Fairness in Algorithmic Targeting : An Economic Analysis

With algorithmic targeting getting increasingly common, the issue of fairness is quickly coming the fore. For example, financial firms have been accused of targeting minorities with pricey loans. Likewise, online job advertisements have disproportionately been targeted to male candidates, disadvantaging females. In order to examine such potentially unfair practices and their possible remedies, we develop an economic model where a firm sells two products and targets each potential consumer with one of them. Targeting is personalized, that is, not all consumers are recommended the same product. Each consumer decides whether to accept the firm's recommendation or to reject it and look for alternatives, or to reject it altogether and not do business with the firm at all. We explain why, if the firm has different levels of information on various demographic groups, it might have an incentive to target them in a differential manner, resulting in uneven welfare across consumer groups. Such unequal treatment is often viewed as discrimination in the policy parlance. We also explain why enforcing equal treatment might not yield the intended effect and in what circumstances it could even become counterproductive. Our results have important implications for ethical deployment of algorithmic targeting