Search

We discuss explanations for formal (abstract and structured) argumentation – the question whether and why a certain argument or claim can be accepted (or not) under various extension-based semantics. We introduce a flexible framework, which can act as the basis for many different types of explanations. For example, we can have simple or comprehensive explanations in terms of arguments for or against a claim, arguments that (indirectly) defend a claim, the evidence (knowledge base) that supports or is incompatible with a claim, and so on. We show how selection based on necessity and sufficiency can be captured in our basic framework and discuss a real-life application.

Computational argumentation is a promising research area, yet there is a gap between theoretical contributions and practical applications. Bridging this gap could potentially raise interest in this topic even more. We argue that one part of the bridge could be an open-source package of implementations of argumentation algorithms, visualised in a web interface. Therefore we present a new release of PyArg, providing various new argumentation-based functionalities – including multiple generators, a learning environment, implementations of theoretical papers and a showcase of a practical application – in a new interface with improved accessibility.

Logical argumentation is a well-known approach to modeling non-monotonic reasoning with conflicting information. In this paper we provide a comprehensive postulate-based study of properties of logical argumentation frameworks and a full characterization of their semantics and inference relations. In this way we identify well-behaved formal argumentative models of drawing logically justified inferences from a given set of possibly conflicting defeasible, as well as strict assumptions. Given some desiderata in terms of rationality postulates, we consider the conditions that an argumentation framework should fulfill for the desiderata to hold. One purpose of this approach is to assist designers to “plug-in” pre-defined formalisms according to actual needs. To this end, we present a classification of argumentation frameworks relative to the types of attacks they implement. In turn, for each class we determine which desiderata are satisfied. Our study is highly abstract, supposing only a minimal set of requirements on the considered underlying deductive systems, and in this way covering a broad range of formalisms, including classical, intuitionistic and modal logics.

Reasoning under incomplete information is an important research direction in AI argumentation. Most computational advances in this direction have so-far focused on abstract argumentation frameworks. Development of computational approaches to reasoning under incomplete information in structured formalisms remains to-date to a large extent a challenge. We address this challenge by studying the so-called stability and relevance problems---with the aim of analyzing aspects of resilience of acceptance statuses in light of new information---in the central structured formalism of ASPIC+. Focusing on the case of the grounded semantics and an ASPIC+ fragment motivated through application scenarios, we develop exact ASP-based algorithms for stability and relevance in incomplete ASPIC+ theories, and pinpoint the complexity of reasoning about stability (coNP-complete) and relevance (Sigma_2^P-complete), further justifying our ASP-based approaches. Empirically, the algorithms exhibit promising scalability, outperforming even a recent inexact approach to stability, with our ASP-based iterative approach being the first algorithm proposed for reasoning about relevance in ASPIC+.

This thesis develops a human-centred explanation method for rule-based automated decision-making systems in the legal domain. The research consists of theoretical exploration and practical implementation. Theoretical research establishes a framework for developing explanation methods, representing its key internal components (content, communication and adaptation) and external factors (decision-making system, human recipient and domain). Further investigation of human-centred research highlights the importance of considering both the recipient’s knowledge and goals. Besides, we found that one way to accomplish this is by creating a question-driven explanation method and visualising the decision-making process to aid understanding. Accordingly, the proposed explanation method involves representing a decision model in a graph database to be able to both question and visualise it. This proposed explanation method is implemented for a real-world scenario, generating tailored explanations for different target groups. The evaluation highlights the method’s ability to answer specific questions but identifies limitations in handling logical checks and hypothetical scenarios. Future research can focus on improving these aspects and exploring additional reasoning properties and customisable interfaces to adapt the method to recipients’ evolving needs.

With the growing usage of artificial intelligence (AI) in daily life, explainable systems become more important. Explainable AI (XAI), which is a set of tools and frameworks to help you understand and interpret predictions made by AI, has risen in popularity due to this. Formal argumentation is a suitable tool that can be used to model contrastive explanations for (X)AI systems. In this thesis, the concept of contrastiveness will be modelled in various formal argumentation settings. The definition for contrastiveness is based on definitions found in literature from the field of the social sci- ences and humanities. First, an extensive literature research is conducted to find suitable definitions for contrastiveness, then those concepts are modelled in Dung’s abstract argumentation frameworks, preference-based argumentation frameworks and finally in the structured argumentation setting of ASPIC+. These definitions are then evaluated at the hand of examples based on real-life situations. This work successfully models various notions of contrastiveness in various formal argumentation set- tings and paves the way for the further study and application of contrastiveness in (argumentative) XAI.

In this thesis it is examined how argumentative dialogues can be adapted for use in Explainable Artificial Intelligence (XAI) and this thesis shows the possibilities on how XAI can be improved in terms of explanation transfer. This is achieved by creating an explanatory dialogue system which integrates findings from the social sciences. In this thesis, requirements of XAI for explanations are examined first. Subsequently, existing formal dialogue systems are discussed and finally, a new formal dialogue system is constructed. This new dialogue system implements various aspects: it allows for different user-types (lay users and experts), it allows for personalized explanations in terms of various question and explanation styles, plus, various explanation functions are usable. Lastly, a user study is performed in which the preferred amount of explanation of participants is measured. The results show that lay users tend to request more explanation than experts

We propose a human-centred explanation method for rule-based automated decision-making systems in the legal domain. Firstly, we establish a conceptual framework for developing explanation methods, representing its key internal components (content, communication and adaptation) and external dependencies (decision-making system, human recipient and domain). Secondly, we propose an explanation method that uses a graph database to enable question-driven explanations and multimedia display. This way, we can tailor the explanation to the user. Finally, we show how our conceptual framework is applicable to a real-world scenario at the Dutch Tax and Customs Administration and implement our explanation method for this scenario., Comment: This is the full version of a demo at the 36th International Conference on Legal Knowledge and Information Systems (JURIX'23)

Modeling contrastive explanations for the use in artificial intelligence (AI) applications is an important research branch within the field of explainable AI (XAI). However, most of the existing contrastive XAI approaches are not based on the findings in the literature from the social sciences on contrastiveness in human reasoning and human explanations. In this work we collect the various types of contrastiveness proposed in the literature and model these with formal argumentation. The result is a variety of argumentation-based methods for contrastive explanations, based on the available literature and applicable in a wide variety of AI-applications.

In this paper we study everyday explanations for classification tasks with formal argumentation. Everyday explanations describe how humans explain in day-to-day life, which is important when explaining decisions of AI systems to lay users. We introduce EVAX, a model-agnostic explanation method for classifiers with which contrastive, selected and social explanations can be generated. The resulting explanations can be adjusted in their size and retain high fidelity scores (an average of 0.95)

Explainable artificial intelligence (XAI) has gained increasing interest in recent years in the argumentation community. In this paper we consider this topic in the context of logic-based argumentation, showing that the latter is a particularly promising paradigm for facilitating explainable AI. In particular, we provide two representations of abductive reasoning by sequent-based argumentation frameworks and show that such frameworks successfully cope with related challenges, such as the handling of synonyms, justifications, and logical equivalences.

We introduce PyArg, a Python-based solver and explainer for both abstract argumentation and ASPIC+. A large variety of extension-based semantics allows for flexible evaluation and several explanation functions are available.

We explore the computational complexity of stability and relevance in incomplete argumentation frameworks (IAFs), abstract argumentation frameworks that encode qualitative uncertainty by distinguishing between certain and uncertain arguments and attacks. IAFs can be specified by, e.g., making uncertain arguments or attacks certain; the justification status of arguments in an IAF is determined on the basis of the certain arguments and attacks. An argument is stable if its justification status is the same in all specifications of the IAF. For arguments that are not stable in an IAF, the relevance problem is of interest: which uncertain arguments or attacks should be investigated for the argument to become stable? We redefine stability and define relevance for IAFs and study their complexity.

In argument-based inquiry, agents jointly construct arguments supporting or attacking a topic claim to find out if the claim can be accepted given the agents’ knowledge bases. While such inquiry systems can be used for various forms of automated information intake, several efficiency issues have so far prevented widespread application. In this paper, we aim to tackle these efficiency issues by exploring the notion of stability: can additional information change the justification status of the claim under discussion? Detecting stability is not tractable for every input, since the problem is CoNP-complete, yet in practical applications it is essential to guarantee efficient computation. This makes approximation a viable alternative. We present a sound approximation algorithm that recognises stability for many inputs in polynomial time and discuss several of its properties. In particular, we show that the algorithm is sound and identify constraints on the input under which it is complete. As a final contribution of this paper, we describe how the proposed algorithm is used in three different case studies at the Netherlands Police.

In this paper we discuss contrastive explanations for formal argumentation – the question why one argument (the fact) can be accepted, whilst another argument (the foil) cannot be accepted. We show under which conditions contrastive explanations in abstract argumentation are meaningful, and how argumentation allows us to make implicit foils explicit.

There has been an upswing in the research field of explainable artificial intelligence (XAI) of methods aimed at explaining opaque artificial intelligence (AI) systems and their decisions. A recent, promising approach involves the use of formal argumentation to explain machine learning (ML) applications. In this thesis we investigate that approach; we aim to gain understanding of the value of argumentation for XAI. In particular, we explore how well argumentation can produce everyday explanations. Everyday explanations describe how humans explain in day to-day life and are claimed to be important for explaining decisions of AI systems to end-users. First, we conceptually show how argumentative explanations can be posed as everyday explanations. Afterward, we demonstrate that current argumentative explanation methods compute explanations that already contain some, but not all properties of everyday explanations. Finally, we present everyday argumentative explanations, or EVAX, which is a model-agnostic method that computes local explanations for ML models. These explanations can be adjusted in their size and retain high fidelity scores (an average of 0.95) on four different datasets and four different ML models. In addition, the explanations incorporate the main characteristics of everyday explanations and help in achieving the objectives of XAI.

There has been an upswing in the research field of explainable artificial intelligence (XAI) of methods aimed at explaining opaque artificial intelligence (AI) systems and their decisions. A recent, promising approach involves the use of formal argumentation to explain machine learning (ML) applications. In this thesis we investigate that approach; we aim to gain understanding of the value of argumentation for XAI. In particular, we explore how well argumentation can produce everyday explanations. Everyday explanations describe how humans explain in day to-day life and are claimed to be important for explaining decisions of AI systems to end-users. First, we conceptually show how argumentative explanations can be posed as everyday explanations. Afterward, we demonstrate that current argumentative explanation methods compute explanations that already contain some, but not all properties of everyday explanations. Finally, we present everyday argumentative explanations, or EVAX, which is a model-agnostic method that computes local explanations for ML models. These explanations can be adjusted in their size and retain high fidelity scores (an average of 0.95) on four different datasets and four different ML models. In addition, the explanations incorporate the main characteristics of everyday explanations and help in achieving the objectives of XAI.

Enforcement, adjusting an argumentation framework such that a certain set of arguments becomes acceptable, is an important research topic within the study of dynamic argumentation, but one that has been little studied for structured argumentation. In this paper we study enforcement in a general structured argumentation setting. In particular, we study conditions on the argumentation setting and the knowledge base that ensure (or prevent) the acceptability of sets of formulas for structured argumentation frameworks.

In this paper we provide a detailed analysis of the inference process induced by logical argumentation frameworks. The frameworks may be defined with respect to any propositional language and logic, different arguments that represent deductions in the logic, various support-based attack relations between arguments, and all the complete Dung-style semantics for the frameworks. We show that, ultimately, for characterizing the inference process with respect to a given framework, extension-based semantics may be divided into two types: single-extension and multiple-extension, which induce respective kinds of entailment relations. These entailments are further classified by the way they tolerate new information (nonmonotonicity-related properties) and maintain conflicts among arguments (inconsistency-related properties).

In this paper, we discuss necessary and sufficient explanations – the question whether and why a certain argument or claim can be accepted (or not) – for abstract and structured argumentation. Given a framework with which explanations for argumentation-based conclusions can be derived, we study necessity and sufficiency: what (sets of) arguments are necessary or sufficient for the (non-)acceptance of an argument or claim? We will show that necessary and sufficient explanations can be strictly smaller than minimal explanations, while still providing all the reasons for a conclusion and we discuss their usefulness in a real-life application.

As AI systems are increasingly applied in real-life situations, it is essential that such systems can give explanations that provide insight into the underlying decision models and techniques. Thus, users can understand, trust and validate the system, and experts can verify that the system works as intended. At the Dutch National Police several applications based on computational argumentation are in use, with police analysts and Dutch citizens as possible users. In this paper we show how a basic framework of explanations aimed at explaining argumentation-based conclusions can be applied to these applications at the police.

We discuss explanations for formal (abstract and structured) argumentation-the question of whether and why a certain argument or claim can be accepted (or not) under various extension-based semantics. We introduce a flexible framework, which can act as the basis for many different types of explanations. For example, we can have simple or comprehensive explanations in terms of arguments for or against a claim, arguments that (indirectly) defend a claim, the evidence (knowledge base) that supports or is incompatible with a claim, and so on. We show how different types of explanations can be captured in our basic framework, discuss a real-life application and formally compare our framework to existing work.