Algorithm Blog #1: Exception to Data Driven Rules
Published on:
Some thoughts on Data Driven Rules.
Case Study
The Right to Be an Exception to a Data-Driven Rule.
The article defines a data-driven rule as “decision rules that define the relationship between the inputs and outputs of data-driven decision aids.” That is a pretty technical definition and be hard to understand. To put it in simpler terms, a data-driven rule is a ruling that is based on the analysis of actual data, rather than using human bias or intuition to make the decision. So what is a data-driven exception, then? The article doesn’t have a full definition of a data-driven exception. Although, It can be understood that a data-driven exception is an exception made to when a data-driven ruling should be used. It is a case where a data-driven ruling may not be as reliable, there might be more concerns within the specific case for which it might be beneficial to deviate from the process and have further review from a human source. An exception is not the same as an error. While an error may sometimes be cause for an exception, an exception is not always an error. That is because an error indicates that something is wrong, there was a mistake made. An exception may just need more attention to ensure that a mistake is not made.
A data-driven ruling is focused on fact, on data, on numbers. It is a ruling based off of evidence fed through the algorithm. A simple algorithm does not have the same attributes that a person does. An algorithm cannot take into account feelings like a human can, this is both a positive and a negative attribute. The article points out that while humans make mistakes, it is done in a diverse way. Whereas, with data-driven rulings, the mistakes may be more consistent within the algorithm and possibly even systemic, causing long-term harm.
Individualization is a key part of of the framework that “can be used to determine whether a data-driven rule respects or infringes on an individual’s right to be an exception” within data-driven rulings. So, what is individualization and why is it important? Individualization is adapting a rule so that it works werll within the specific case that is being considered or examined. It ensures that the algorithm is suited for the case that may need extra attention. That’s what makes it important, it allows for data-driven rulings to still be used in cases where it may have previously been more worrisome to use. However, that is not to say that there do not exist downsides of individualization. Downsides include privacy concerns as well as the current data-driven methods which are unable to individualize in the same way as humans.
When evaluating data-driven rules, it is important to also consider uncertainty. Uncertainty is a key factor in upholding the correct ruling. Having uncertainty in a ruling means adding extra concern and care into what the decision in the end, is. There are two types of uncertainty that are at play here: Epistemic and Aleatoric. Epsistemic uncertainty is reducible, it is prevalent there is a lack of knowledge at play, and is systemic. On the other hand, aleatoric uncertainty cannot be reduced. It is statisitical and comes from randomness that occurs in an event. Only one of these types can bw removed, epistemic. The article does mention that fully eliminating uncertainty is not the solution. Rather, the article suggests that “the levels of individualization and certainty be balanced against the risk of harm.” Ultimately, the article concludes that no matter how much an individualized ruling be, the decision to inflict harm should follow if and only if there is a strong level of certainty that coexists. Certainty is the most important aspect to consider.
Discussion Questions: How can it be determined whether to use a data-driven ruling versus when to use a human based ruling? What might that determination process look like? What scenario or case should data-driven rulings be drawn the line at? Is it ethical for these to be used in court? Is is ethical for banks to use them? How can ethical frameworks help develop data-driven rulings to ensure they work as well as they can for as many people as possible?
Overall, I enjoyed this excercise. I think coming up with discussion questions, although it was challenging, encouraged me to think more critically about the reading in a way that I may not have otherwise. It also furthered my understanding of the reading. I learned from the assignment and am glad with the way that it was set up.