technique that randomly sets a fraction of activations to zero during training. Dropout is intended to reduce reliance on particular activations and can help limit overfitting. Its effect depends on the architecture, rate and training setup; validation is still needed.
E
- Bereitstellung
Edge AI
Running AI models on or near devices such as industrial controllers, cameras, vehicles and mobile phones. Edge deployment can support local processing and limited-connectivity use, but latency, offline behaviour and privacy depend on the complete system. Device security, data storage, telemetry, updates and any remote dependencies still need qualification.
- Techniken
Embedding
A numerical representation of data (text, images, etc.) in a high-dimensional vector space where semantically similar items are mathematically close together. Embeddings are the foundational building block for semantic search, RAG pipelines, and recommendation systems.
- Architektur
Embedding Layer
The initial layer of a neural network that maps discrete tokens (words, image patches) to dense continuous vectors that the model can process. The quality of an embedding layer significantly affects downstream task performance.
- Grundlagen
Embodied AI
AI systems situated in a physical body or robot that perceive the world through sensors and act on it through actuators. Embodied AI bridges software intelligence and the physical world, making it central to robotics, autonomous vehicles, and smart manufacturing.
- Architektur
Encoder-Decoder
A two-part neural network architecture where the encoder compresses an input into a latent representation and the decoder generates an output from it. Used in translation, summarisation, and speech recognition systems.
- Governance
EU AI Act
The world's first comprehensive AI regulation, adopted by the European Union in 2024 and phasing in through 2027–2028 (prohibitions from February 2025, GPAI obligations from August 2025, stand-alone high-risk systems from 2 December 2027, and high-risk AI embedded in regulated products from 2 August 2028). It classifies AI systems by risk level—unacceptable, high, limited, and minimal—and imposes obligations proportionate to risk, from outright bans to conformity assessments and transparency requirements.
- Governance
Explainability (XAI)
The ability to make an AI system's outputs and relevant limitations understandable to the people using or affected by it. Explanation methods should fit the decision, audience and available evidence. EU AI Act Article 13 addresses transparency for deployers of high-risk systems; Article 86 provides explanation rights for specified affected decisions, subject to its scope and exceptions. Neither prescribes one universal XAI method.
- Grundlagen
Explainable AI (XAI)
Methods that help people interpret and evaluate an AI system's outputs, behaviour and limitations. Options include interpretable models, feature attribution, examples and other explanations, selected for the task and audience. Validate whether an explanation is faithful and useful; using an XAI technique alone does not establish transparency, safety or compliance.
F
- Governance
Fairness (AI)
The property of an AI system that ensures its outputs and decisions do not systematically disadvantage individuals on the basis of protected characteristics such as gender, ethnicity, or disability. Multiple mathematical definitions of fairness exist and may be in tension with one another.
- Bereitstellung
Feature Store
A platform for creating, storing, sharing, and serving ML features across training and inference pipelines. Feature stores can support feature reuse and consistency, but do not eliminate training-serving skew by themselves. Transformation versions, timestamps, freshness and the serving path still need validation.
- Grundlagen
Federated Learning
A collaborative learning approach in which participants train on locally held data and coordinate model updates rather than pooling their training datasets. Actual data flows depend on the architecture. Privacy depends on protections for data, updates, access and outputs; federated training alone does not establish confidentiality or GDPR compliance.
- Architektur
Feed-Forward Network (FFN)
The fully connected sublayer within each transformer block that applies two linear transformations with a non-linearity between them. FFN layers are responsible for storing factual associations and account for the majority of a transformer's parameters.
- Techniken
Few-Shot Learning
A prompting or training approach that uses a small number of labelled examples for a task. In-context few-shot prompting supplies examples in the prompt without updating model weights. Its quality and cost should be assessed against the task, prompt length and alternative approaches.
- Techniken
Fine-tuning
The process of taking a pre-trained AI model and further training it on a specific dataset to adapt it for a particular task or domain. Fine-tuning can deliver significantly better performance than prompting alone for specialised enterprise workflows.
- Grundlagen
Foundation Model
A large AI model trained on broad data at scale that can be adapted to a wide range of downstream tasks through fine-tuning or prompting. GPT-4, Claude, and Gemini are examples; under the EU AI Act, providers of GPAI models face specific transparency obligations.
- Business
Fractional Chief AI Officer
A senior Chief AI Officer engaged on a part-time or project basis to provide executive AI leadership without the cost of a full-time hire. Fractional Chief AI Officers are particularly valuable for mid-market companies that need board-level AI strategy, vendor evaluation, and team upskilling but cannot yet justify a permanent role.
G
- Governance
GDPR + AI
Applying data-protection requirements when AI systems process personal data. Relevant responsibilities include lawfulness, purpose limitation, data minimisation, security and data-subject rights. Articles 13–15 address information and access, including specified automated-decision information; Article 22 separately restricts solely automated decisions with legal or similarly significant effects, subject to exceptions and safeguards. Obligations depend on the processing and the organisation's role.
- Techniken
Generative Adversarial Network (GAN)
A framework consisting of two neural networks—a generator and a discriminator—that compete against each other. The generator creates synthetic data; the discriminator tries to distinguish real from fake. GANs produce highly realistic images and are also studied in the context of deepfake detection.
- Grundlagen
Generative AI
AI models capable of creating new content—text, images, code, audio, or video—rather than just classifying existing data. Generative AI is driving a productivity wave in enterprise workflows but also introduces governance challenges around accuracy and intellectual property.
- Business
Generative AI for Enterprise
The deployment of generative AI, such as language models, image generators and code assistants, in enterprise workflows. Examples include internal assistants, customer-facing chatbots, content production and software development. Both enterprise and consumer products need appropriate controls; enterprise integration adds requirements tied to organisational data, permissions, responsibilities and operating processes.
- Governance
GPAI (General Purpose AI)
General-purpose AI models, such as large language models, that can support a wide range of tasks. The EU AI Act distinguishes technical documentation for authorities, information for downstream providers, a copyright policy, and a public summary of training content. Some documentation duties have qualified open-source exemptions; models with systemic risk have additional obligations.
- Architektur
GPT Architecture
A family of autoregressive, decoder-only transformer architectures pre-trained to predict the next token. GPT models are the basis for ChatGPT and most commercial LLM APIs. Their autoregressive design makes them excellent text generators but less efficient for classification tasks than encoder models.
H
- Techniken
Hallucination
The tendency of LLMs to generate plausible-sounding but factually incorrect or fabricated content. Hallucination is a primary enterprise risk for generative AI and is mitigated through RAG, grounding techniques, output validation, and human-in-the-loop review processes.
- Governance
High-Risk AI System
An AI system classified under the EU AI Act as posing significant risks to health, safety, or fundamental rights. Applicable requirements include conformity assessment, technical documentation and human oversight. Registration depends on the classification route and use case: Article 49 covers specified Annex III systems, with critical-infrastructure systems under Annex III point 2 registered at national level.
- Governance
Human Oversight
Human supervision that enables people to understand, monitor and, where appropriate, challenge or intervene in an AI system's operation. EU AI Act Article 14 requires oversight measures proportionate to a high-risk system's risks, autonomy and use context. Article 26 assigns deployers responsibility for competent, trained and authorised oversight personnel with adequate support.
- Business
Human-AI Collaboration
Workflows that combine human judgement and domain knowledge with AI-supported analysis or execution. Define which decisions remain human, the system's permitted actions and the process for review and escalation. Evaluate the combined workflow against suitable human-only and AI-only baselines for the same task, using quality, risk, time and cost measures. Report improvements only when demonstrated by that evaluation.
- Governance
Human-in-the-Loop (HITL)
A workflow in which an authorised person meaningfully reviews a specified AI-assisted decision before it takes effect. Its suitability depends on the decision, risks and applicable obligations. GDPR Article 22 restricts solely automated decisions with legal or similarly significant effects, subject to exceptions and safeguards; it does not impose this workflow on every high-stakes use case.
- Governance
Human-on-the-Loop (HOTL)
A workflow in which people supervise an AI system and can intervene, without approving every action in advance. Suitability depends on the system's intended use, risks, intervention time and the person's ability to act effectively. Monitoring alone does not establish compliance with human oversight requirements or make a high-risk use acceptable.
- Business
Hyperion Physical AI Product System
The Hyperion Physical AI Product System is a decision framework with five lifecycle stages: Strategy, Discovery, Productization, Launch and Scale. Six cross-cutting lenses connect user value, economics, AI and data, system architecture, assurance, and deployment and operations to the evidence for an accountable decision. Field evidence can reopen an earlier decision. Product Decision Review, Product Leadership Mission and Executive Product Leadership are independent commercial engagements; each is selected and scoped for the product, situation and decision authority required. A lifecycle stage is not a requirement to buy an engagement.
I
- Bereitstellung
Inference
The process of running a trained model on new data to produce predictions or generated outputs. Inference cost and latency are the dominant operational concerns in production AI, particularly for large generative models that can cost cents per request at scale.
- Bereitstellung
Inference Cost
The compute and financial cost of running a model to produce a single prediction or generated response. Inference cost is often the dominant AI operational expenditure at scale and is managed through model compression, caching, quantization, and batching strategies.
- Governance
Instructions for Use (AI Act)
Documentation provided by AI system providers to deployers explaining the system's intended purpose, performance, limitations, human oversight requirements, and maintenance obligations. Under the EU AI Act, instructions for use are mandatory for all high-risk AI systems.
- Governance
ISO 42001
The international standard for AI management systems, published in 2023. ISO 42001 provides organisations with a framework to establish, implement, maintain, and continually improve responsible AI governance—complementary to, but distinct from, EU AI Act legal requirements.
K
- Governance
Kill Switch
A control that allows an authorised person to interrupt an AI system or disable specified actions. The intended stopping behaviour needs application-specific design and validation. EU AI Act Article 14 includes intervention or interruption through a stop control or similar procedure, as appropriate and proportionate; a generic software switch does not itself demonstrate safety or compliance.
- Techniken
Knowledge Distillation
A training technique in which a student model learns from a teacher model’s outputs or representations. It can support smaller models for constrained deployments, but latency and quality must be evaluated for the intended task. Training-data rights and the teacher’s licence or service terms also affect whether the planned use is permitted.
- Techniken
Knowledge Graph
A structured representation of entities and their relationships. Knowledge graphs can provide explicit relationships and provenance for enterprise retrieval or analysis, but their usefulness depends on source accuracy, maintenance and access controls. Connecting a graph to an LLM does not by itself verify generated answers.
- Bereitstellung
Kubernetes for ML
Using the Kubernetes container orchestration platform to manage scalable, fault-tolerant AI/ML workloads in production. Kubernetes enables auto-scaling inference services, GPU resource management, and rolling model updates with minimal downtime.
L
- Techniken
Large Language Model (LLM)
AI models trained on vast amounts of text data that can understand and generate human-like text. Examples include GPT-4, Claude, and Llama. LLMs power modern chatbots, content generation, and code assistance tools.
- Architektur
Layer Normalization
A technique that normalises activations across the feature dimension within a single layer, stabilising training of deep networks. Layer norm is standard in transformer architectures, enabling effective training at billion-parameter scale.
- Bereitstellung
LLMOps
An extension of MLOps focused on the operational challenges specific to large language models: prompt management, context handling, latency optimisation, cost control, hallucination monitoring, and compliance logging. LLMOps tooling is rapidly maturing in 2025–26.
- Architektur
Logit
The raw, unnormalised score produced by a model's final layer before the softmax function is applied. Logits are useful for calibration, temperature scaling, and understanding a model's confidence distribution across possible output classes.
- Techniken
LoRA (Low-Rank Adaptation)
A parameter-efficient fine-tuning method that injects trainable low-rank matrices into transformer layers instead of updating all model weights. LoRA enables cost-effective domain adaptation of large models on modest hardware, making enterprise fine-tuning accessible without full GPU clusters.
M
- Grundlagen
Machine Learning (ML)
A subset of AI that enables systems to automatically learn and improve from experience without being explicitly programmed. ML algorithms build models based on training data to make predictions or decisions.
- Techniken
Mixture of Experts (MoE)
A model architecture where different sub-networks ("experts") specialise in different types of inputs, and a gating network routes each token to the most relevant experts. MoE enables very large model capacity at lower inference cost—Mixtral and GPT-4 are believed to use this approach.
- Business
ML Engineer
An engineer who specialises in building, deploying, and maintaining machine learning systems in production. ML engineers bridge the gap between data scientists' experimental models and the robust, scalable infrastructure required for enterprise AI.
- Architektur
MLOps
The practice of combining machine learning, DevOps, and data engineering to support deployment, monitoring, and maintenance of ML models in production. MLOps provides processes and controls for reliability, scaling and reproducibility; those properties still require validation of the actual data, software, infrastructure and operating procedures.
- Governance
Model Card
A short document accompanying a trained AI model that describes its intended uses, performance across demographic groups, limitations, and ethical considerations. Model cards promote transparency and are considered best practice under frameworks like the EU AI Act and ISO 42001.
- Bereitstellung
Model Compression
Techniques such as quantization, distillation, pruning, and low-rank factorisation that aim to reduce model size or computation. Compression introduces trade-offs that depend on the model, task, runtime and hardware. Validate quality, resource use and latency for the resulting configuration before treating it as suitable for deployment.
- Bereitstellung
Model Drift
A change in the data or relationships a deployed model encounters that can degrade its performance. Monitoring for drift is an MLOps practice whose checks depend on the task and risk. For high-risk AI systems, the EU AI Act’s post-market monitoring duties require relevant performance information; not every distribution change is a failure or reportable incident.
- Bereitstellung
Model Registry
A centralised repository for versioning, cataloguing, and managing trained ML models throughout their lifecycle—from experimentation through staging to production. A model registry is the foundation of reproducible, auditable AI deployments.
- Bereitstellung
Model Serving
The process of deploying trained ML models to production environments where they can receive inputs and return predictions at scale. Model serving infrastructure must address throughput, latency, versioning, and cost while meeting SLAs.
- Architektur
Multi-Head Attention
An extension of the attention mechanism that runs multiple attention functions in parallel, allowing the model to attend to information from different representation subspaces simultaneously. Multi-head attention is a core component of every transformer-based model.
- Grundlagen
Multimodal AI
AI systems that process and reason across multiple types of data simultaneously—text, images, audio, and video. Multimodal models enable richer enterprise applications such as document understanding that combines tables, charts, and prose.
N
- Grundlagen
Narrow AI
AI systems designed and trained to perform a single, specific task—such as image classification, text translation, or fraud detection. All commercially deployed AI today is narrow AI, which is why domain-specific governance and risk management remain practical priorities.
- Grundlagen
Natural Language Processing (NLP)
A branch of AI that enables computers to understand, interpret, and generate human language. NLP powers applications like chatbots, translation services, sentiment analysis, and text summarization.
- Grundlagen
Neural Network
A computing system inspired by biological neural networks in the brain. It consists of interconnected nodes (neurons) organized in layers that process information and can learn patterns from data.
- Governance
NIST AI RMF
The US National Institute of Standards and Technology AI Risk Management Framework—a voluntary framework organised around four core functions: Govern, Map, Measure, and Manage. Widely adopted by enterprises globally as a complement to EU AI Act compliance programmes.
O
- Bereitstellung
On-Premise AI
Deploying AI models on an organisation’s own premises rather than relying on an external model API. This can address specific control, connectivity or latency requirements, but it does not itself ensure privacy or compliance. External processing is not categorically prohibited by GDPR; assess the actual processing, recipients, access arrangements and any international transfers.
- Bereitstellung
Online Learning
A training paradigm where a model is continuously updated with new data as it arrives, rather than in discrete batch retraining cycles. Online learning is used in recommendation systems and fraud detection where data distributions shift rapidly.
- Bereitstellung
ONNX (Open Neural Network Exchange)
An open format for representing machine learning models that enables interoperability across frameworks and hardware. Exporting to ONNX allows models trained in PyTorch to run efficiently on TensorRT, OpenVINO, or other optimised runtimes.
P
- Techniken
PEFT (Parameter-Efficient Fine-Tuning)
A family of techniques, including LoRA, prefix tuning and adapters, that adapt pre-trained models while updating a limited subset of parameters or adding trainable components. This can reduce adaptation and checkpoint-storage requirements compared with full fine-tuning. Actual memory, compute and quality depend on the method and configuration.
- Business
Physical AI
AI systems that perceive and act in the physical world through robotics, autonomous vehicles, smart manufacturing equipment, and industrial sensors. Physical AI integrates computer vision, reinforcement learning, and real-time control to automate tasks previously requiring human dexterity and judgment in unstructured environments.
- Architektur
Physical AI system architecture
A reference architecture for reasoning about a Physical AI system: device and control, embedded platform, perception and robotics, intelligence and knowledge, and orchestration and operations, with separate safety and cybersecurity responsibilities. Orchestration makes coordination, authorization, human review and recovery requirements explicit. The architecture does not establish that a particular implementation is safe, secure or production-qualified.
- Business
Pilot Project
A limited real-world deployment of an AI solution that tests performance, gathers user feedback, and refines the approach before broader rollout. A successful pilot de-risks the full investment and builds organisational confidence in AI.
- Architektur
Positional Encoding
A technique that injects information about the position of each token in a sequence into the transformer, compensating for the architecture's lack of inherent sequence awareness. Modern models use learned or rotary positional encodings (RoPE) to support long context windows.
- Governance
Post-Market Monitoring
The ongoing collection and analysis of information about a deployed AI system’s performance and risks. Under Article 72 of the EU AI Act, providers of high-risk systems must establish proportionate post-market monitoring. Serious-incident reporting is a separate Article 73 obligation with defined conditions; a drift observation is not automatically a reportable serious incident.
- Business
Production AI
AI systems that are live in enterprise environments, processing real transactions and influencing business decisions at scale. Production AI requires robust MLOps practices, monitoring, compliance controls, and SLAs—a significantly higher bar than prototype or pilot systems.
- Governance
Prohibited AI
AI practices prohibited by Article 5 of the EU AI Act under defined conditions, including certain harmful manipulation, exploitation and social scoring. Some biometric practices are also prohibited. The law sets specific scopes and exceptions, including for certain law-enforcement uses of real-time remote biometric identification; classification requires checking the actual use against the legal text.
- Bereitstellung
Prompt Cache
A technique that reuses cached computation for a repeated prompt prefix, such as attention key-value states. Benefits depend on cache eligibility, hit rate, eviction, implementation and any provider pricing rules. A cache does not remove the need to check access boundaries and treatment of sensitive prompt data.
- Techniken
Prompt Engineering
The practice of designing and optimising input prompts to elicit the best results from AI models, especially LLMs. Structured prompting—using system instructions, examples, and constraints—can dramatically improve output quality without changing the underlying model.
- Business
Proof of Concept (POC)
A small-scale implementation that demonstrates the technical feasibility of an AI solution without full production investment. POCs validate whether a chosen approach can work before committing resources to pilot and production phases.
- Governance
Purpose Limitation
The GDPR principle that personal data are collected for specified, explicit and legitimate purposes and not processed further in an incompatible way. Reuse for AI training requires assessing the existing purpose, legal basis, relevant safeguards and, where Article 6(4) applies, compatibility with the original purpose. Compatible reuse can rely on the original legal basis; consent or legislation may provide other routes subject to their conditions and other GDPR duties.
Q
R
- Bereitstellung
Ray Serve
An open-source, scalable model serving framework built on the Ray distributed computing library. Ray Serve supports complex inference pipelines with model composition, dynamic batching, and Python-native deployment, making it popular for LLM serving.
- Bereitstellung
Real-Time Inference
Serving model predictions with low latency in response to individual live requests, typically within milliseconds to seconds. Real-time inference is required for customer-facing applications like chatbots, fraud detection, and autonomous control systems.
- Grundlagen
Reinforcement Learning (RL)
A machine learning paradigm where an agent learns by interacting with an environment and receiving reward or penalty signals. RL is the foundation of game-playing systems like AlphaGo and increasingly powers robotics, logistics optimisation, and dynamic pricing.
- Grundlagen
Reinforcement Learning from Human Feedback (RLHF)
A training technique that refines language model behaviour by learning from human preferences rather than fixed labels. RLHF is a primary method used to align LLMs like ChatGPT and Claude with desired values and reduce harmful outputs.
- Architektur
Residual Connection
A shortcut connection that adds a layer's input directly to its output, allowing gradients to flow more easily during training and enabling very deep networks. Residual connections are essential in transformers and were pioneered in the ResNet image model family.
- Techniken
Retrieval-Augmented Generation (RAG)
A technique that retrieves information from external knowledge sources for an LLM to use when generating a response. RAG can provide current, task-specific context without retraining the model. Accuracy depends on retrieval, source quality, access controls and generation behaviour; retrieved context does not guarantee a correct or supported answer.
- Business
Revenue AI
AI applications intended to support revenue, such as recommendations, pricing assistance, lead qualification and marketing. Incremental value should be tested against a suitable baseline with attention to data rights, customer effects and business constraints. Where appropriate, controlled experiments can help distinguish a causal effect from correlation.
- Governance
Right to Explanation
An individual's ability to obtain an understandable explanation of certain automated decisions affecting them. For automated decision-making covered by GDPR Article 22, Articles 13(2)(f), 14(2)(g) and 15(1)(h) address information about the logic, significance and expected consequences, alongside Article 22's restrictions and safeguards. The applicable right and required explanation depend on the processing and legal context.
S
- Techniken
Self-Supervised Learning
A training paradigm where a model generates its own supervision signal from unlabelled data—for example, by predicting masked words in a sentence. Self-supervised pre-training on internet-scale text or images is what enables foundation models to acquire broad world knowledge.
- Techniken
Semantic Search
A search approach that uses representations of meaning to retrieve results beyond exact keyword matches, often through embeddings and vector search. Its relevance should be evaluated against lexical and hybrid retrieval on the actual corpus and queries; semantic similarity alone does not establish factual relevance or permission to access a result.
- Bereitstellung
Shadow Mode
A deployment technique where a new model processes production traffic in parallel with the existing system but its predictions are logged rather than served to users. Shadow mode can reveal model behaviour on operational data before cutover, but data access, logging, resource use and integration still require controls; it does not establish deployment safety.
- Business
SME AI Activation
Applying product leadership to an SME’s AI opportunity with scope and resources suited to the business. Hyperion’s Product Decision Review, Product Leadership Mission and Executive Product Leadership address different decision or leadership needs and can be commissioned independently. Production readiness depends on the agreed scope, evidence, integration and operational ownership.
- Architektur
Softmax
A mathematical function that converts a vector of raw logits into a probability distribution summing to 1. Softmax is applied at the output of classifiers and language models to produce token probabilities used in sampling and decoding.
- Architektur
Sparse Model
A model architecture that activates only a subset of its parameters for any given input, rather than the full network. Sparse models—enabled by Mixture of Experts designs—achieve larger total capacity while keeping per-inference compute manageable.
- Grundlagen
Supervised Learning
A machine learning approach where the model is trained on labelled input-output pairs. The model learns a mapping intended to support predictions on new data. Performance on unseen examples depends on the training data, task and evaluation; supervised learning is used for tasks such as classification and regression.
T
- Governance
Technical Documentation (AI Act)
A mandatory artefact under the EU AI Act that high-risk AI providers must prepare and maintain, detailing the system's purpose, architecture, training methodology, validation data, performance metrics, risk management measures, and post-market monitoring plan.
- Techniken
Temperature (LLM)
A parameter that changes token-selection probabilities during sampling. Lower values concentrate probability on more likely tokens; higher values spread it more widely. Some runtimes use a zero setting for greedy decoding, but temperature alone does not guarantee reproducible output across runtimes, hardware or model versions. Supported settings depend on the implementation.
- Bereitstellung
Token Budget
The total number of tokens allocated for a model request, encompassing both input (prompt + context) and output. Managing token budgets is central to controlling inference cost in production LLM applications, especially when processing long documents or maintaining conversational history.
- Techniken
Tokenization
The process of splitting raw text into smaller units called tokens—typically sub-words or word-pieces—that serve as the basic input unit for language models. Token count determines both model context limits and API pricing, making tokenization an important operational consideration.
- Techniken
Top-p Sampling (Nucleus Sampling)
A decoding strategy that restricts the model's next-token choices to the smallest set of tokens whose cumulative probability exceeds a threshold p. Used alongside temperature, top-p sampling balances output diversity and coherence in production LLM deployments.
- Grundlagen
Training (AI)
The process of exposing a model to large volumes of labelled or unlabelled data and adjusting its internal parameters to minimise prediction errors. Training is computationally intensive, often performed on GPU clusters, and constitutes the major cost in foundation model development.
- Techniken
Transfer Learning
A technique in which a model pre-trained on one task or dataset is adapted for a different, related task. Reusing learned representations can reduce the need to train from scratch, but the benefit depends on task similarity, data quality and adaptation. Evaluate the adapted model on the intended domain.
- Architektur
Transformer
The dominant neural network architecture for language, vision, and multimodal AI, introduced in the 2017 "Attention Is All You Need" paper. Transformers use self-attention to process all tokens in parallel, enabling training on internet-scale data and powering every major LLM in use today.
- Bereitstellung
Triton Inference Server
NVIDIA's open-source inference serving software that supports multiple frameworks (TensorRT, ONNX, PyTorch, TensorFlow) on GPU infrastructure. Triton is widely used in enterprise deployments requiring maximum throughput from GPU hardware.
U
- Governance
Unacceptable Risk AI
The highest risk tier under the EU AI Act, covering AI applications whose potential harms to society are deemed too severe to permit. Systems in this category are prohibited from being placed on the market or put into service in the EU.
- Grundlagen
Unsupervised Learning
A machine learning approach that discovers structure in unlabelled data. Common techniques include clustering (grouping similar items) and dimensionality reduction. It is often used for anomaly detection and exploratory data analysis.
V
- Architektur
Vector Database
A specialised database designed to store and efficiently search high-dimensional embedding vectors using approximate nearest-neighbour algorithms. Vector databases are the infrastructure backbone of RAG systems and semantic search applications.
- Architektur
Vision Transformer (ViT)
An adaptation of the transformer architecture to image data, treating fixed-size image patches as tokens. ViTs now outperform convolutional networks on many computer vision benchmarks and are used in medical imaging, satellite analysis, and industrial quality control.
Z
Hyperions kanonische Frameworks
Zwei Frameworks verankern jedes Projekt — die Architektur, für die wir bauen, und das Modell, nach dem wir liefern.
Physical-AI-Systemarchitektur
Gerät und Steuerung · Embedded-Plattform · Wahrnehmung und Robotik · Intelligenz und Wissen · Orchestrierung und Betrieb · Sicherheits- und Cybersicherheitsverantwortung
Eine Referenzsicht auf Geräte und Steuerung, eingebettete Software, Wahrnehmung, Intelligenz und Betrieb, mit getrennten Verantwortlichkeiten für Schutzfunktionen und Cybersicherheit. Orchestrierung dient dazu, Anforderungen an Koordination, Berechtigungen, menschliche Prüfung und Wiederherstellung für das konkrete System festzulegen.
Vollständige Definition lesenHyperion Physical AI Product System
Review der Produktentscheidung · Mission für Produktführung · Executive Produktführung
Ein Entscheidungsrahmen mit fünf Lebenszyklusphasen und sechs phasenübergreifenden Perspektiven. Neue Erkenntnisse können eine frühere Entscheidung wieder öffnen. Review der Produktentscheidung, Mission für Produktführung und Executive Produktführung sind unabhängig wählbare Mandate. Ihr Umfang richtet sich nach Produkt, Situation und erforderlicher Entscheidungsbefugnis.
Vollständige Definition lesen
Hilfe beim Verständnis von KI Benötigt?
Buchen Sie ein Physical-AI-Eignungsgespräch, um zu besprechen, wie diese KI-Konzepte auf Ihre Branche und Ihre Herausforderungen anwendbar sind.