The Dual Role of Abstracting over the Irrelevant in Symbolic Explanations: Cognitive Effort vs. Understanding

TL;DR

The paper investigates how to render symbolic explanations in Answer Set Programming (ASP) more understandable by abstracting over irrelevant details. It introduces a formal notion of $\chi$-irrelevance, with two operations—Removal (mapping atoms to $\top$) and Clustering (mapping atoms to a common representative)—to produce simplified explanations. An empirical study across three domains shows that clustering improves accuracy while removal reduces cognitive effort, confirming a double benefit of abstraction for human-centered symbolic AI. The work demonstrates that ASP-based explanations can be aligned with cognitive processes, providing practical guidance for explainable AI in high-stakes settings.

Abstract

Explanations are central to human cognition, yet AI systems often produce outputs that are difficult to understand. While symbolic AI offers a transparent foundation for interpretability, raw logical traces often impose a high extraneous cognitive load. We investigate how formal abstractions, specifically removal and clustering, impact human reasoning performance and cognitive effort. Utilizing Answer Set Programming (ASP) as a formal framework, we define a notion of irrelevant details to be abstracted over to obtain simplified explanations. Our cognitive experiments, in which participants classified stimuli across domains with explanations derived from an answer set program, show that clustering details significantly improve participants' understanding, while removal of details significantly reduce cognitive effort, supporting the hypothesis that abstraction enhances human-centered symbolic explanations.

Figures (5)

• Figure 1: Blocksworld example
• Figure 2: Explanations obtained by xclingo
• Figure 3: Example of a learning phase instance as presented in the study. The table lists all attributes and values. The provided explanation is the default without any abstraction.
• Figure 4: Study Procedure Overview. The order of the three domains and the test phase, i.e. the order of the positive (illustrative in green) and negative (resp. in red) stimuli were random to balance out possible position effects.
• Figure 5: Mean accuracy (left) and mean answering times (right) across the abstraction levels. Error bars represent $\pm$1 SEM. Significant effect observed for pairwise comparison between the groups are marked with $\star$ ($p < 0.05$) and $\star\star$ ($p < 0.01$)

Theorems & Definitions (5)

• Example 1
• Example 2: Ex. \ref{['ex:block']} cont.
• Definition 1: $\chi$-irrelevance
• Example 3
• Example 4: Ex. \ref{['ex:toy']} ctd