KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Algorithmic traceability
Ability to track and document each step of an AI system's decision-making process, from input data to the final result.
Audit logging
Systematic and chronological recording of all interactions, parameters, and decisions of an AI model to enable post-hoc analysis.
Accountability
Formal mechanism for assigning responsibility for algorithmic decisions, including identification of actors and accountability processes.
Decision mapping
Structured visualization of an AI's decision path, identifying decision points and associated weights.
Algorithmic proof
Formal documentation establishing that an algorithmic decision complies with defined criteria and regulatory constraints.
AI certification
Independent evaluation process validating the compliance of an AI system with ethical, technical, and regulatory standards.
Model documentation
Complete record of the characteristics, limitations, training data, and performance metrics of an AI model.
Record of processing activities
Centralized database listing all algorithmic processing and their purposes, mandatory under GDPR.
Ethical impact assessment
Systematic evaluation of the potential consequences of an AI system on fundamental rights and societal values.
Accountability metrics
Quantitative indicators measuring the level of accountability and traceability of an AI system in its operational environment.
Accountability chain
Documented sequence of actors and actions contributing to an algorithmic decision, allowing for the identification of responsibilities.
Compliance audit
Systematic examination of an AI system to verify its compliance with legal, regulatory, and ethical requirements.
Procedural transparency
Complete disclosure of methods, algorithms, and parameters used by an AI system to make decisions.
Right to explanation
Legal right allowing individuals to obtain an understandable explanation of automated decisions affecting them.
Decision reconstruction
Ability to exactly reproduce the conditions that led to a specific algorithmic decision for analysis.
Version archiving
Systematic preservation of different versions of an AI model and their parameters for temporal traceability.