Glossary
The vocabulary of thinking well alongside AI — the biases, failure modes, and habits worth being able to name. Clear definitions, in plain English.
Anchoring
The tendency to rely too heavily on the first piece of information you encounter when making a judgement. When working with AI, the model's opening answer can become your anchor, quietly narrowing the options you go on to consider.
RADAR: Autonomy →Answer engine optimisation (AEO)
Structuring content so that AI systems and answer engines can understand, trust, and cite it directly — through clear definitions, structured data, and original facts — rather than optimising only to appear in a ranked list of links.
Automation bias
The tendency to over-trust automated systems: accepting their output even when it is wrong, and even when it conflicts with other evidence or your own knowledge.
RADAR: Reliance →Base rate neglect
Ignoring general background frequencies — how common something actually is — in favour of vivid, specific detail, which skews probability judgements.
Calibration
The match between how confident you are and how often you turn out to be right. Good calibration means being confident when correct and uncertain when not; overconfidence is high confidence paired with frequent error.
RADAR: Reliance →Cognitive offloading
Delegating mental work — remembering, calculating, reasoning — to an external tool. Useful in moderation, but heavy offloading to AI can erode the very skill being offloaded.
RADAR: Autonomy →Confirmation bias
The tendency to seek out, favour, and remember information that confirms what you already believe, while discounting what doesn't.
Critical thinking
The disciplined practice of deciding what to believe or do by examining the reasoning and evidence behind a claim, rather than accepting it at face value. In practice: clarify the question, surface assumptions, weigh evidence, consider alternatives, and check your own reasoning.
Read the guide →Deepfake
Synthetic audio, image, or video, generated or altered by AI, that convincingly depicts events or statements which never actually occurred.
RADAR: Detection →Dunning–Kruger effect
A pattern in which people with low competence in an area overestimate their ability — partly because the skills needed to perform well are the same ones needed to recognise performing badly.
Epistemic humility
Holding your beliefs with an honest awareness that they might be wrong, and a genuine willingness to revise them in the light of better evidence.
Falsifiability
The property of a claim that it could, in principle, be shown false by some possible observation. Claims that nothing could ever disprove carry little evidential weight.
Framing effect
When the way information is presented — its wording or emphasis — changes the conclusion people draw from the same underlying facts. Asking “is it safe?” and “what are the dangers?” can yield very differently weighted answers.
RADAR: Awareness →Hallucination (AI)
When an AI model produces confident, plausible-sounding content that is false or fabricated — invented facts, figures, or citations. Also called confabulation. The fabrication typically looks exactly like a genuine answer.
RADAR: Detection →Motivated reasoning
Reasoning steered towards a conclusion you already want to reach — marshalling arguments to defend it rather than to test it. An agreeable AI can amplify this considerably.
RADAR: Resistance →Prompt-shopping
Rephrasing a question over and over until the AI finally gives the answer you were hoping for. It feels like refining the prompt; it is really selecting for agreement rather than truth.
RADAR: Awareness →Provenance
The documented origin and chain of custody of a piece of information or media. Checking provenance means tracing a claim, quote, or image back to its primary source rather than to somewhere it was merely repeated.
RADAR: Detection →RADAR
The five-dimension framework used on this site for judgement at the human–AI interface: Reliance, Autonomy, Detection, Awareness, and Resistance.
The framework →Steelmanning
Engaging with the strongest possible version of an opposing argument rather than a weak caricature of it — the opposite of attacking a straw man. A core move for genuinely testing your own view.
Sycophancy (AI)
The tendency of AI systems to tell users what they want to hear — agreeing, flattering, or reversing position under push-back — because agreeable responses tend to be rewarded during training.
RADAR: Resistance →Synthetic media
Content — text, image, audio, or video — that has been generated or substantially altered by AI. See also: deepfake.
RADAR: Detection →