This work presents a comprehensive literature review of the current state of the subfield of XAI that consist of causality-motivated post-hoc XAI methods, and a causal framework for categorising XAI is introduced, and three types of post-hoc XAI methods are identified: observational methods, internally causal methods and externally causal methods.
Abstract
Today, the use of Artificial Intelligence (AI) is rapidly increasing in many areas of society. While model performance on various tasks continue to impress, it does so at the cost of increased model complexity, such that most state-of-the-art AI models are effectively black boxes. Where human-made decisions typically are accompanied by human-understandable explanations detailing the reasoning behind the decision, incorporating advanced AI as part of a decision-making process reduces the transparency of that process significantly. Yet, the ability to explain decisions is essential for there to be understanding and trust. As a response to this, Explainable Artificial Intelligence (XAI) has emerged as a field that aims to provide explanations of model behaviour. Methods categorised as post-hoc are designed to generate explanations for black box models after training, at no cost to model performance. In parallel with this, extensive work has been done in the field of causality to formalise the structure of human-understandable, causal explanations. This work presents a comprehensive literature review of the current state of the subfield of XAI that consist of causality-motivated post-hoc XAI methods. In order to clearly define causal XAI, a causal framework for categorising XAI is introduced, and three types of post-hoc XAI methods are identified: observational methods, internally causal methods and externally causal methods. Finally, externally causal XAI is argued a promising direction for reliable and understandable post-hoc XAI, with the ability to generate counterfactual explanations using a meaningful vocabulary, in line with the definition of counterfactual used in causal theory.
An analytical model is developed that incorporates the defining features of human and machine intelligence, capturing the limited but flexible nature of human cognition with imperfect machine recommendations, and represents how AI-based explanations influence the DM’s belief in the algorithm’s predictive quality.
Tamer Boyacı, Caner Canyakmaz, Francis de Véricourt· Management Sciences· 0 citations
Through the application of Artificial intelligence (AI), there has been the establishment of the technology as a cornerstone in current decision systems in essential fields like health, finance, transport, industrial automation, and the governance of citizens. Although traditional AI and machine learning frameworks, specifically deep learning models, have shown outstanding predictive performance, their nature is not enlightened by default, and as such, they have cast considerable doubt on the issues of trust, accountability, fairness, and regulatory compliance. It is thanks to this limitation that Explainable Artificial Intelligence (XAI) as a paradigm has emerged, aimed at ensuring that AI-driven decisions are easy to understand and interpret by the human stakeholders without major performance reduction. This paper is a thorough and stepwise analysis of how XAI was created in current decision systems. It starts with the historical contextualization of the development of AI, as an expert system driven by rules, to a data-driven black-box model and the increasing demand to explain decision-making processes. The paper provides a critical literature review of the current research on XAI methods classifying them into model-intrinsic and post-hoc methods of explanation, and discussing their relevance to various fields. An elaborate methodology is suggested, that incorporates explainability protocols into the AI choice channel, such as information pre-processing, model order, explanation creation and human-centered assessment. Additionally, the paper evaluates the experimental findings and case-based debates on how XAI enhances transparency, end-user trust, compliance with regulations, and system resilience. Popular explainability methods are also compared and evaluated including SHAP, LIME, saliency maps, and rule extraction. The results indicate that explainable models, in addition to increasing interpretability, can also help to improve debugging, bias detection and ethical AI deployment. The paper ends by recommending the current challenges, areas of open research, and future roles of XAI in the development of responsible and human-centered intelligent decision systems.
Meena Krishnan· International Journal of Mod...· 0 citations
Explainable Artificial Intelligence (XAI) seeks to explain how an Artificial Intelligence (AI) system arrived at a particular decision. We propose''Rule of Thumb''(RoT) explanations, a new approach to XAI based upon a novel formulation that identifies the most relevant features for predicting the behaviour of an AI system, for a particular datapoint. We show how RoT is well-suited to enable XAI in: (a) zero-shot classification using large language models (LLMs), (b) auditing of opaque AI systems without model access, and (c) the use of AI in scientific discovery. Additionally, RoT meets specific requirements from leading AI regulations, provides a familiar interface and visualisations for XAI practitioners, is model-agnostic, and is substantially faster than alternatives. Code available at: https://github.com/KaiRawal/Rule-of-Thumb-Explaining-Artificial-Intelligence-Systems-using-Partial-Information
Kai Rawal, D. Onitiu, Brent Mittelstadt et al.· 0 citations
XAI provides a powerful framework for responsible AI development, challenges such as the performance-interpretability trade-off, lack of standardized evaluation metrics, and potential for human misinterpretation remain areas of active research.
P. Pradhan, Amol Rajmane, C. patil· Journal of image processing...· 0 citations
This work explored powerful explainers which use rules, where explanation instability stemming from training data becomes more apparent, and found that the rule-based method employed, BARBE, sharply increased in fidelity and stability when trained with the modified process, making BARBE+PBC which exceeded other methods that improve stability like S-LIME and LORE.