Exceptions to Data-Driven Rules
A reflection on when individuals should be able to challenge algorithmic recommendations that fail to capture their context.
The Right to Be an Exception to a Data-Driven Rule
Data-driven rules are algorithmic systems that guide decisions in areas such as hiring, healthcare, credit scoring, and parole. This blog post argues that people should sometimes have a right to be treated as exceptions to these rules.
An exception is appropriate when applying a general model to a particular person produces an unfair or misleading result. For example, while working with job seekers without college degrees, I saw how applicant tracking systems systems could undervalue nontraditional experience. A candidate with strong practical skills and a commendable work ethic might be automatically rejected if the model treats formal education as the main signal of readiness.
Data-driven rules can be useful when population patterns genuinely improve decisions, such as using clinical data to guide treatment and prevention. However, in areas shaped by culture, language, unequal access, and institutional history, models can amplify exclusion. Not every human story fits a pattern.
There are benefits to individualization. It can protect people with nonstandard profiles, like older adults and migrants, who may be experiencing language barriers, tech illiteracy, and may not be represented in the algorithm design. At the same time, exceptions must be handled carefully. Human review can introduce bias, and excessive scrutiny can become invasive.
Uncertainty is useful here. A model should not be treated as authoritative when confidence is low or when the case differs meaningfully from the population used to build it. Centering uncertainty makes human accountability visible.
The question that interests me is: when data-driven rules fail to recognize human diversity, who should have the power to grant exceptions? Designers can improve models, institutions can audit decisions, and affected individuals can appeal outcomes. A fair system likely needs all three.
This blog helped me think more carefully about algorithmic fairness. Models that influence people's lives require evaluation frameworks, appeal processes, and critical attention to the contexts in which they are used.