Where does bandit collide?

In sequential decision making a bandit is a choice problem in which an agent repeatedly picks one option from several whose payoff odds are unknown, and must balance trying options to learn their value against favoring the one that has paid best so far. Thompson's 1933 On the likelihood that one unknown probability exceeds another in view of the evidence of two samples gave the first procedure for choosing between two uncertain options as evidence builds up, and Robbins's 1952 Some aspects of the sequential design of experiments set out the general problem now called the multi-armed bandit, named after the slot machine or one-armed bandit whose lever a gambler keeps pulling. The referent here is a mathematical model of allocation under uncertainty, not a rural outlaw admired by villagers or a predatory ruler in an economic model.[1][2]

Unit

One arm-pull: a single choice among options with unknown payoff, one round of a repeated game

Recovery means

Not applicable

Measured by

Scored by cumulative reward, or by regret, the shortfall between the reward earned and the reward of always choosing the best option

A social bandit is an outlaw whom the rural community treats not as a common criminal but as a champion, avenger, or fighter for justice, tolerated, sheltered, and admired by the peasantry he comes from. The historian Eric Hobsbawm introduced the idea in his 1959 Primitive Rebels and developed it fully in his 1969 Bandits, arguing that figures of the Robin Hood type recur across agrarian societies under strain. The referent is a real person and a social role documented by historians, not a decision problem in statistics or an actor in a model of taxation.[3][4]

Unit

One outlaw cast as a peasant hero, studied as a recurring social role

Recovery means

Not applicable

Measured by

Traced through court records, ballads, folklore, and memoir that show how a community regarded the outlaw

In political economy a bandit is a way of modeling a ruler as a predator who takes wealth from a population. Olson's 1993 Dictatorship, Democracy, and Development distinguished the roving bandit, who steals what he can and moves on, from the stationary bandit, who settles in one place, turns theft into regular taxation, and therefore gains an interest in the prosperity he taxes. The referent is an actor in an economic model of predation and government, not an outlaw sheltered by villagers or a lever on a slot machine.[5]

Unit

One predatory actor in a model of extraction: roving or stationary

Recovery means

Not applicable

Measured by

Expressed through the actor's tax or theft take and its effect on investment and output in the model

In fisheries science, bandit refers to mobile fleets and traders who deplete one marine resource after another and move on before local institutions can respond. This meaning comes from Berkes (2006) in Science.[6]

Unit

Recovery means

Measured by