Francesca Toni

Avinash Kori, Ben Glocker, Francesca Toni

Most of the current explainability techniques focus on capturing the importance of features in input space. However, given the complexity of models and data-generating processes, the resulting explanations are far from being `complete', in that they lack an indication of feature interactions and visualization of their `effect'. In this work, we propose a novel twin-surrogate explainability framework to explain the decisions made by any CNN-based image classifier (irrespective of the architecture). For this, we first disentangle latent features from the classifier, followed by aligning these features to observed/human-defined `context' features. These aligned features form semantically meaningful concepts that are used for extracting a causal graph depicting the `perceived' data-generating process, describing the inter- and intra-feature interactions between unobserved latent features and observed `context' features. This causal graph serves as a global model from which local explanations of different forms can be extracted. Specifically, we provide a generator to visualize the `effect' of interactions among features in latent space and draw feature importance therefrom as local explanations. Our framework utilizes adversarial knowledge distillation to faithfully learn a representation from the classifiers' latent space and use it for extracting visual explanations. We use the styleGAN-v2 architecture with an additional regularization term to enforce disentanglement and alignment. We demonstrate and evaluate explanations obtained with our framework on Morpho-MNIST and on the FFHQ human faces dataset. Our framework is available at \url{https://github.com/koriavinash1/GLANCE-Explanations}.

Ainkaran Santhirasekaram, Avinash Kori, Andrea Rockall et al.

Explanations for \emph{black-box} models help us understand model decisions as well as provide information on model biases and inconsistencies. Most of the current explainability techniques provide a single level of explanation, often in terms of feature importance scores or feature attention maps in input space. Our focus is on explaining deep discriminative models at \emph{multiple levels of abstraction}, from fine-grained to fully abstract explanations. We achieve this by using the natural properties of \emph{hyperbolic geometry} to more efficiently model a hierarchy of symbolic features and generate \emph{hierarchical symbolic rules} as part of our explanations. Specifically, for any given deep discriminative model, we distill the underpinning knowledge by discretisation of the continuous latent space using vector quantisation to form symbols, followed by a \emph{hyperbolic reasoning block} to induce an \emph{abstraction tree}. We traverse the tree to extract explanations in terms of symbolic rules and its corresponding visual semantics. We demonstrate the effectiveness of our method on the MNIST and AFHQ high-resolution animal faces dataset. Our framework is available at \url{https://github.com/koriavinash1/SymbolicInterpretability}.

Junqi Jiang, Francesco Leofante, Antonio Rago et al.

The use of counterfactual explanations (CFXs) is an increasingly popular explanation strategy for machine learning models. However, recent studies have shown that these explanations may not be robust to changes in the underlying model (e.g., following retraining), which raises questions about their reliability in real-world applications. Existing attempts towards solving this problem are heuristic, and the robustness to model changes of the resulting CFXs is evaluated with only a small number of retrained models, failing to provide exhaustive guarantees. To remedy this, we propose Δ-robustness, the first notion to formally and deterministically assess the robustness (to model changes) of CFXs for neural networks. We introduce an abstraction framework based on interval neural networks to verify the Δ-robustness of CFXs against a possibly infinite set of changes to the model parameters, i.e., weights and biases. We then demonstrate the utility of this approach in two distinct ways. First, we analyse the Δ-robustness of a number of CFX generation methods from the literature and show that they unanimously host significant deficiencies in this regard. Second, we demonstrate how embedding Δ-robustness within existing methods can provide CFXs which are provably robust.

Avinash Kori, Francesco Locatello, Fabio De Sousa Ribeiro et al.

The extraction of modular object-centric representations for downstream tasks is an emerging area of research. Learning grounded representations of objects that are guaranteed to be stable and invariant promises robust performance across different tasks and environments. Slot Attention (SA) learns object-centric representations by assigning objects to \textit{slots}, but presupposes a \textit{single} distribution from which all slots are randomly initialised. This results in an inability to learn \textit{specialized} slots which bind to specific object types and remain invariant to identity-preserving changes in object appearance. To address this, we present \emph{\textsc{Co}nditional \textsc{S}lot \textsc{A}ttention} (\textsc{CoSA}) using a novel concept of \emph{Grounded Slot Dictionary} (GSD) inspired by vector quantization. Our proposed GSD comprises (i) canonical object-level property vectors and (ii) parametric Gaussian distributions, which define a prior over the slots. We demonstrate the benefits of our method in multiple downstream tasks such as scene generation, composition, and task adaptation, whilst remaining competitive with SA in popular object discovery benchmarks.

Antonio Rago, Hengzhi Li, Francesca Toni

As the field of explainable AI (XAI) is maturing, calls for interactive explanations for (the outputs of) AI models are growing, but the state-of-the-art predominantly focuses on static explanations. In this paper, we focus instead on interactive explanations framed as conflict resolution between agents (i.e. AI models and/or humans) by leveraging on computational argumentation. Specifically, we define Argumentative eXchanges (AXs) for dynamically sharing, in multi-agent systems, information harboured in individual agents' quantitative bipolar argumentation frameworks towards resolving conflicts amongst the agents. We then deploy AXs in the XAI setting in which a machine and a human interact about the machine's predictions. We identify and assess several theoretical properties characterising AXs that are suitable for XAI. Finally, we instantiate AXs for XAI by defining various agent behaviours, e.g. capturing counterfactual patterns of reasoning in machines and highlighting the effects of cognitive biases in humans. We show experimentally (in a simulated environment) the comparative advantages of these behaviours in terms of conflict resolution, and show that the strongest argument may not always be the most effective.

Hamed Ayoobi, Nico Potyka, Francesca Toni

Neural networks (NNs) have various applications in AI, but explaining their decisions remains challenging. Existing approaches often focus on explaining how changing individual inputs affects NNs' outputs. However, an explanation that is consistent with the input-output behaviour of an NN is not necessarily faithful to the actual mechanics thereof. In this paper, we exploit relationships between multi-layer perceptrons (MLPs) and quantitative argumentation frameworks (QAFs) to create argumentative explanations for the mechanics of MLPs. Our SpArX method first sparsifies the MLP while maintaining as much of the original structure as possible. It then translates the sparse MLP into an equivalent QAF to shed light on the underlying decision process of the MLP, producing global and/or local explanations. We demonstrate experimentally that SpArX can give more faithful explanations than existing approaches, while simultaneously providing deeper insights into the actual reasoning process of MLPs.

Junqi Jiang, Jianglin Lan, Francesco Leofante et al.

Counterfactual Explanations (CEs) have received increasing interest as a major methodology for explaining neural network classifiers. Usually, CEs for an input-output pair are defined as data points with minimum distance to the input that are classified with a different label than the output. To tackle the established problem that CEs are easily invalidated when model parameters are updated (e.g. retrained), studies have proposed ways to certify the robustness of CEs under model parameter changes bounded by a norm ball. However, existing methods targeting this form of robustness are not sound or complete, and they may generate implausible CEs, i.e., outliers wrt the training dataset. In fact, no existing method simultaneously optimises for closeness and plausibility while preserving robustness guarantees. In this work, we propose Provably RObust and PLAusible Counterfactual Explanations (PROPLACE), a method leveraging on robust optimisation techniques to address the aforementioned limitations in the literature. We formulate an iterative algorithm to compute provably robust CEs and prove its convergence, soundness and completeness. Through a comparative experiment involving six baselines, five of which target robustness, we show that PROPLACE achieves state-of-the-art performances against metrics on three evaluation aspects.

Nico Potyka, Xiang Yin, Francesca Toni

Random forests are decision tree ensembles that can be used to solve a variety of machine learning problems. However, as the number of trees and their individual size can be large, their decision making process is often incomprehensible. In order to reason about the decision process, we propose representing it as an argumentation problem. We generalize sufficient and necessary argumentative explanations using a Markov network encoding, discuss the relevance of these explanations and establish relationships to families of abductive explanations from the literature. As the complexity of the explanation problems is high, we discuss a probabilistic approximation algorithm and present first experimental results.

Ha Thanh Nguyen, Randy Goebel, Francesca Toni et al.

Negation is a fundamental aspect of natural language, playing a critical role in communication and comprehension. Our study assesses the negation detection performance of Generative Pre-trained Transformer (GPT) models, specifically GPT-2, GPT-3, GPT-3.5, and GPT-4. We focus on the identification of negation in natural language using a zero-shot prediction approach applied to our custom xNot360 dataset. Our approach examines sentence pairs labeled to indicate whether the second sentence negates the first. Our findings expose a considerable performance disparity among the GPT models, with GPT-4 surpassing its counterparts and GPT-3.5 displaying a marked performance reduction. The overall proficiency of the GPT models in negation detection remains relatively modest, indicating that this task pushes the boundaries of their natural language understanding capabilities. We not only highlight the constraints of GPT models in handling negation but also emphasize the importance of logical reliability in high-stakes domains such as healthcare, science, and law.

Xiang Yin, Nico Potyka, Francesca Toni

Argumentative explainable AI has been advocated by several in recent years, with an increasing interest on explaining the reasoning outcomes of Argumentation Frameworks (AFs). While there is a considerable body of research on qualitatively explaining the reasoning outcomes of AFs with debates/disputes/dialogues in the spirit of extension-based semantics, explaining the quantitative reasoning outcomes of AFs under gradual semantics has not received much attention, despite widespread use in applications. In this paper, we contribute to filling this gap by proposing a novel theory of Argument Attribution Explanations (AAEs) by incorporating the spirit of feature attribution from machine learning in the context of Quantitative Bipolar Argumentation Frameworks (QBAFs): whereas feature attribution is used to determine the influence of features towards outputs of machine learning models, AAEs are used to determine the influence of arguments towards topic arguments of interest. We study desirable properties of AAEs, including some new ones and some partially adapted from the literature to our setting. To demonstrate the applicability of our AAEs in practice, we conclude by carrying out two case studies in the scenarios of fake news detection and movie recommender systems.

Dekai Zhang, Francesca Toni, Matthew Williams

Recent research has shown the potential for neural networks to improve upon classical survival models such as the Cox model, which is widely used in clinical practice. Neural networks, however, typically rely on data that are centrally available, whereas healthcare data are frequently held in secure silos. We present a federated Cox model that accommodates this data setting and also relaxes the proportional hazards assumption, allowing time-varying covariate effects. In this latter respect, our model does not require explicit specification of the time-varying effects, reducing upfront organisational costs compared to previous works. We experiment with publicly available clinical datasets and demonstrate that the federated model is able to perform as well as a standard model.

Benjamin Irwin, Antonio Rago, Francesca Toni

We introduce Forecasting Argumentation Frameworks (FAFs), a novel argumentation-based methodology for forecasting informed by recent judgmental forecasting research. FAFs comprise update frameworks which empower (human or artificial) agents to argue over time about the probability of outcomes, e.g. the winner of a political election or a fluctuation in inflation rates, whilst flagging perceived irrationality in the agents' behaviour with a view to improving their forecasting accuracy. FAFs include five argument types, amounting to standard pro/con arguments, as in bipolar argumentation, as well as novel proposal arguments and increase/decrease amendment arguments. We adapt an existing gradual semantics for bipolar argumentation to determine the aggregated dialectical strength of proposal arguments and define irrational behaviour. We then give a simple aggregation function which produces a final group forecast from rational agents' individual forecasts. We identify and study properties of FAFs and conduct an empirical evaluation which signals FAFs' potential to increase the forecasting accuracy of participants.

Fabrizio Russo, Francesca Toni

Neural networks have proven to be effective at solving machine learning tasks but it is unclear whether they learn any relevant causal relationships, while their black-box nature makes it difficult for modellers to understand and debug them. We propose a novel method overcoming these issues by allowing a two-way interaction whereby neural-network-empowered machines can expose the underpinning learnt causal graphs and humans can contest the machines by modifying the causal graphs before re-injecting them into the machines. The learnt models are guaranteed to conform to the graphs and adhere to expert knowledge, some of which can also be given up-front. By building a window into the model behaviour and enabling knowledge injection, our method allows practitioners to debug networks based on the causal structure discovered from the data and underpinning the predictions. Experiments with real and synthetic tabular data show that our method improves predictive performance up to 2.4x while producing parsimonious networks, up to 7x smaller in the input layer, compared to SOTA regularised networks.

Alexander Gaskell, Yishu Miao, Lucia Specia et al.

Recent efforts within the AI community have yielded impressive results towards "soft theorem proving" over natural language sentences using language models. We propose a novel, generative adversarial framework for probing and improving these models' reasoning capabilities. Adversarial attacks in this domain suffer from the logical inconsistency problem, whereby perturbations to the input may alter the label. Our Logically consistent AdVersarial Attacker, LAVA, addresses this by combining a structured generative process with a symbolic solver, guaranteeing logical consistency. Our framework successfully generates adversarial attacks and identifies global weaknesses common across multiple target models. Our analyses reveal naive heuristics and vulnerabilities in these models' reasoning capabilities, exposing an incomplete grasp of logical deduction under logic programs. Finally, in addition to effective probing of these models, we show that training on the generated samples improves the target model's performance.

Ha-Thanh Nguyen, Randy Goebel, Francesca Toni et al.

The evolution of Generative Pre-trained Transformer (GPT) models has led to significant advancements in various natural language processing applications, particularly in legal textual entailment. We present an analysis of GPT-3.5 (ChatGPT) and GPT-4 performances on COLIEE Task 4 dataset, a prominent benchmark in this domain. The study encompasses data from Heisei 18 (2006) to Reiwa 3 (2021), exploring the models' abilities to discern entailment relationships within Japanese statute law across different periods. Our preliminary experimental results unveil intriguing insights into the models' strengths and weaknesses in handling legal textual entailment tasks, as well as the patterns observed in model performance. In the context of proprietary models with undisclosed architectures and weights, black-box analysis becomes crucial for evaluating their capabilities. We discuss the influence of training data distribution and the implications on the models' generalizability. This analysis serves as a foundation for future research, aiming to optimize GPT-based models and enable their successful adoption in legal information extraction and entailment applications.

Emanuele De Angelis, Maurizio Proietti, Francesca Toni

Recently, ABA Learning has been proposed as a form of symbolic machine learning for drawing Assumption-Based Argumentation frameworks from background knowledge and positive and negative examples. We propose a novel method for implementing ABA Learning using Answer Set Programming as a way to help guide Rote Learning and generalisation in ABA Learning.

Avinash Kori, Ben Glocker, Francesca Toni

An effective way to obtain different perspectives on any given topic is by conducting a debate, where participants argue for and against the topic. Here, we propose a novel debate framework for understanding and explaining a continuous image classifier's reasoning for making a particular prediction by modeling it as a multiplayer sequential zero-sum debate game. The contrastive nature of our framework encourages players to learn to put forward diverse arguments during the debates, picking up the reasoning trails missed by their opponents and highlighting any uncertainties in the classifier. Specifically, in our proposed setup, players propose arguments, drawn from the classifier's discretized latent knowledge, to support or oppose the classifier's decision. The resulting Visual Debates collect supporting and opposing features from the discretized latent space of the classifier, serving as explanations for the internal reasoning of the classifier towards its predictions. We demonstrate and evaluate (a practical realization of) our Visual Debates on the geometric SHAPE and MNIST datasets and on the high-resolution animal faces (AFHQ) dataset, along standard evaluation metrics for explanations (i.e. faithfulness and completeness) and novel, bespoke metrics for visual debates as explanations (consensus and split ratio).

Stylianos Loukas Vasileiou, Ashwin Kumar, William Yeoh et al.

We present a novel framework designed to extend model reconciliation approaches, commonly used in human-aware planning, for enhanced human-AI interaction. By adopting a structured argumentation-based dialogue paradigm, our framework enables dialectical reconciliation to address knowledge discrepancies between an explainer (AI agent) and an explainee (human user), where the goal is for the explainee to understand the explainer's decision. We formally describe the operational semantics of our proposed framework, providing theoretical guarantees. We then evaluate the framework's efficacy ``in the wild'' via computational and human-subject experiments. Our findings suggest that our framework offers a promising direction for fostering effective human-AI interactions in domains where explainability is important.

Antonio Rago, Pietro Baroni, Francesca Toni

Causal models are playing an increasingly important role in machine learning, particularly in the realm of explainable AI. We introduce a conceptualisation for generating argumentation frameworks (AFs) from causal models for the purpose of forging explanations for the models' outputs. The conceptualisation is based on reinterpreting desirable properties of semantics of AFs as explanation moulds, which are means for characterising the relations in the causal model argumentatively. We demonstrate our methodology by reinterpreting the property of bi-variate reinforcement as an explanation mould to forge bipolar AFs as explanations for the outputs of causal models. We perform a theoretical evaluation of these argumentative explanations, examining whether they satisfy a range of desirable explanatory and argumentative properties.

Piyawat Lertvittayakumjorn, Francesca Toni

Recent works in Explainable AI mostly address the transparency issue of black-box models or create explanations for any kind of models (i.e., they are model-agnostic), while leaving explanations of interpretable models largely underexplored. In this paper, we fill this gap by focusing on explanations for a specific interpretable model, namely pattern-based logistic regression (PLR) for binary text classification. We do so because, albeit interpretable, PLR is challenging when it comes to explanations. In particular, we found that a standard way to extract explanations from this model does not consider relations among the features, making the explanations hardly plausible to humans. Hence, we propose AXPLR, a novel explanation method using (forms of) computational argumentation to generate explanations (for outputs computed by PLR) which unearth model agreements and disagreements among the features. Specifically, we use computational argumentation as follows: we see features (patterns) in PLR as arguments in a form of quantified bipolar argumentation frameworks (QBAFs) and extract attacks and supports between arguments based on specificity of the arguments; we understand logistic regression as a gradual semantics for these QBAFs, used to determine the arguments' dialectic strength; and we study standard properties of gradual semantics for QBAFs in the context of our argumentative re-interpretation of PLR, sanctioning its suitability for explanatory purposes. We then show how to extract intuitive explanations (for outputs computed by PLR) from the constructed QBAFs. Finally, we conduct an empirical evaluation and two experiments in the context of human-AI collaboration to demonstrate the advantages of our resulting AXPLR method.

Guilherme Paulino-Passos, Francesca Toni

Recent work shows issues of consistency with explanations, with methods generating local explanations that seem reasonable instance-wise, but that are inconsistent across instances. This suggests not only that instance-wise explanations can be unreliable, but mainly that, when interacting with a system via multiple inputs, a user may actually lose confidence in the system. To better analyse this issue, in this work we treat explanations as objects that can be subject to reasoning and present a formal model of the interactive scenario between user and system, via sequences of inputs, outputs, and explanations. We argue that explanations can be thought of as committing to some model behaviour (even if only prima facie), suggesting a form of entailment, which, we argue, should be thought of as non-monotonic. This allows: 1) to solve some considered inconsistencies in explanation, such as via a specificity relation; 2) to consider properties from the non-monotonic reasoning literature and discuss their desirability, gaining more insight on the interactive explanation scenario.

Nico Potyka, Xiang Yin, Francesca Toni

There is broad agreement in the literature that explanation methods should be faithful to the model that they explain, but faithfulness remains a rather vague term. We revisit faithfulness in the context of continuous data and propose two formal definitions of faithfulness for feature attribution methods. Qualitative faithfulness demands that scores reflect the true qualitative effect (positive vs. negative) of the feature on the model and quanitative faithfulness that the magnitude of scores reflect the true quantitative effect. We discuss under which conditions these requirements can be satisfied to which extent (local vs global). As an application of the conceptual idea, we look at differentiable classifiers over continuous data and characterize Gradient-scores as follows: every qualitatively faithful feature attribution method is qualitatively equivalent to Gradient-scores. Furthermore, if an attribution method is quantitatively faithful in the sense that changes of the output of the classifier are proportional to the scores of features, then it is either equivalent to gradient-scoring or it is based on an inferior approximation of the classifier. To illustrate the practical relevance of the theory, we experimentally demonstrate that popular attribution methods can fail to give faithful explanations in the setting where the data is continuous and the classifier differentiable.

Francesca Toni, Nico Potyka, Markus Ulbricht et al.

ProbLog is a popular probabilistic logic programming language/tool, widely used for applications requiring to deal with inherent uncertainties in structured domains. In this paper we study connections between ProbLog and a variant of another well-known formalism combining symbolic reasoning and reasoning under uncertainty, i.e. probabilistic argumentation. Specifically, we show that ProbLog is an instance of a form of Probabilistic Abstract Argumentation (PAA) that builds upon Assumption-Based Argumentation (ABA). The connections pave the way towards equipping ProbLog with alternative semantics, inherited from PAA/PABA, as well as obtaining novel argumentation semantics for PAA/PABA, leveraging on prior connections between ProbLog and argumentation. Further, the connections pave the way towards novel forms of argumentative explanations for ProbLog's outputs.

Francis Rhys Ward, Francesco Belardinelli, Francesca Toni

We define a novel neuro-symbolic framework, argumentative reward learning, which combines preference-based argumentation with existing approaches to reinforcement learning from human feedback. Our method improves prior work by generalising human preferences, reducing the burden on the user and increasing the robustness of the reward model. We demonstrate this with a number of experiments.

Enrico Pontelli, Stefania Costantini, Carmine Dodaro et al.

This volume contains the Technical Communications presented at the 39th International Conference on Logic Programming (ICLP 2023), held at Imperial College London, UK from July 9 to July 15, 2023. Technical Communications included here concern the Main Track, the Doctoral Consortium, the Application and Systems/Demo track, the Recently Published Research Track, the Birds-of-a-Feather track, the Thematic Tracks on Logic Programming and Machine Learning, and Logic Programming and Explainability, Ethics, and Trustworthiness.

Emanuele De Angelis, Fabio Fioravanti, Maria Chiara Meo et al.

Assumption-based Argumentation (ABA) is a well-established form of structured argumentation. ABA frameworks with an underlying atomic language are widely studied, but their applicability is limited by a representational restriction to ground (variable-free) arguments and attacks built from propositional atoms. In this paper, we lift this restriction and propose a novel notion of constrained ABA (CABA), whose components, as well as arguments built from them, may include constrained variables, ranging over possibly infinite domains. We define non-ground semantics for CABA, in terms of various notions of non-ground attacks. We show that the new semantics conservatively generalise standard ABA semantics.

Hamed Ayoobi, Nico Potyka, Francesca Toni

We propose ProtoArgNet, a novel interpretable deep neural architecture for image classification in the spirit of prototypical-part-learning as found, e.g., in ProtoPNet. While earlier approaches associate every class with multiple prototypical-parts, ProtoArgNet uses super-prototypes that combine prototypical-parts into a unified class representation. This is done by combining local activations of prototypes in an MLP-like manner, enabling the localization of prototypes and learning (non-linear) spatial relationships among them. By leveraging a form of argumentation, ProtoArgNet is capable of providing both supporting (i.e. `this looks like that') and attacking (i.e. `this differs from that') explanations. We demonstrate on several datasets that ProtoArgNet outperforms state-of-the-art prototypical-part-learning approaches. Moreover, the argumentation component in ProtoArgNet is customisable to the user's cognitive requirements by a process of sparsification, which leads to more compact explanations compared to state-of-the-art approaches.

Antonio Rago, Bence Palfi, Purin Sukpanichnant et al.

AI-driven tools for healthcare are widely acknowledged as potentially beneficial to health practitioners and patients, e.g. the QCancer regression tool for cancer risk prediction. However, for these tools to be trusted, they need to be supplemented with explanations. We examine how explanations' content and format affect user comprehension and trust when explaining QCancer's predictions. Regarding content, we deploy the SHAP and Occlusion-1 explanation methods. Regarding format, we present SHAP explanations, conventionally, as charts (SC) and Occlusion-1 explanations as charts (OC) as well as text (OT), to which their simpler nature lends itself. We conduct experiments with two sets of stakeholders: the general public (representing patients) and medical students (representing healthcare practitioners). Our experiments showed higher subjective comprehension and trust for Occlusion-1 over SHAP explanations based on content. However, when controlling for format, only OT outperformed SC, suggesting this trend is driven by preferences for text. Other findings corroborated that explanation format, rather than content, is often the critical factor.

Xiang Yin, Nico Potyka, Francesca Toni

There is a growing interest in understanding arguments' strength in Quantitative Bipolar Argumentation Frameworks (QBAFs). Most existing studies focus on attribution-based methods that explain an argument's strength by assigning importance scores to other arguments but fail to explain how to change the current strength to a desired one. To solve this issue, we introduce counterfactual explanations for QBAFs. We discuss problem variants and propose an iterative algorithm named Counterfactual Explanations for Quantitative bipolar Argumentation frameworks (CE-QArg). CE-QArg can identify valid and cost-effective counterfactual explanations based on two core modules, polarity and priority, which help determine the updating direction and magnitude for each argument, respectively. We discuss some formal properties of our counterfactual explanations and empirically evaluate CE-QArg on randomly generated QBAFs.

Purin Sukpanichnant, Anna Rapberger, Francesca Toni

Peer review is an essential process to determine the quality of papers submitted to scientific conferences or journals. However, it is subjective and prone to biases. Several studies have been conducted to apply techniques from NLP to support peer review, but they are based on black-box techniques and their outputs are difficult to interpret and trust. In this paper, we propose a novel pipeline to support and understand the reviewing and decision-making processes of peer review: the PeerArg system combining LLMs with methods from knowledge representation. PeerArg takes in input a set of reviews for a paper and outputs the paper acceptance prediction. We evaluate the performance of the PeerArg pipeline on three different datasets, in comparison with a novel end-2-end LLM that uses few-shot learning to predict paper acceptance given reviews. The results indicate that the end-2-end LLM is capable of predicting paper acceptance from reviews, but a variant of the PeerArg pipeline outperforms this LLM.

Stylianos Loukas Vasileiou, Antonio Rago, Francesca Toni et al.

Computational argumentation offers formal frameworks for transparent, verifiable reasoning but has traditionally been limited by its reliance on domain-specific information and extensive feature engineering. In contrast, LLMs excel at processing unstructured text, yet their opaque nature makes their reasoning difficult to evaluate and trust. We argue that the convergence of these fields will lay the foundation for a new paradigm: Argumentative Human-AI Decision-Making. We analyze how the synergy of argumentation framework mining, argumentation framework synthesis, and argumentative reasoning enables agents that do not just justify decisions, but engage in dialectical processes where decisions are contestable and revisable -- reasoning with humans rather than for them. This convergence of computational argumentation and LLMs is essential for human-aware, trustworthy AI in high-stakes domains.

Deniz Gorur, Antonio Rago, Francesca Toni

Argument mining (AM) is the process of automatically extracting arguments, their components and/or relations amongst arguments and components from text. As the number of platforms supporting online debate increases, the need for AM becomes ever more urgent, especially in support of downstream tasks. Relation-based AM (RbAM) is a form of AM focusing on identifying agreement (support) and disagreement (attack) relations amongst arguments. RbAM is a challenging classification task, with existing methods failing to perform satisfactorily. In this paper, we show that general-purpose Large Language Models (LLMs), appropriately primed and prompted, can significantly outperform the best performing (RoBERTa-based) baseline. Specifically, we experiment with two open-source LLMs (Llama-2 and Mistral) with ten datasets.

Xiang Yin, Nico Potyka, Francesca Toni

Explaining the strength of arguments under gradual semantics is receiving increasing attention. For example, various studies in the literature offer explanations by computing the attribution scores of arguments or edges in Quantitative Bipolar Argumentation Frameworks (QBAFs). These explanations, known as Argument Attribution Explanations (AAEs) and Relation Attribution Explanations (RAEs), commonly employ removal-based and Shapley-based techniques for computing the attribution scores. While AAEs and RAEs have proven useful in several applications with acyclic QBAFs, they remain largely unexplored for cyclic QBAFs. Furthermore, existing applications tend to focus solely on either AAEs or RAEs, but do not compare them directly. In this paper, we apply both AAEs and RAEs, to Truth Discovery QBAFs (TD-QBAFs), which assess the trustworthiness of sources (e.g., websites) and their claims (e.g., the severity of a virus), and feature complex cycles. We find that both AAEs and RAEs can provide interesting explanations and can give non-trivial and surprising insights.

Aniol Civit, Antonio Rago, Antonio Andriella et al.

Gradual argumentation is a field of symbolic AI which is attracting attention for its ability to support transparent and contestable AI systems. It is considered a useful tool in domains such as decision-making, recommendation, debate analysis, and others. The outcomes in such domains are usually dependent on the arguments' base scores, which must be selected carefully. Often, this selection process requires user expertise and may not always be straightforward. On the other hand, organising the arguments by preference could simplify the task. In this work, we introduce \emph{Base Score Extraction Functions}, which provide a mapping from users' preferences over arguments to base scores. These functions can be applied to the arguments of a \emph{Bipolar Argumentation Framework} (BAF), supplemented with preferences, to obtain a \emph{Quantitative Bipolar Argumentation Framework} (QBAF), allowing the use of well-established computational tools in gradual argumentation. We outline the desirable properties of base score extraction functions, discuss some design choices, and provide an algorithm for base score extraction. Our method incorporates an approximation of non-linearities in human preferences to allow for better approximation of the real ones. Finally, we evaluate our approach both theoretically and experimentally in a robotics setting, and offer recommendations for selecting appropriate gradual semantics in practice.

Gabriel Freedman, Adam Dejl, Adam Gould et al.

Claim verification is an important problem in high-stakes settings, including health and finance. When information underpinning claims is incomplete or conflicting, uncertain answers may be more appropriate than binary true or false classifications. In all cases, faithful explanations of the considerations determining the final verdict are crucial. We introduce inference-time argumentation (ITA), a trainable neurosymbolic framework for ternary claim verification in which a formal argumentation semantics giving the strength of claims is used both (i) to guide LLM training as models learn to generate arguments and assign them base scores (representing intrinsic strengths) and (ii) to compute ternary (true/false/uncertain) predictions from generated, scored arguments. As a result, at training time, argument generation and scoring can be optimised according to the quality of the induced argumentative predictions. Moreover, at inference time, the final prediction is faithful, by construction, to the arguments and scores determining the verdict, rather than being justified by a potentially unfaithful post-hoc reasoning trace as in conventional reasoning models. We finally show that, on two datasets for ternary claim verification, ITA improves upon argumentative baselines and can perform competitively against non-argumentative direct-prediction baselines, while providing verdicts that are computed deterministically from explicit, inspectable argumentative structures.

Emanuele De Angelis, Maurizio Proietti, Francesca Toni

Assumption-based Argumentation (ABA) is advocated as a unifying formalism for various forms of non-monotonic reasoning, including logic programming. It allows capturing defeasible knowledge, subject to argumentative debate. While, in much existing work, ABA frameworks are given up-front, in this paper we focus on the problem of automating their learning from background knowledge and positive/negative examples. Unlike prior work, we newly frame the problem in terms of brave reasoning under stable extensions for ABA. We present a novel algorithm based on transformation rules (such as Rote Learning, Folding, Assumption Introduction and Fact Subsumption) and an implementation thereof that makes use of Answer Set Programming. Finally, we compare our technique to state-of-the-art ILP systems that learn defeasible knowledge.

Adam Dejl, Dekai Zhang, Hamed Ayoobi et al.

Artificial Neural Networks (ANNs) often represent conflicts between features, arising naturally during training as the network learns to integrate diverse and potentially disagreeing inputs to better predict the target variable. Despite their relevance to the ``reasoning'' processes of these models, the properties and implications of conflicts for understanding and explaining ANNs remain underexplored. In this paper, we develop a rigorous theory of conflicts in ANNs and demonstrate their impact on ANN explainability through two case studies. In the first case study, we use our theory of conflicts to inspire the design of a novel feature attribution method, which we call Conflict-Aware Feature-wise Explanations (CAFE). CAFE separates the positive and negative influences of features and biases, enabling more faithful explanations for models applied to tabular data. In the second case study, we take preliminary steps towards understanding the role of conflicts in out-of-distribution (OOD) scenarios. Through our experiments, we identify potentially useful connections between model conflicts and different kinds of distributional shifts in tabular and image data. Overall, our findings demonstrate the importance of accounting for conflicts in the development of more reliable explanation methods for AI systems, which are crucial for the beneficial use of these systems in the society.

Preesha Gehlot, Anna Rapberger, Fabrizio Russo et al.

Assumption-Based Argumentation (ABA) is a powerful structured argumentation formalism, but exact computation of extensions under stable semantics is intractable for large frameworks. We present the first Graph Neural Network (GNN) approach to approximate credulous acceptance in ABA. To leverage GNNs, we model ABA frameworks via a dependency graph representation encoding assumptions, claims and rules as nodes, with heterogeneous edge labels distinguishing support, derive and attack relations. We propose two GNN architectures - ABAGCN and ABAGAT - that stack residual heterogeneous convolution or attention layers, respectively, to learn node embeddings. Our models are trained on the ICCMA 2023 benchmark, augmented with synthetic ABAFs, with hyperparameters optimised via Bayesian search. Empirically, both ABAGCN and ABAGAT outperform a state-of-the-art GNN baseline that we adapt from the abstract argumentation literature, achieving a node-level F1 score of up to 0.71 on the ICCMA instances. Finally, we develop a sound polynomial time extension-reconstruction algorithm driven by our predictor: it reconstructs stable extensions with F1 above 0.85 on small ABAFs and maintains an F1 of about 0.58 on large frameworks. Our work opens new avenues for scalable approximate reasoning in structured argumentation.

Adam Dejl, Matthew Williams, Francesca Toni

Large language models (LLMs) exhibit strong general capabilities, but their deployment in high-stakes domains is hindered by their opacity and unpredictability. Recent work has taken meaningful steps towards addressing these issues by augmenting LLMs with post-hoc reasoning based on computational argumentation, providing faithful explanations and enabling users to contest incorrect decisions. However, this paradigm is limited to pre-defined binary choices and only supports local contestation for specific instances, leaving the underlying decision logic unchanged and prone to repeated mistakes. In this paper, we introduce ArgEval, a framework that shifts from instance-specific reasoning to structured evaluation of general decision options. Rather than mining arguments solely for individual cases, ArgEval systematically maps task-specific decision spaces, builds corresponding option ontologies, and constructs general argumentation frameworks (AFs) for each option. These frameworks can then be instantiated to provide explainable recommendations for specific cases while still supporting global contestability through modification of the shared AFs. We investigate the effectiveness of ArgEval on treatment recommendation for glioblastoma, an aggressive brain tumour, and show that it can produce explainable guidance aligned with clinical practice.

Adam Gould, Guilherme Paulino-Passos, Seema Dadhania et al.

In the pursuit of enhancing the efficacy and flexibility of interpretable, data-driven classification models, this work introduces a novel incorporation of user-defined preferences with Abstract Argumentation and Case-Based Reasoning (CBR). Specifically, we introduce Preference-Based Abstract Argumentation for Case-Based Reasoning (which we call AA-CBR-P), allowing users to define multiple approaches to compare cases with an ordering that specifies their preference over these comparison approaches. We prove that the model inherently follows these preferences when making predictions and show that previous abstract argumentation for case-based reasoning approaches are insufficient at expressing preferences over constituents of an argument. We then demonstrate how this can be applied to a real-world medical dataset sourced from a clinical trial evaluating differing assessment methods of patients with a primary brain tumour. We show empirically that our approach outperforms other interpretable machine learning models on this dataset.

Junqi Jiang, Tom Bewley, Salim I. Amoukou et al.

Test-time scaling improves large language models' (LLMs) performance by allocating more compute budget during inference. To achieve this, existing methods often require intricate modifications to prompting and sampling strategies. In this work, we introduce representation consistency (RC), a test-time scaling method for aggregating answers drawn from multiple candidate responses of an LLM regardless of how they were generated, including variations in prompt phrasing and sampling strategy. RC enhances answer aggregation by not only considering the number of occurrences of each answer in the candidate response set, but also the consistency of the model's internal activations while generating the set of responses leading to each answer. These activations can be either dense (raw model activations) or sparse (encoded via pretrained sparse autoencoders). Our rationale is that if the model's representations of multiple responses converging on the same answer are highly variable, this answer is more likely to be the result of incoherent reasoning and should be down-weighted during aggregation. Importantly, our method only uses cached activations and lightweight similarity computations and requires no additional model queries. Through experiments with four open-source LLMs and four reasoning datasets, we validate the effectiveness of RC for improving task performance during inference, with consistent accuracy improvements (up to 4%) over strong test-time scaling baselines. We also show that consistency in the sparse activation signals aligns well with the common notion of coherent reasoning.

Jun Rui Lee, Sadegh Emami, Michael David Hollins et al.

Feature attribution (FA) methods are widely used in explainable AI (XAI) to help users understand how the inputs of a machine learning model contribute to its outputs. However, different FA models often provide disagreeing importance scores for the same model. In the absence of ground truth or in-depth knowledge about the inner workings of the model, it is often difficult to meaningfully determine which of the different FA methods produce more suitable explanations in different contexts. As a step towards addressing this issue, we introduce the open-source XAI-Units benchmark, specifically designed to evaluate FA methods against diverse types of model behaviours, such as feature interactions, cancellations, and discontinuous outputs. Our benchmark provides a set of paired datasets and models with known internal mechanisms, establishing clear expectations for desirable attribution scores. Accompanied by a suite of built-in evaluation metrics, XAI-Units streamlines systematic experimentation and reveals how FA methods perform against distinct, atomic kinds of model reasoning, similar to unit tests in software engineering. Crucially, by using procedurally generated models tied to synthetic datasets, we pave the way towards an objective and reliable comparison of FA methods.

Avinash Kori, Antonio Rago, Francesca Toni

Deep learning models are powerful image classifiers but their opacity hinders their trustworthiness. Explanation methods for capturing the reasoning process within these classifiers faithfully and in a clear manner are scarce, due to their sheer complexity and size. We provide a solution for this problem by defining a novel method for explaining the outputs of image classifiers with debates between two agents, each arguing for a particular class. We obtain these debates as concrete instances of Free Argumentative eXchanges (FAXs), a novel argumentation-based multi-agent framework allowing agents to internalise opinions by other agents differently than originally stated. We define two metrics (consensus and persuasion rate) to assess the usefulness of FAXs as argumentative explanations for image classifiers. We then conduct a number of empirical experiments showing that FAXs perform well along these metrics as well as being more faithful to the image classifiers than conventional, non-argumentative explanation methods. All our implementations can be found at https://github.com/koriavinash1/FAX.

Lihu Chen, Xiang Yin, Francesca Toni

Understanding the internal thinking process of Large Language Models (LLMs) and the cause of hallucinations remains a key challenge. To this end, we introduce latent debate, a novel framework for interpreting model predictions through the lens of implicit internal arguments. Unlike the current work of self-consistency and multi-agent debate, which relies on explicit debates among multiple answers or multiple models, latent debate captures the hidden supporting and attacking signals that arise within a single model during a single inference. We first present a model- and task-agnostic conceptual framework, and then instantiate it symbolically to approximate the thinking process of LLMs on True/False prediction tasks. Empirical studies demonstrate that latent debate is a faithful structured surrogate model that has highly consistent predictions with the original LLM. Beyond interpretability, we demonstrate that latent debate provides a strong baseline for hallucination detection. Further analysis reveals strong correlations between hallucinations and debate patterns, such as a high degree of latent debates in the middle layers is linked to a higher risk of hallucinations. These findings position latent debate as a potential framework for understanding internal mechanisms of LLMs, especially for scenarios where internal (dis)agreements appear during the inference steps.

Dekai Zhang, Matthew Williams, Francesca Toni

Neural networks (NNs) can learn to rely on spurious signals in the training data, leading to poor generalisation. Recent methods tackle this problem by training NNs with additional ground-truth annotations of such signals. These methods may, however, let spurious signals re-emerge in deep convolutional NNs (CNNs). We propose Targeted Activation Penalty (TAP), a new method tackling the same problem by penalising activations to control the re-emergence of spurious signals in deep CNNs, while also lowering training times and memory usage. In addition, ground-truth annotations can be expensive to obtain. We show that TAP still works well with annotations generated by pre-trained models as effective substitutes of ground-truth annotations. We demonstrate the power of TAP against two state-of-the-art baselines on the MNIST benchmark and on two clinical image datasets, using four different CNN architectures.

Adam Gould, Francesca Toni

Deep learning has become the dominant approach for creating high capacity, scalable models across diverse data modalities. However, because these models rely on a large number of learned parameters, tightly couple feature extraction with task objectives, and often lack explicit reasoning mechanisms, it is difficult for humans to understand how they arrive at their predictions. Understanding what representations emerge and why they arise from the training data remains an open challenge. We introduce Deep Arguing, a novel neurosymbolic approach that integrates deep learning with argumentation construction and reasoning for interpretable classification with different data modalities. In our approach deep neural networks construct an argumentation structure wherein data points support their assigned label and attack different ones. Using differentiable argumentation semantics for reasoning, the model is trained end-to-end to jointly learn feature representation and argumentative interactions. This results in argumentation structures providing faithful case-based explanations for predictions. Structure constraints over the argumentation graph guide learning, improving both interpretability and predictive performance. Experiments with tabular and imaging datasets show that Deep Arguing achieves performance competitive with standard baselines whilst offering interpretable argumentative reasoning.

Ming Pok Ng, Junqi Jiang, Gabriel Freedman et al.

Leveraging outputs from multiple large language models (LLMs) is emerging as a method for harnessing their power across a wide range of tasks while mitigating their capacity for making errors, e.g., hallucinations. However, current approaches to combining insights from multiple LLMs often involve unstructured interactions (e.g., free debate), resulting in model generations that are not faithfully justifiable. In this work, we introduce MArgE, a novel framework to provide formal structure to the evidence from each LLM, in the form of a tree of extracted arguments, for the task of claim verification. We use a variant of Argumentative LLMs (ArgLLMs), i.e. LLMs driven by frameworks and semantics from the field of computational argumentation, to construct structured argument trees for given claims. This process creates an inspectable pathway from the initial arguments to the final claim verification decisions, providing a faithful justification thereof. We show experimentally that MArgE can significantly outperform single LLMs, including three open-source models (4B to 8B parameters), GPT-4o-mini and existing ArgLLMs, as well as prior methods for unstructured multi-LLM debates. We thus demonstrate the advantages of incorporating formal, argumentative reasoning mechanisms when combining multiple LLM outputs.

Lihu Chen, Shuojie Fu, Gabriel Freedman et al.

A significant and growing number of published scientific articles is found to involve fraudulent practices, posing a serious threat to the credibility and safety of research in fields such as medicine. We propose Pub-Guard-LLM, the first large language model-based system tailored to fraud detection of biomedical scientific articles. We provide three application modes for deploying Pub-Guard-LLM: vanilla reasoning, retrieval-augmented generation, and multi-agent debate. Each mode allows for textual explanations of predictions. To assess the performance of our system, we introduce an open-source benchmark, PubMed Retraction, comprising over 11K real-world biomedical articles, including metadata and retraction labels. We show that, across all modes, Pub-Guard-LLM consistently surpasses the performance of various baselines and provides more reliable explanations, namely explanations which are deemed more relevant and coherent than those generated by the baselines when evaluated by multiple assessment methods. By enhancing both detection performance and explainability in scientific fraud detection, Pub-Guard-LLM contributes to safeguarding research integrity with a novel, effective, open-source tool.

Guilherme Paulino-Passos, Francesca Toni

Case-based reasoning is known to play an important role in several legal settings. In this paper we focus on a recent approach to case-based reasoning, supported by an instantiation of abstract argumentation whereby arguments represent cases and attack between arguments results from outcome disagreement between cases and a notion of relevance. In this context, relevance is connected to a form of specificity among cases. We explore how relevance can be learnt automatically in practice with the help of decision trees, and explore the combination of case-based reasoning with abstract argumentation (AA-CBR) and learning of case relevance for prediction in legal settings. Specifically, we show that, for two legal datasets, AA-CBR and decision-tree-based learning of case relevance perform competitively in comparison with decision trees. We also show that AA-CBR with decision-tree-based learning of case relevance results in a more compact representation than their decision tree counterparts, which could be beneficial for obtaining cognitively tractable explanations.

Junqi Jiang, Francesco Leofante, Antonio Rago et al.

Counterfactual explanations (CEs) are advocated as being ideally suited to providing algorithmic recourse for subjects affected by the predictions of machine learning models. While CEs can be beneficial to affected individuals, recent work has exposed severe issues related to the robustness of state-of-the-art methods for obtaining CEs. Since a lack of robustness may compromise the validity of CEs, techniques to mitigate this risk are in order. In this survey, we review works in the rapidly growing area of robust CEs and perform an in-depth analysis of the forms of robustness they consider. We also discuss existing solutions and their limitations, providing a solid foundation for future developments.