Defining Explanation and Explanatory Depth in XAI

Buijsman, Stefan

doi:10.1007/s11023-022-09607-9

Cited by 20 publications

(14 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As Rodu and Baiocchi write, "some of the most successful algorithms are so complex that no person can describe the mathematical features of the algorithm" [54]. This has led to the rise of a whole movement in AI, called explainable AI or XAI [55][56][57][58], best reflected in DARPA's large project with the same name [59].…”

Section: Explainable Ai Vs Black-box Aimentioning

confidence: 99%

Identity of AI

Devedžić

2022

Discov Artif Intell

View full text Add to dashboard Cite

With the explosion of Artificial Intelligence (AI) as an area of study and practice, it has gradually become very difficult to mark its boundaries precisely and specify what exactly it encompasses. Many other areas of study are interwoven with AI, and new research and development topics that require interdisciplinary approach frequently attract attention. In addition, several AI subfields and topics are home to long-time controversies that give rise to seemingly never-ending debates that further obfuscate the entire area of AI and make its boundaries even more indistinct. To tackle such problems in a systematic way, this paper introduces the concept of identity of AI (viewed as an area of study) and discusses its dynamics, controversies, contradictions, and opposing opinions and approaches, coming from different sources and stakeholders. The concept of identity of AI emerges as a set of characteristics that shape up the current outlook on AI from epistemological, philosophical, ethical, technological, and social perspectives.

show abstract

Section: Explainable Ai Vs Black-box Aimentioning

confidence: 99%

Identity of AI

Devedžić

2022

Discov Artif Intell

View full text Add to dashboard Cite

show abstract

“…"), and why-questions (e.g., "Why did event that happen?"). Along those lines, Buijsman (2022) reports the properties a good explanation should have: first, a rule answering why we got a specific output, and second a counterfactual component aimed at answering why X occurred rather than Y. Furthermore, Buijsman also conceptualised the depth of an explanation in terms of the abstractness of variables and generality.…”

Section: Explanations In Philosophymentioning

confidence: 99%

“…order to provide a good explanation (Buijsman 2022). From the philosophy of science literature, it is possible to derive that a satisfactory explanation should have two components such that it provides answers to contrastive whyquestions (Woodward 2003).…”

Section: Introductionmentioning

confidence: 99%

“…Several explanation methods have already been proposed but they only focus on one of the two aspects argued by Woodward (2003) and Buijsman (2022). Approaches like LIME (Ribeiro, Singh, and Guestrin 2016) and Grad-CAM (Selvaraju et al 2020), aim to answer the why aspect of explanations by finding which regions within images a model regards as more important.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

CHIME: Causal Human-in-the-Loop Model Explanations

Biswas

Corti

Buijsman

et al. 2022

HCOMP

View full text Add to dashboard Cite

Explaining the behaviour of Artificial Intelligence models has become a necessity. Their opaqueness and fragility are not tolerable in high-stakes domains especially. Although considerable progress is being made in the field of Explainable Artificial Intelligence, scholars have demonstrated limits and flaws of existing approaches: explanations requiring further interpretation, non-standardised explanatory format, and overall fragility. In light of this fragmentation, we turn to the field of philosophy of science to understand what constitutes a good explanation, that is, a generalisation that covers both the actual outcome and, possibly multiple, counterfactual outcomes. Inspired by this, we propose CHIME: a human-in-the-loop, post-hoc approach focused on creating such explanations by establishing the causal features in the input. We first elicit people's cognitive abilities to understand what parts of the input the model might be attending to. Then, through Causal Discovery we uncover the underlying causal graph relating the different concepts. Finally, with such a structure, we compute the causal effects different concepts have towards a model's outcome. We evaluate the Fidelity, Coherence, and Accuracy of the explanations obtained with CHIME with respect to two state-of-the-art Computer Vision models trained on real-world image data sets. We found evidence that the explanations reflect the causal concepts tied to a model's prediction, both in terms of causal strength and accuracy.

show abstract

“…In recent work, Buijsman (2022) also considers the Pearl-Woodward framework in the context of XAI, including its relationship to standard counterfactual approaches. This important work differs from the study undertaken here in a couple of ways.…”

Section: Introductionmentioning

confidence: 99%

Explainable AI and Causal Understanding: Counterfactual Approaches Considered

Baron

2023

Minds & Machines

View full text Add to dashboard Cite

The counterfactual approach to explainable AI (XAI) seeks to provide understanding of AI systems through the provision of counterfactual explanations. In a recent systematic review, Chou et al. (Inform Fus 81:59–83, 2022) argue that the counterfactual approach does not clearly provide causal understanding. They diagnose the problem in terms of the underlying framework within which the counterfactual approach has been developed. To date, the counterfactual approach has not been developed in concert with the approach for specifying causes developed by Pearl (Causality: Models, reasoning, and inference. Cambridge University Press, 2000) and Woodward (Making things happen: A theory of causal explanation. Oxford University Press, 2003). In this paper, I build on Chou et al.’s work by applying the Pearl-Woodward approach. I argue that the standard counterfactual approach to XAI is capable of delivering causal understanding, but that there are limitations on its capacity to do so. I suggest a way to overcome these limitations.

show abstract

Defining Explanation and Explanatory Depth in XAI

Cited by 20 publications

References 42 publications

Identity of AI

Identity of AI

CHIME: Causal Human-in-the-Loop Model Explanations

Explainable AI and Causal Understanding: Counterfactual Approaches Considered

Contact Info

Product

Resources

About