Algorithm Ethics

Talk by Majken Sanders and Joerg Blumtritt given at Strata + Hadoop World Conference 2015, London.

Ethics is how ‘to decide in a morally right way’. Algorithms are usually regarded as something deterministic and mathematical, not to contain ethics: Eratosthenes’ sieve, for example, will give you all prime numbers up to a given maximum. Every other prime-checking algorithm will come to the same solution. A number is prime or not.

But there is a different kind of algorithm that is far more common in our daily life: calculations to find a solution for some task that other people might have done differently and with different outcomes. These algorithms contain value judgments, choices, or decisions made on how to deal with tasks according to social, cultural, or legal rules or personal persuasion. Obvious examples are credit scoring or pricing of a retail offer. However, there are a multitude of hidden ethical algorithms that are far more pervasive. When an ad network’s targeting system selects which ads we see and which we don’t, we might not find that too important. But a search engine deciding what it regards as relevant to us affects the information we see and what we miss. And medical images like MRIs might even affect our life with their many implicit parameters that are not visible to the physician.

There are three basic types of value judgments in algorithms: 1) Choosing a method, 2) Setting parameters, 3) Deciding how to deal with uncertainty and misclassification. All three judgments are quite regularly not made explicit. For many applications, the only way to understand these presumptions is to “open the black box” – hence to hack.

We will present some of these value judgments, discuss their consequences, and propose a cause to deal with them on a personal as well as on a business level.