Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transferring Domain Knowledge with (X)AI-Based Learning Systems

Philipp Spitzer, Niklas Kühl, Marc Goutier, Manuel Kaschura, Gerhard Satzger

TL;DR

This paper investigates transferring expert tacit knowledge to novices via an (X)AI-based learning system trained on experts' past decisions and coupled with visual explanations. In a between-subject online study with 249 participants performing mammography image classification, the XAI condition yielded higher learning performance across ten rounds, with last-round accuracy (84.65%) exceeding the baseline (82.37%), while learning times remained comparable. The study also demonstrates that cognitive style moderates the learning effect: visual learners generally perform better, but non-visual learners gain relatively more from the explanations, indicating a nuanced interaction between explanation modality and user characteristics. These findings suggest that (X)AI-based learning systems can serve as scalable alternatives to traditional training and should be tailored to individual cognitive styles to maximize knowledge transfer in high-stakes domains.

Abstract

In numerous high-stakes domains, training novices via conventional learning systems does not suffice. To impart tacit knowledge, experts' hands-on guidance is imperative. However, training novices by experts is costly and time-consuming, increasing the need for alternatives. Explainable artificial intelligence (XAI) has conventionally been used to make black-box artificial intelligence systems interpretable. In this work, we utilize XAI as an alternative: An (X)AI system is trained on experts' past decisions and is then employed to teach novices by providing examples coupled with explanations. In a study with 249 participants, we measure the effectiveness of such an approach for a classification task. We show that (X)AI-based learning systems are able to induce learning in novices and that their cognitive styles moderate learning. Thus, we take the first steps to reveal the impact of XAI on human learning and point AI developers to future options to tailor the design of (X)AI-based learning systems.

Transferring Domain Knowledge with (X)AI-Based Learning Systems

TL;DR

This paper investigates transferring expert tacit knowledge to novices via an (X)AI-based learning system trained on experts' past decisions and coupled with visual explanations. In a between-subject online study with 249 participants performing mammography image classification, the XAI condition yielded higher learning performance across ten rounds, with last-round accuracy (84.65%) exceeding the baseline (82.37%), while learning times remained comparable. The study also demonstrates that cognitive style moderates the learning effect: visual learners generally perform better, but non-visual learners gain relatively more from the explanations, indicating a nuanced interaction between explanation modality and user characteristics. These findings suggest that (X)AI-based learning systems can serve as scalable alternatives to traditional training and should be tailored to individual cognitive styles to maximize knowledge transfer in high-stakes domains.

Abstract

In numerous high-stakes domains, training novices via conventional learning systems does not suffice. To impart tacit knowledge, experts' hands-on guidance is imperative. However, training novices by experts is costly and time-consuming, increasing the need for alternatives. Explainable artificial intelligence (XAI) has conventionally been used to make black-box artificial intelligence systems interpretable. In this work, we utilize XAI as an alternative: An (X)AI system is trained on experts' past decisions and is then employed to teach novices by providing examples coupled with explanations. In a study with 249 participants, we measure the effectiveness of such an approach for a classification task. We show that (X)AI-based learning systems are able to induce learning in novices and that their cognitive styles moderate learning. Thus, we take the first steps to reveal the impact of XAI on human learning and point AI developers to future options to tailor the design of (X)AI-based learning systems.
Paper Structure (15 sections, 5 figures, 1 table)

This paper contains 15 sections, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Theoretical Development
  4. Study Design
  5. Data selection
  6. Recruitment
  7. Development of the XAI system
  8. Study design
  9. Results
  10. Impact of XAI on novices' learning
  11. Impact of cognitive style in (X)AI-based learning systems
  12. Discussion and Limitation
  13. Implications for research and practice
  14. Limitations and outlook
  15. Conclusion

Figures (5)

  • Figure 1: Research model.
  • Figure 2: Example of an image without visual explanation (left) and the same image with corresponding explanation (right). The explanation highlights the relevant region with cancer cells.
  • Figure 3: Comparison of the learning curves in both conditions.
  • Figure 4: Comparison of the learning times in both conditions.
  • Figure 5: Interaction effect of visual cognitive style with example-based learning.