C16: Identify effective explanations – a hands-on course

Abstract

Artificial intelligence (AI) is increasingly integrated into consumer and business applications. A prominent area within AI, machine learning (ML), brings significant advantages but also introduces a key challenge: uncertainty surrounding the correctness of its outcomes from a human perspective. To foster trust in these systems, it is essential to provide clear and traceable explanations that support the human to understand why specific outcomes are determined.

However, identifying "effective" explanations is not straightforward. In this context, "effective" refers to explanations that are tailored to the needs and goals of a specific user group. Currently, there is no widely accepted framework or methodology for systematically identifying such tailored explanations.

This course introduces a methodology specifically designed to address this gap by helping participants develop user-centric explanations for AI-based systems. The methodology, refined over several years of industrial research, offers a hands-on, step-by-step approach, guiding participants through the process of identifying effective explanations for a particular target audience. By the end of the course, participants will have identified the explanations with its structure and content needed and applied in selected ML-based applications fostering the trust of the selected target user group. We refer here to the structured explanation content as a mental model for explanations—which can serve as the foundation for UX design.

We encourage participants to come to the course with their own application and target user group in mind. Given the practical nature of the course, it is recommended that participants from the same organization join as a small team to maximize collaboration during the exercises and produce a useful result. Exercises are conducted in groups of 3 – 4 participants.

Course Content

The objective of the course is the creation of a mental model for explanations for an application and a target user group for each participant. 

• Introduction (Uncertainty premise, reasons for uncertainty)
• Understanding the explanation model
• Understanding the method to identify and evaluate effective explanations
• Step 1: Specify application and target user group
• Step 2: Identify user goals
• Step 3: Identify user tasks
• Step 4: Mark the user tasks that require explanations
• Step 5: Identify inputs and outputs per user task
• Step 6: Answer backward-looking explainability questions
• Step 7: Answer forward-looking explainability questions
• Step 8: Create the mental model for explanations
• Step 9: Group presentation and wrap-up