Algorithmic Contract Design

No algorithm is an island -- algorithms constantly interact with self-interested players.

Algorithmic Game Theory (AGT) has thus far concentrated on one aspect of this: the algorithm's input is reported by such players.

Incentivizing truthful reports is the focus of mechanism design in economics, and in AGT, algorithmic mechanism design became a hugely successful research area.

We propose to apply the algorithmic lens to a different but no less important field in economics called contract design, recognized by the 2016 Nobel Prize. The essence of a contract is to incentivize players' actions (rather than reports).

It is thus extremely relevant to another way in which algorithms interact with players -- the algorithm's output is carried out through their actions.

We refer to the new research area that will emerge as algorithmic contract design (ACD).We aim to lay the theoretical foundations for ACD. Typically, computational environments are more complex than traditional economics ones.

Key complexities are:1)A rich choice of actions makes computing an optimal contract nontrivial; 2)The optimal contract can be unintuitive and brittle;3)A one-size-fits-all contract is suboptimal for a diverse player population;4)Multiple contracts can undermine each other; 5)Traditional contract formats can be too weak.We will tackle these complexities, designing the next generation of algorithmic incentive schemes for strategic action -- tractable, simple/robust, personalized and coordinated -- and develop new contract formats en route.

The potential impact of ACD is far-reaching: First, it will prevent traditional algorithms from failing due to selfish action choices.

Second, given the current influence of algorithms on behavior, it will help achieve a more socially-efficient allocation of effort.

Applications include traditional contracts moving to online platforms, like freelancing, as well as novel data-driven incentive schemes for domains like digital healthcare.