5 ideas
Subjects prefer betting on an urn with a known 50/50 red-black split over an urn with an unknown split, whether the bet is on red or on black. In the three-color version (30 red, 60 black-or-yellow in unknown proportion), people prefer betting on red over black, yet prefer 'black or yellow' over 'red or yellow'. No assignment of subjective probabilities can rationalize both preferences.
The typical urn choices violate Savage's Sure-Thing Principle, which says that preferences between two acts should not depend on outcomes the acts share. Adding the yellow payoff to both options reverses people's preferences. So their behavior cannot be represented as maximizing expected utility under any single subjective probability distribution.
Ambiguity is the quality of information about probabilities: its amount, reliability, and consensus. It is not the same as the probabilities themselves. Two situations can carry identical best-estimate odds yet differ sharply in how confident one can be in those odds, and people respond systematically to that difference.
Many thoughtful people, including decision theorists, keep their ambiguity-averse choices even after the axiom violation is explained to them. They defend those choices on reflection rather than treating them as mistakes. This suggests the Savage axioms are not self-evidently normative, and that avoiding ambiguity can be a defensible pattern of reasoning rather than a bias to correct.
When choices seem inconsistent with any single set of beliefs, ask how much the decider trusts their own probability estimates. Low confidence explains avoiding unknown options, paying for information that does not change expected value, and cautious policy choices in novel situations. These are responses to ambiguity, not errors in estimating odds.