The Faces Behind the Data Points: Combating AI Bias in Product Development

Summary: Companies have increasingly leveraged AI to build data products. While this can potentially lead to many exciting innovations, there may be biases in AI. One problem is with commonly-used facial analysis algorithms, which tend to fail more often specifically on female faces of color. In the presentation, I outline: *Examples of how AI biases led to failures in products *Why AI is often biased *The results of an audit on facial analysis / object detection algorithms I conducted, in which I found discrepancies in the detection of black female faces *5 ways companies can combat AI bias in product development, in order to ensure high quality data products.