Shomron Jacob,
Iterate.ai, Head of Applied Machine Learning & Platform
Biases in AI development pose significant challenges to achieving inclusive outcomes. This presentation explores the sources of bias, their impact on AI efforts, and how to avoid them.
One significant factor contributing to bias is the reliance on datasets like LAION-5B, which may inadvertently amplify social biases due to flawed automated filtering processes. The complexities of these tasks and the diversity of cultural perspectives can also hinder the accuracy of classifiers. Gartner predicts that up to 85% of AI projects will produce false results due to bias until 2030...further emphasizing the critical need to eliminate biases in training data and algorithms.
This AI DevWorld session highlights recommended anti-bias AI frameworks, such as the Aletheia Framework from Rolls Royce, Deloitte's AI framework, and Naveen Joshi's AI framework, all of which provide comprehensive practices for developing trustworthy and unbiased AI solutions.
The session will also discuss using bias-detection toolkits, like IBM's AI Fairness 360 as an open-source toolkit for detecting, reporting, and mitigating bias and discrimination in ML models (among others).
By embracing anti-bias frameworks and utilizing bias-detection toolkits, businesses can proactively address AI bias, minimize embarrassing brand damage, and ensure accurate contextualization in their AI-driven applications.
