Algorithmic Accountability

In the previous section I tried to articulate some of the myriad ways that algorithms can exert power though decisions they make in prioritizing, classifying, associating, and filtering information. This inspires various questions we might use as a basis for beginning to investigate an algorithm:

• What is the basis for a prioritization decision? Is it fair and just, or discriminatory?

• What are the criteria built into a ranking, classification, or association, and are they politicized or biased in some consequential way? What are the limits to measuring and operationalizing the criteria used by the algorithm?

• What are the limits of an algorithm and when is it known to break down or fail? For instance: What types of errors are made in classification? How has the algorithm been tuned to privilege false positive or false negative errors? Does that tuning benefit one set of stakeholders over another? What are thresholds used in classification decisions? What kind of uncertainty is there in the classifier?

• What are the potential biases of the training data used in a classifying algorithm? How has the algorithm evolved with that data? What types of parameters or data were used to initiate the algorithm?

• How are the semantics and similarity functions defined in an association algorithm? Do those definitions have implications for the interpretation or connotation of those associations?

• Are there some pieces of information that are differentially over- emphasized or excluded by the algorithm? What are the editorial criteria of the algorithm and is such filtering warranted? What are the implications of that filtering?

  From the list of questions above it should be clear that there are a number of human influences embedded into algorithms, such as criteria choices, training data, semantics, and interpretation. Any investigation must there- fore consider algorithms as objects of human creation and take into account intent, including that of any group or institutional processes that may have influenced their design.

  It’s with this concept in mind that I transition into devising a strategy to characterize the power exerted by an algorithm. I’ll start first with an examination of transparency, and how it may or may not be useful in characterizing algorithms. Then I’ll move into how you might employ reverse engineering in the investigation of algorithms, including both theoretical thinking and practical use cases that illustrate the technique. I conclude the section with certain methodological details that might inform future practice in developing an investigative reporting “beat” on algorithms, including issues of how to identify algorithms for investigation, sample them, and find stories.