AI Glossary

The tendency of people to overvalue and uncritically accept the outputs of an automated system, which undermines the effectiveness of human oversight.

The difficulty that the internal decision-making processes of complex models are very hard for humans to explain, which complicates both oversight and the proving of liability.

A term for an AI agent deployed within an organization to perform tasks independently in a manner similar to a human employee, raising questions of liability and oversight.

An AI system falling into the high-risk category under the AI Act, subject to strict requirements on risk management, data quality, documentation, transparency and human oversight.

An arrangement in which the AI proposes a course of action or output, but its execution must be approved by a human before it is carried out; a form of human oversight.

A set of measures enabling a human to monitor the operation of an AI system and to intervene where necessary.

A principle under which a human must have real, not merely formal, control over a high-risk system, including the ability to intervene in and override the system.

The risk that an advanced system trained to optimize a certain goal begins to circumvent control mechanisms or human oversight in order to achieve the goal; particularly monitored in agentic systems.

A concept of AI that is lawful, ethical, and technically robust; it includes, among other things, respect for fundamental rights, transparency, fairness, and human oversight.