Control variable strategies can go wrong and controls can make estimates worse rather than better. In this chapter, we discuss why control variables always deserve skepticism and require specific justification. We discuss criteria for plausible controls, bad controls, and why one should not control for everything we can measure (aka kitchen sink models). There are common conditions in which controlling for a variable causes bias, and therefore when excluding that control variable reduces bias.

An extensive literature details how businesses and governments use the information they collect. We focus on the relatively recent use of artificial intelligence (machine learning, predictive analytics) to analyze the information. We focus on two features of AI-driven surveillance. Broad-based predictions: What happens in virtually any area of one’s life (what stores you shop in, what you post on social media, and so on) may serve as input to predictions affecting virtually any other area (what you pay for insurance, whether you get hired, what schools you get in to, and so on). Data feedback without error correction: AI-driven decisions affect what happens to people in the future, and that information feeds back into the systems as input for subsequent decisions. Current AI systems often lack mechanisms for effective detection and correction of errors. These two features create fairness problems.

We show how to create an informational norm that constrains the use of proxies in AI-driven surveillance. The norm creates privacy in public by implementing conditions on the flow of information that control the use of that information as proxy variables. There are a variety of possible conditions. The task is to choose one that implements an acceptable tradeoff between informational privacy and information processing. The norm we propose ensures that the use of proxies is fair and to that extent implements an acceptable tradeoff between informational privacy and information processing. There is, of course, more to finding such a tradeoff than just ensuring that the use of proxies is fair, and the process we describe creates a forum for addressing tradeoff issues in general. Our focus for the moment, however, is on fairness. The example of Sally from Chapter 2 illustrates the questions that arise. Sally’s car insurance premiums increase as a result of her bankruptcy.