Algorithmic Biases Talk

When you ask a computer a question, either through a search engine or a large language model, you usually receive a definitive answer in fractions of a second. Surety and speed can be mistaken for accuracy and impartiality. Search engines, algorithms, and large language models are susceptible to bias. The sources of the bias may be errors, bad training data, or even manipulation. We'll look at the sources, causes, and implications of algorithmic bias through the works of writers such at Roger McNamee (author of "Zucked"), Cathy O'Neil (author of "Weapons of Math Destruction"), and Dr. Joy Buolamwini (author of "Unmasking AI" and founder of The Algorithmic Justice League), as well as more recent research on Large Language Models. Algorithmic Bias appears in lending decisions, the legal system (for example, sentencing decisions), hiring decisions, facial recognition, and can appear in virtually any computerized process. The building of "data voodoo dolls" used by advertisers (described in McNamee), the use of bad training data (O'Neil), and the weaknesses of facial recognition technology (Dr. Buolamwini) all have the ability to strongly influence the directions of our lives. Yet these processes are typically opaque and unregulated, hidden behind proprietary algorithms and data sets, and potentially used without disclosure. Awareness of these risks is a vital first step to change, and the rise of Generative AI adds additional urgency.