Towards more ‘fair’ AI: algorithmic intent in a disparate society — part 6

Can ‘fair’ algorithms help address unfair policies offline? Part 6 of 9 (let’s say…)

Algorithms are axiomatic.

To review our discussion about algorithmic attempts at ‘fairness’ as this relates to European legislation today thus far:

1. Algorithms use data to establish ground truths and make predictions.
2. Useful data is SMART and ephemeral, making context and perspective critical.
3. Context and perspective are by definition limited as well as limiting.
4. Law is constantly susceptible to human inconsistencies in interpretation as well as misunderstandings, miscommunications, and, of course, plain old human disagreement.

Algorithms themselves cut through the antagonistic fog created by mixing all of the above. Algorithm logic is axiomatic — algorithms solve a selected issue as efficiently as possible using the data at their immediate disposal.

This can cause problems when, for instance, an algorithm amplifies a pre-existing bias. Algorithms are, for now, indifferent to normative arguments; they sort through pre-existing data and make calculations based on what has been prioritized in that data. This makes sense if the algorithm sorts…