AI Ethics, Bias, and Responsible Innovation
with Kesha Williams • March 25, 2026
Episode overview
What happens when the data you feed an AI system is already broken — and no one stops to ask why?
In this episode of The Pluralsight Podcast, Kesha Williams — AI ethicist, AWS Hero, and 30-year tech veteran — makes the case that building powerful AI systems isn't enough. Building responsible ones is the only real standard that matters.
Kesha traces her focus on AI ethics back to a single project: a crime prediction model that exposed how easily biased data can corrupt a machine learning system before a single line of code is written. From there, she breaks down the three types of bias teams face — data, algorithmic, and interpretation — why interpretation bias is the one most teams are still getting wrong, and what model drift means for organizations that think their work is done once a model ships.
We also get into AI governance in the age of agents, why the ability to roll back an AI action may be the most underrated capability in any AI stack, and what an AI Center of Excellence actually looks like in practice.
If you're building AI systems — or leading teams that do — this conversation is a practical and honest look at where things go wrong, and what it actually takes to get them right.
Want to go deeper? Check out our weekly newsletters focused on Security, Cloud, and AI.
Follow Pluralsight on Linkedin and join the conversation.
Connect directly with Kesha Williams on LinkedIn.
Questions or comments? podcast@pluralsight.com
Chapters
00:00:33 — Introduction: Kesha Williams, AWS AI Hero
00:01:05 — Kesha's 30-year journey and spotting emerging tech early
00:02:51 — The moment that changed everything: building a crime prediction model
00:04:18 — Pre-crime, Minority Report, and bias hiding in UK stop-and-search data
00:05:44 — The Clear News AI case study: how bias shapes what a nation reads
00:07:57 — The three types of bias — and why interpretation bias is now the hardest
00:09:16 — Role play: interpretation bias and the home loan example
00:11:53 — Red flags: why skipping model retraining silently reintroduces bias
00:13:21 — Favorite tools: SageMaker Clarify, AI Fairness 360, and Fairlearn
00:14:22 — SHAP and LIME: making model decisions explainable
00:15:28 — Agentic AI governance: visibility, guardrails, and rollback
00:18:09 — Accountability and the case for an AI Center of Excellence
00:20:53 — Skills engineers need to prioritize: prompt engineering and LLM literacy
00:22:37 — The mindset of learners who thrive: curiosity and innovation
00:24:32 — No-code platforms, citizen developers, and guardrails
00:25:28 — Where to find Kesha: LinkedIn and Pluralsight