Paper 1159-2021
Author
Jason Brinkley - Abt Associates
Abstract
The field of algorithm development for machine learning/artificial intelligence has seen rapid changes in recent years. And the rise in algorithm development has increased concerns of rampant bias in implementation. Many businesses turned to AI as a mechanism to create or evaluate things faster and cheaper with more accuracy than human beings. This hyper-focus on speed and accuracy has created numerous instances of AI with biases; there are many examples in the news and research articles of algorithms that have been shown to have strong gender or racial biases. There is a tug of war going on between the data science and policymaker communities around the need to better evaluate and audit AI to discover such biases before widespread rollout. The main issue is the need for methodological innovations and our existing tools are not well-suited for evaluating AI. We will spend some time looking at the algorithm landscape and discuss novel innovations, such as humble, fair or interpretable AI in a way that is digestible to a general audience. The focal point of the talk is on determining what we need to ask of AI in the future. The new world of asking AI to make predictions through a lens of equity necessitates the need for new evaluation measures that move beyond these original goals. The talk is designed for both scientists and policymakers.
Watch the Presentation
Watch Auditing Algorithms: The Need for New Tools to Combat Bias in Artificial Intelligence as presented by the author on the SAS Users channel on YouTube.
Click image to register for webinar