Option A: Accuracy

The Basics of Consequentialism

Consequentialism, particularly utilitarianism, focuses on the outcomes of actions, aiming to maximise overall benefits and minimise harm.

In the context of an algorithm like COMPAS, a consequentialist approach could prioritise maximising overall accuracy insofar this approach yields the best aggregate outcome for society: fewer total errors and, ideally, enhanced public safety. From this perspective, maximising accuracy serves the collective good, as a highly accurate system is likely to make fewer mistakes overall, which could mean better allocation of resources and more just outcomes in many cases.

Fine-Grained Consequentialist Assessment

Understanding this consequentialist approach requires taking into account the following aspects:

Aggregated Benefits: By maximising accuracy, the system is intended to provide reliable guidance to judges and other decision-makers, leading to fewer incorrect assessments and, in theory, a safer society. Accurate predictions mean that individuals who are genuinely high-risk are more likely to be flagged, potentially preventing harm and contributing to social stability.

Consideration of Error Rates: It is noteworthy that a consequentialist approach can still take into account the kinds of mistakes being made and the severity of potential errors (as we discussed in the context of error balance). For instance, a false positive (incorrectly labeling someone as high risk and thus potentially keeping them in prison) is a significant harm—especially if that person’s freedom and reputation are compromised without fault. Thus, while the priority remains on maximising overall accuracy, this approach does not completely disregard error rates. Rather, it seeks to limit severe mistakes where possible, acknowledging that certain types of errors (like unwarranted imprisonment) carry especially high moral and social costs. But the consequentialist approach would consider error rates only insofar as they are relevant to an assessment of the overall aggregation of benefits and harms.

Key Points

Trade-Offs with Error Rates: A consequentialist system optimised for accuracy might still show unequal error rates across demographic groups. For example, if African-American defendants have a higher false positive rate, the consequentialist might argue that, while unfortunate, this trade-off is justified if the system’s overall predictive reliability contributes to greater societal benefit. 

Net Positive Outcome: Even if some error disparities remain, the maximisation of overall accuracy could arguably yield a net positive outcome, with fewer false negatives (e.g., correctly identifying high-risk individuals), which enhances public safety. This approach emphasises aggregate benefit, but it does not entirely ignore individual harms—particularly in cases where errors could lead to significant personal consequences like wrongful incarceration.

Potential Criticisms

Disproportionate Impact on Minorities: If maximising accuracy results in a higher error rate for certain groups, such as African-Americans, consequentialists must grapple with the moral acceptability of this disparity. Is the aggregate benefit worth the potential for unequal burdens across groups? This remains an ethical tension within consequentialism, particularly when societal benefit comes at the expense of fairness for some.

Risk of Overlooking Structural Inequalities: Even with an emphasis on reducing severe errors, maximising accuracy could perpetuate biases present in historical data, reinforcing systemic inequities. A pure consequentialist view may risk insufficient attention to how these structural factors create unequal starting points for different groups.

Let's quickly recap: