Where does credit collide?

In social psychology, credit is the store of goodwill a group member builds up through competence and early conformity to the group's norms, which can later be 'spent' to deviate from those norms without losing standing: Hollander's idiosyncrasy credit. Hollander's 1958 Conformity, status, and idiosyncrasy credit set out how status earns a person latitude to innovate or dissent. The referent is an earned social licence to depart from expectations, tracked by how much non-conformity a member can sustain before sanction, and it is a wholly different construct from a bookkeeping entry or an algorithm's problem of assigning reward.[1]

Unit

The accumulated standing a member can spend on deviating from norms

Recovery means

Not applicable

Measured by

Inferred from status, prior conformity, and competence, and tested by tolerance for a member's subsequent non-conformity

In artificial intelligence and reinforcement learning, the credit-assignment problem is the difficulty of deciding which of many earlier decisions, actions, or internal components deserves the credit (or blame) for an eventual outcome, which is hard when the reward arrives long after the choices that produced it. Minsky's 1961 Steps toward artificial intelligence named the problem and placed it at the centre of machine learning. The referent is the attribution of responsibility for an outcome across a sequence or a structure, a computational construct unrelated to an accountant's right-hand entry or a group member's earned social latitude.[2]

Unit

The share of responsibility for an outcome assigned to each prior action or component

Recovery means

Not applicable

Measured by

Addressed algorithmically through temporal-difference returns, eligibility traces, and backpropagated error gradients