From the course: Learning XAI: Explainable Artificial Intelligence

Bias in AI and its impact

From the course: Learning XAI: Explainable Artificial Intelligence

Start my 1-month free trial Buy for my team

Bias in AI and its impact

“

- Bias in AI. Many AI practitioners hear the word bias and they immediately take pause. The idea of bias in AI is an arguing point for many leaders across the industry. If you're taking this course, you're likely the type of leader who's aware that bias in AI happens, and who wants to learn the tools to mitigate it. Well, I'm glad you're here. Let's lay foundation for this course by understanding what we mean exactly when we talk about this topic. Bias in AI is when an AI model returns a skewed, unfavorable, unfair, or unrepresentative output, likely due to underlying human-created data with these biases. While many of these biases are reflection of the society that the data comes from, the nature of how these probabilistic AI models behave, that is models that make predictions based on probability, makes bias matching more amplified. The impacts that bias in AI can have on the public are over-reaching. Generative AI, which is the focus for this course, is particularly rife with biased outputs that can have long-term effects on users, businesses, and their customers. Examples can be as extreme as parroting racist slurs or creating harmful deep fakes targeting marginalized individuals or even seemingly innocuous as not saying the name David Mayer. Do you all remember that? This happened due to an internal flag making it impossible for anyone with that name to get service on a particular AI platform, or even text generators that perpetuate gendered stereotypes in business communications and image generators that reinforce limited cultural representations. These biases can deeply impact how users express themselves and how their work is perceived. So how does that impact you? When building generative AI systems, it's important to keep your end user in mind. Understand how bias may harm your customers, even if that bias is unintended. Thankfully, many responsible AI leaders have worked really hard to determine and refine ways to mitigate these biases so that it can be less likely to show up in a deployed model. Researchers like Dr. Timnit Gebru, Dr. Alex Hanna, and Dr. Joy Buolamwini, just to name a few, dedicate their careers to uncovering bias in generative AI and coming up with mitigation techniques. Together we'll incorporate these techniques from some of the greats and ensure that our generative AI applications work fairly for all of us.

Contents