DIKU Bits: Navigating the Right to Explanation in AI: Methods and Technical Challenges

Speaker

Pepa Kostadinova Atanasova, Postdoc in the Natural Language processing (NLP) section

Title

Navigating the Right to Explanation in AI: Methods and Technical Challenges

Abstract

As machine learning "black boxes" increasingly influence critical decisions, the development of methods to reveal their decision-making processes is crucial. This raises an important question: Are the current explanation methods sufficient to meet this growing demand?

This talk will present diagnostic methods that uncover vulnerabilities and shortcomings of explanation methods. We will particularly study examples of the faithfulness of free-text explanations produced by language models.

Which courses do you teach?

(BSc and MSc) – I have previously thought/was a teaching assistant for Introduction to Natural Language Processing and Fair and Transparent Machine Learning Methods.

Which technology/research/projects/startup are you excited to see the evolution of?

I’d be excited to see the evolution of robust and accurate explainable AI.