What Is a Proxy and Why Is It a Problem?

Concerns with so-called ‘proxies’ have become ubiquitous in recent debates about algorithmic fairness. Yet it is far from clear what makes something a proxy and why it poses a problem. In this talk, Solon Barocas will differentiate between two different notions of proxies: One concerned with the choice of prediction target and the other concerned with the choice of input features. In the first case, the concern is that a chosen target can be a systematically biased measure — that is, a poor proxy — for the true outcome of interest and that even accurately predicting the target can be unfair. In the second case, the concern seems to be that even when a legally proscribed feature is not provided directly as an input into a machine learning model, discrimination on that basis may persist because non-proscribed features are correlated with — that is, serve as a proxy for — the proscribed feature. After disentangling these two different notions, Barocas will show that it can be surprisingly difficult, in both cases, to pin down whether something is a proxy and whether that should be cause for concern.

To attend this in-person event, you will need to register in advance and provide:

• Acceptable proof of vaccination (vaccine card/certificate, a copy or photo of vaccine card/certificate or electronic NYS Excelsior Pass or NJ Docket Pass)
• Photo ID
• Eventbrite ticket confirmation email with QR code
• Simons Foundation Health Screening Questionnaire approval email

Guests are expected to complete these requirements each time they visit the Simons Foundation and entrance will not be granted without this documentation.

On-site registration will not be permitted. Walk-in entry will be denied.

Barocas is a principal researcher in the New York City lab of Microsoft Research and an adjunct assistant professor in the department of information science at Cornell University. His research explores ethical and policy issues in artificial intelligence, particularly fairness in machine learning, methods for bringing accountability to automated decision-making, and the privacy implications of inference. He co-founded the ACM conference on Fairness, Accountability, and Transparency (FAccT).