Abstract: The stories of bias in AI are everywhere: Amazon's recruiting tool, Apple's credit card limits, Google's facial recognition, and dozens more. The quick solution is just to blame the algorithm and its designers. However, as data scientists, its incumbent on us to understand the true source of the bias and improve the underlying process.
AI does not create bias alone; it exposes the latent bias present in the system it was designed to imitate. We need to reframe the conversation around bias in AI to instead identify it as the first step in building a more ethical system.
In this talk, we show how machine learning can make the implicit bias of a human institution explicit. Bias becomes diagnosable, correctable, and ultimately preventable in a way that cannot be replicated in human decision-making, which is opaque and difficult to change. Bias is not new, but AI represents a new toolset to measure and change it.
The goal is not only to provide you a theoretical understanding of bias, but a practical plan that you can start to implement right away. After all, it’s not whether or not you have bias in your institution, but how you plan to handle it.
Bio: Jett Oristaglio is the Data Science and Product Lead of Trusted AI at DataRobot. He has a background in Cognitive Science, with focuses in computer vision, neuro-ethics, and transcendent states of consciousness. His primary mission at DataRobot is to answer the questions: "What is everything you need in order to trust a decision-making system with your life? And what tools can we build to automated that process as comprehensively as possible?"