Towards Ethical Intelligence: Bias Mitigation in AI

The second part of a two talk series...

Zoom Link
[https://us06web.zoom.us/j/84725585840?pwd=IH8LxF2MlgYO4doCvV6eGvi3dNjh1v.1](https://www.google.com/url?q=https://us06web.zoom.us/j/84725585840?pwd%3DIH8LxF2MlgYO4doCvV6eGvi3dNjh1v.1&sa=D&source=calendar&ust=1709648158635758&usg=AOvVaw2LJq0Y-0cYeGkU6vIasqR2)

Summary
As artificial intelligence (AI) continues to shape our technological landscape, it is crucial to address the inherent biases that may manifest within AI systems. While it is common knowledge that data is biased more often than not, few accept to acknowledge the bias in their predictions, and fewer try to mitigate it.

The session commences with a brief historical overview of biases in AI as we will already have covered the topic during January's presentation.
The presentation then focuses on the detection and measurement of bias in classification models. Multiple approaches will be explored, ranging from assessing model performance metrics to evaluating prediction rankings and ensuring model fairness.

We will then go over some of the mitigation techniques developed as part of recent projects. This will focus mostly on mitigating consumer data and helping brands deploy fair marketing campaigns and reach underserved markets.

The presentation concludes with a broader examination of bias mitigation in other fields of AI, such as image recognition and natural language processing.

About the Speaker
Thibault Dody, Data Scientist at Faraday