Abstract In this paper, we introduce a post-hoc and local explainable AI method tailored for Knowledge Graph Embedding (KGE) models. These models are essential to Knowledge Graph Completion yet criticized for their black-box nature. Despite their success in capturing the semantics of knowledge graphs through high-dimensional latent representations, their inherent complexity poses substantial challenges to explainability. While existing methods like Kelpie use resource-intensive perturbation to explain KGE models, our approach directly decodes the latent representations encoded by KGE models, leveraging the smoothness of the embeddings, which follows the principle that similar embeddings reflect similar behaviours within the Knowledge Graph, meaning that nodes are similarly embedded because their graph neighbourhood looks similar. This principle is commonly referred to as smoothness. By identifying symbolic structures, in the form of triples, within the subgraph neighborhoods of similarly embedded entities, our method identifies the statistical regularities on which the models rely and translates these insights into human-understandable symbolic rules and facts. This bridges the gap between the abstract representations of KGE models and their predictive outputs, providing clear and interpretable insights. The contributions include a novel post-hoc and local explainable AI method for KGE models, which provides immediate and faithful explanations without retraining, thereby facilitating real-time application on large-scale knowledge graphs. The method’s flexibility enables the generation of rule-based, instance-based, and analogy-based explanations, meeting diverse user needs. Extensive evaluations show the effectiveness of our approach in delivering faithful and well-localized explanations, enhancing the transparency and trustworthiness of KGE models.

In contrast to vast academic efforts to study AI security, few real-world reports of AI security incidents exist. Released incidents prevent a thorough investigation of the attackers' motives, as crucial information about the company and AI application is missing. As a consequence, it often remains unknown how to avoid incidents. We tackle this gap and combine previous reports with freshly collected incidents to a small database of 32 AI security incidents. We analyze the attackers' target and goal, influencing factors, causes, and mitigations. Many incidents stem from non-compliance with best practices in security and privacy-enhancing technologies. In the case of direct AI attacks, access control may provide some mitigation, but there is little scientific work on best practices. Our paper is thus a call for action to address these gaps.

Despite the large body of academic work on machine learning security, little is known about the occurrence of attacks on machine learning systems in the wild. In this paper, we report on a quantitative study with 139 industrial practitioners. We analyze attack occurrence and concern and evaluate statistical hypotheses on factors influencing threat perception and exposure. Our results shed light on real-world attacks on deployed machine learning. On the organizational level, while we find no predictors for threat exposure in our sample, the amount of implement defenses depends on exposure to threats or expected likelihood to become a target. We also provide a detailed analysis of practitioners’ replies on the relevance of individual machine learning attacks, unveiling complex concerns like unreliable decision making, business information leakage, and bias introduction into models. Finally, we find that on the individual level, prior knowledge about machine learning security influences threat perception. Our work paves the way for more research about adversarial machine learning in practice, but yields also insights for regulation and auditing.

The interest in explainable artificial intelligence has grown strongly in recent years because of the need to convey safety and trust in the ‘how’ and ‘why’ of automated decision-making to users. While a plethora of approaches has been developed, only a few focus on how to use domain knowledge and how this influences the understanding of explanations by users. In this paper, we show that by using ontologies we can improve the human understandability of global post-hoc explanations, presented in the form of decision trees. In particular, we introduce Trepan Reloaded, which builds on Trepan, an algorithm that extracts surrogate decision trees from black-box models. Trepan Reloaded includes ontologies, that model domain knowledge, in the process of extracting explanations to improve their understandability. We tested the understandability of the extracted explanations by humans in a user study with four different tasks. We evaluate the results in terms of response times and correctness, subjective ease of understanding and confidence, and similarity of free text responses. The results show that decision trees generated with Trepan Reloaded, taking into account domain knowledge, are significantly more understandable throughout than those generated by standard Trepan. The enhanced understandability of post-hoc explanations is achieved with little compromise on the accuracy with which the surrogate decision trees replicate the behaviour of the original neural network models.

Explainability in Artificial Intelligence has been revived as a topic of active research by the need of conveying safety and trust to users in the `how' and `why' of automated decision-making. Whilst a plethora of approaches have been developed for post-hoc explainability, only a few focus on how to use domain knowledge, and how this influences the understandability of global explanations from the users' perspective. In this paper, we show how ontologies help the understandability of global post-hoc explanations, presented in the form of symbolic models. In particular, we build on Trepan, an algorithm that explains artificial neural networks by means of decision trees, and we extend it to include ontologies modeling domain knowledge in the process of generating explanations. We present the results of a user study that measures the understandability of decision trees using a syntactic complexity measure, and through time and accuracy of responses as well as reported user confidence and understandability. The user study considers domains where explanations are critical, namely, in finance and medicine. The results show that decision trees generated with our algorithm, taking into account domain knowledge, are more understandable than those generated by standard Trepan without the use of ontologies.

Rationality plays a key role in both the study of human reasoning and Artificial Intelligence (AI). Certain notions of rationality have been adopted in AI as guides for the development of intelligent machines and these notions have been given a normative function. The notions of rationality in AI are often taken to be closely related to conceptions of rationality in human contexts. In this paper, we argue that the normative role of rationality differs in the human and artificial contexts. While rationality in human-focused fields of study is normative, prescribing how humans ought to reason, the normative conception in AI is built on a notion of human rationality which is descriptive, not normative, in the human context, as AI aims at building agents which reason as humans do. In order to make this point, we review prominent notions of rationality used in psychology, cognitive science, and (the history of) philosophy, as well as in AI, and discuss some factors that contributed to rationality being assigned the differing normative statuses in the differing fields of study. We argue that while ‘rationality’ is a normative notion in both AI and in human reasoning, the normativity of the AI conception of ‘rationality’ is grounded in a descriptive account of human rationality.

During the 1980s Michie defined Machine Learning in terms of two orthogonal axes of performance: predictive accuracy and comprehensibility of generated hypotheses. Since predictive accuracy was readily measurable and comprehensibility not so, later definitions in the 1990s, such as Mitchell’s, tended to use a one-dimensional approach to Machine Learning based solely on predictive accuracy, ultimately favouring statistical over symbolic Machine Learning approaches. In this paper we provide a definition of comprehensibility of hypotheses which can be estimated using human participant trials. We present two sets of experiments testing human comprehensibility of logic programs. In the first experiment we test human comprehensibility with and without predicate invention. Results indicate comprehensibility is affected not only by the complexity of the presented program but also by the existence of anonymous predicate symbols. In the second experiment we directly test whether any state-of-the-art ILP systems are ultra-strong learners in Michie’s sense, and select the Metagol system for use in humans trials. Results show participants were not able to learn the relational concept on their own from a set of examples but they were able to apply the relational definition provided by the ILP system correctly. This implies the existence of a class of relational concepts which are hard to acquire for humans, though easy to understand given an abstract explanation. We believe improved understanding of this class could have potential relevance to contexts involving human learning, teaching and verbal interaction.

Concept blending – a cognitive process which allows for the combination of certain elements (and their relations) from originally distinct conceptual spaces into a new unified space combining these previously separate elements, and enables reasoning and inference over the combination – is taken as a key element of creative thought and combinatorial creativity. In this article, we summarise our work towards the development of a computational-level and algorithmic-level account of concept blending, combining approaches from computational analogy-making and case-based reasoning (CBR). We present the theoretical background, as well as an algorithmic proposal integrating higher-order anti-unification matching and generalisation from analogy with amalgams from CBR. The feasibility of the approach is then exemplified in two case studies.