Where does reward collide?

In common law a reward is a sum publicly offered for the performance of a specified act, such as returning lost property or giving information that leads to a conviction, and the offer becomes a binding contract once someone performs the act. Williams v Carwardine (1833) held that a person who did the act could claim the posted reward even though it was not her main motive, treating the notice as an offer that performance accepts. Carlill v Carbolic Smoke Ball Co (1893) made this the textbook case of a unilateral contract, where a promise to the world is accepted by conduct rather than by a return promise. The referent is a legal offer enforceable in court, not a signal recorded in the brain, a number an algorithm maximises, or the pay an employer gives for a job.[1][2]

Unit

One posted reward offer, defined by the act it names and the sum promised

Recovery means

Not applicable

Measured by

Established in litigation by whether a valid offer was made and whether the claimant performed the named act while the offer was in force

In physiology a reward is a stimulus or event that an animal will work to obtain and that strengthens the behaviour producing it, handled by dedicated circuits in the brain. Olds and Milner's 1954 Positive reinforcement produced by electrical stimulation of septal area and other regions of rat brain found that rats would press a lever repeatedly to stimulate particular brain regions, identifying neural sites whose activity acts as reward. Schultz, Dayan, and Montague's 1997 A neural substrate of prediction and reward showed that midbrain dopamine neurons fire in proportion to the gap between received and predicted reward, the reward prediction error. The referent is a biological stimulus and the neural response it drives, measured in the laboratory, not a clause enforced by a court, a term in an optimisation, or a line in a compensation plan.[3][4]

Unit

A rewarding stimulus and the neural firing it evokes, per trial or event

Recovery means

Not applicable

Measured by

Assessed by how hard an animal works to obtain the stimulus and by recording dopamine neuron firing or brain activity during it

In sequential decision making a reward is the scalar number an agent receives from its environment at each step, and the agent's goal is to choose actions that maximise the total reward it accumulates over time. Howard's 1960 Dynamic Programming and Markov Processes attached a reward to each transition of a Markov process and solved for the policy with the greatest long-run reward. Sutton and Barto's 1998 Reinforcement Learning: An Introduction placed this reward signal at the centre of the field, stating the reward hypothesis that any goal can be framed as maximising expected cumulative reward. The referent is a designer-defined quantity inside an algorithm, not a stimulus recorded in a brain, a payment set by an employer, or an offer enforced as a contract.[5][6]

Unit

A single scalar value delivered to the agent per step or transition

Recovery means

Not applicable

Measured by

Specified by a reward function the designer defines over states and actions, and summed as cumulative or discounted return

In organizations a reward is the pay, benefit, or recognition an employer gives an employee in return for work, and the way these are arranged into a reward system is a central tool for motivating and keeping staff. Lawler's 1971 Pay and Organizational Effectiveness: A Psychological View analysed how the amount of a financial reward, the basis on which it is given, and its perceived fairness shape effort and satisfaction, and separated rewards tied to performance from those given for membership alone. The referent is a resource an employer allocates to a worker, not a stimulus recorded in the brain, a scalar maximised by an algorithm, or an offer enforced as a contract in court.[7]

Unit

One reward allocated to an employee, in money or in kind

Recovery means

Not applicable

Measured by

Captured by compensation levels, pay-for-performance ratios, and survey measures of perceived reward fairness