Reference
Glossary
Plain definitions of the terms used throughout these pages.
Human-centered AI
An approach to designing and deploying AI systems in which the interests, autonomy, and well-being of the people affected are the primary frame of reference — not the capability of the model or the goals of the deploying institution.
Human agency
A person’s capacity to form intentions, weigh options, and act on their own reflective judgment. To protect agency is to leave room for that process; to expand it is to make it easier and richer.
Manipulation
Influence that bypasses or undermines a person’s reflective judgment — typically by exploiting cognitive shortcuts, emotional states, or asymmetries of information. Manipulation works precisely because the person being manipulated does not see it.
Persuasion
Influence that engages a person’s reflective judgment — by giving reasons, evidence, or honest appeals that they can accept or reject on their own terms. Persuasion respects the listener as a thinking being.
Augmented intelligence
A framing in which AI systems are tools that strengthen human reasoning and capability, rather than agents that replace human judgment. The intelligence remains human; the system extends it.
Trustworthy AI
An AI system whose behavior, incentives, and limits are intelligible enough — and stable enough — that a reasonable person can rely on it without surrendering their own judgment to it.