Justice and Fairness in Data Use and Machine Learning

The Information Ethics Roundtable (IER) is a yearly conference that brings together researchers from disciplines such as philosophy, information science, communications, public administration, anthropology, and law to discuss ethical issues such as information privacy, intellectual property, intellectual freedom, and censorship.

The 17th annual Information Ethics Roundtable will explore the relationship between the normative notions of justice and fairness and current practices of data use and machine learning.

Artificial intelligence is now a part of our everyday lives. It allows us to easily find get to a place we have never been before, while avoiding traffic and road work, to communicate with our Chinese friend when we don’t share a common language, and to carry out complex but mind numbing repetitive jobs in factories. But such artificial intelligences can also exhibit what we might call “artificial bias;” that is, machine behavior that, if produced by a person, we would say is biased against particular groups, such as racial minorities. Machine learning using large data sets is one means of achieving AI that is particularly vulnerable to producing biased systems, because it uses data from human behavior that is itself biased. A number of tech companies, such as Google and IBM, and computer science researchers are currently seeking ways to correct for such biases and to produce “fair” algorithms. But a number of fundamental questions about bias, fairness, and even justice still need to be answered if we are to solve this problem. Such as:

• What concepts of fairness and justice in philosophy and other disciplines are most useful for understanding fairness, equality, and justice in data use and machine learning?
• To what extent is it possible to operationalize (or computationalize) different conceptions of fairness and justice within different machine learning techniques?
• Should machine learning based decision-making systems be held to a higher or different standard of fairness and justice before being implemented in industry (e.g. lending) or social services (e.g. child protective services) in comparison to currently accepted practices?
• What is the role of data scientists and computer programmers in correcting for bias? How can machine learning be used in this role?
• Not all biases are problematic; indeed, some are very helpful. What sorts of bias are unjust and why?
• What can modern day programmers of “classifications” learn about avoiding bias from the experience of other disciplines devoted to classification, such as librarianship?
• What can normative research in other areas – for example, with respect to police profiling or immigration/refugee screening – teach about when or under what conditions profiling with machine learning is acceptable?
• What is the relationship between explainability/interpretability in machine learning decision-making and the just use of machine learning in different contexts?

Prior registration is required for attendance at this event. However, registration is free and the conference is open to the public. Thus, we invite you to attend, regardless of whether or not you are formally workshopping or discussing a paper.