Research Papers

RESHAPE: Explaining Accounting Anomalies in Financial Statement Audits by enhancing SHapley Additive exPlanations

Ricardo Müller, Marco Schreyer, Timur Sattarov +1

Detecting accounting anomalies is a recurrent challenge in financial statement audits. Recently, novel methods derived from Deep-Learning (DL) have been proposed to audit the large volumes of a statement's underlying accounting records. However, due to their vast number of parameters, such models exhibit the drawback of being inherently opaque. At the same time, the concealing of a model's inner workings often hinders its real-world application. This observation holds particularly true in financial audits since auditors must reasonably explain and justify their audit decisions. Nowadays, various Explainable AI (XAI) techniques have been proposed to address this challenge, e.g., SHapley Additive exPlanations (SHAP). However, in unsupervised DL as often applied in financial audits, these methods explain the model output at the level of encoded variables. As a result, the explanations of Autoencoder Neural Networks (AENNs) are often hard to comprehend by human auditors. To mitigate this drawback, we propose (RESHAPE), which explains the model output on an aggregated attribute-level. In addition, we introduce an evaluation framework to compare the versatility of XAI methods in auditing. Our experimental results show empirical evidence that RESHAPE results in versatile explanations compared to state-of-the-art baselines. We envision such attribute-level explanations as a necessary next step in the adoption of unsupervised DL techniques in financial auditing.

Recent AI algorithms are black box models whose decisions are difficult to interpret. eXplainable AI (XAI) is a class of methods that seek to address lack of AI interpretability and trust by explaining to customers their AI decisions. The common wisdom is that regulating AI by mandating fully transparent XAI leads to greater social welfare. Our paper challenges this notion through a game theoretic model of a policy-maker who maximizes social welfare, firms in a duopoly competition that maximize profits, and heterogenous consumers. The results show that XAI regulation may be redundant. In fact, mandating fully transparent XAI may make firms and consumers worse off. This reveals a tradeoff between maximizing welfare and receiving explainable AI outputs. We extend the existing literature on method and substantive fronts, and we introduce and study the notion of XAI fairness, which may be impossible to guarantee even under mandatory XAI. Finally, the regulatory and managerial implications of our results for policy-makers and businesses are discussed, respectively.

The classification of internet traffic has become increasingly important due to the rapid growth of today's networks and applications. The number of connections and the addition of new applications in our networks causes a vast amount of log data and complicates the search for common patterns by experts. Finding such patterns among specific classes of applications is necessary to fulfill various requirements in network analytics. Deep learning methods provide both feature extraction and classification from data in a single system. However, these networks are very complex and are used as black-box models, which weakens the experts' trust in the classifications. Moreover, by using them as a black-box, new knowledge cannot be obtained from the model predictions despite their excellent performance. Therefore, the explainability of the classifications is crucial. Besides increasing trust, the explanation can be used for model evaluation gaining new insights from the data and improving the model. In this paper, we present a visual interactive tool that combines the classification of network data with an explanation technique to form an interface between experts, algorithms, and data.

If autonomous AI systems are to be reliably safe in novel situations, they will need to incorporate general principles guiding them to recognize and avoid harmful behaviours. Such principles may need to be supported by a binding system of regulation, which would need the underlying principles to be widely accepted. They should also be specific enough for technical implementation. Drawing inspiration from law, this article explains how negative human rights could fulfil the role of such principles and serve as a foundation both for an international regulatory system and for building technical safety constraints for future AI systems.

RestoreX-AI: A Contrastive Approach towards Guiding Image Restoration via Explainable AI Systems

Aboli Marathe, Pushkar Jain, Rahee Walambe +1

Modern applications such as self-driving cars and drones rely heavily upon robust object detection techniques. However, weather corruptions can hinder the object detectability and pose a serious threat to their navigation and reliability. Thus, there is a need for efficient denoising, deraining, and restoration techniques. Generative adversarial networks and transformers have been widely adopted for image restoration. However, the training of these methods is often unstable and time-consuming. Furthermore, when used for object detection (OD), the output images generated by these methods may provide unsatisfactory results despite image clarity. In this work, we propose a contrastive approach towards mitigating this problem, by evaluating images generated by restoration models during and post training. This approach leverages OD scores combined with attention maps for predicting the usefulness of restored images for the OD task. We conduct experiments using two novel use-cases of conditional GANs and two transformer methods that probe the robustness of the proposed approach on multi-weather corruptions in the OD task. Our approach achieves an averaged 178 percent increase in mAP between the input and restored images under adverse weather conditions like dust tornadoes and snowfall. We report unique cases where greater denoising does not improve OD performance and conversely where noisy generated images demonstrate good results. We conclude the need for explainability frameworks to bridge the gap between human and machine perception, especially in the context of robust object detection for autonomous vehicles.

Predictive algorithms have a powerful potential to offer benefits in areas as varied as medicine or education. However, these algorithms and the data they use are built by humans, consequently, they can inherit the bias and prejudices present in humans. The outcomes can systematically repeat errors that create unfair results, which can even lead to situations of discrimination (e.g. gender, social or racial). In order to illustrate how important is to count with a diverse training dataset to avoid bias, we manipulate a well-known facial expression recognition dataset to explore gender bias and discuss its implications.

The nondeterminism of Deep Learning (DL) training algorithms and its influence on the explainability of neural network (NN) models are investigated in this work with the help of image classification examples. To discuss the issue, two convolutional neural networks (CNN) have been trained and their results compared. The comparison serves the exploration of the feasibility of creating deterministic, robust DL models and deterministic explainable artificial intelligence (XAI) in practice. Successes and limitation of all here carried out efforts are described in detail. The source code of the attained deterministic models has been listed in this work. Reproducibility is indexed as a development-phase-component of the Model Governance Framework, proposed by the EU within their excellence in AI approach. Furthermore, reproducibility is a requirement for establishing causality for the interpretation of model results and building of trust towards the overwhelming expansion of AI systems applications. Problems that have to be solved on the way to reproducibility and ways to deal with some of them, are examined in this work.

Intelligent decision support (IDS) systems leverage artificial intelligence techniques to generate recommendations that guide human users through the decision making phases of a task. However, a key challenge is that IDS systems are not perfect, and in complex real-world scenarios may produce incorrect output or fail to work altogether. The field of explainable AI planning (XAIP) has sought to develop techniques that make the decision making of sequential decision making AI systems more explainable to end-users. Critically, prior work in applying XAIP techniques to IDS systems has assumed that the plan being proposed by the planner is always optimal, and therefore the action or plan being recommended as decision support to the user is always correct. In this work, we examine novice user interactions with a non-robust IDS system -- one that occasionally recommends the wrong action, and one that may become unavailable after users have become accustomed to its guidance. We introduce a novel explanation type, subgoal-based explanations, for planning-based IDS systems, that supplements traditional IDS output with information about the subgoal toward which the recommended action would contribute. We demonstrate that subgoal-based explanations lead to improved user task performance, improve user ability to distinguish optimal and suboptimal IDS recommendations, are preferred by users, and enable more robust user performance in the case of IDS failure

Identifying Biased Subgroups in Ranking and Classification

Eliana Pastor, Luca de Alfaro, Elena Baralis

When analyzing the behavior of machine learning algorithms, it is important to identify specific data subgroups for which the considered algorithm shows different performance with respect to the entire dataset. The intervention of domain experts is normally required to identify relevant attributes that define these subgroups. We introduce the notion of divergence to measure this performance difference and we exploit it in the context of (i) classification models and (ii) ranking applications to automatically detect data subgroups showing a significant deviation in their behavior. Furthermore, we quantify the contribution of all attributes in the data subgroup to the divergent behavior by means of Shapley values, thus allowing the identification of the most impacting attributes.

In Oxford Handbook on AI Governance: The Role of Workers in AI Ethics and Governance

Natalia Luka, JS Tan

While the role of states, corporations, and international organizations in AI governance has been extensively theorized, the role of workers has received comparatively little attention. This chapter looks at the role that workers play in identifying and mitigating harms from AI technologies. Harms are the causally assessed impacts of technologies. They arise despite technical reliability and are not a result of technical negligence but rather of normative uncertainty around questions of safety and fairness in complex social systems. There is high consensus in the AI ethics community on the benefits of reducing harms but less consensus on mechanisms for determining or addressing harms. This lack of consensus has resulted in a number of collective actions by workers protesting how harms are identified and addressed in their workplace. We theorize the role of workers within AI governance and construct a model of harm reporting processes in AI workplaces. The harm reporting process involves three steps, identification, the governance decision, and the response. Workers draw upon three types of claims to argue for jurisdiction over questions of AI governance, subjection, control over the product of labor, and proximate knowledge of systems. Examining the past decade of AI related worker activism allows us to understand how different types of workers are positioned within a workplace that produces AI systems, how their position informs their claims, and the place of collective action in staking their claims. This chapter argues that workers occupy a unique role in identifying and mitigating harms caused by AI systems.

Explaining firm decisions made by algorithms in customer-facing applications is increasingly required by regulators and expected by customers. While the emerging field of Explainable Artificial Intelligence (XAI) has mainly focused on developing algorithms that generate such explanations, there has not yet been sufficient consideration of customers' preferences for various types and formats of explanations. We discuss theoretically and study empirically people's preferences for explanations of algorithmic decisions. We focus on three main attributes that describe automatically-generated explanations from existing XAI algorithms (format, complexity, and specificity), and capture differences across contexts (online targeted advertising vs. loan applications) as well as heterogeneity in users' cognitive styles. Despite their popularity among academics, we find that counterfactual explanations are not popular among users, unless they follow a negative outcome (e.g., loan application was denied). We also find that users are willing to tolerate some complexity in explanations. Finally, our results suggest that preferences for specific (vs. more abstract) explanations are related to the level at which the decision is construed by the user, and to the deliberateness of the user's cognitive style.

In real-world image enhancement, it is often challenging (if not impossible) to acquire ground-truth data, preventing the adoption of distance metrics for objective quality assessment. As a result, one often resorts to subjective quality assessment, the most straightforward and reliable means of evaluating image enhancement. Conventional subjective testing requires manually pre-selecting a small set of visual examples, which may suffer from three sources of biases: 1) sampling bias due to the extremely sparse distribution of the selected samples in the image space; 2) algorithmic bias due to potential overfitting the selected samples; 3) subjective bias due to further potential cherry-picking test results. This eventually makes the field of real-world image enhancement more of an art than a science. Here we take steps towards debiasing conventional subjective assessment by automatically sampling a set of adaptive and diverse images for subsequent testing. This is achieved by casting sample selection into a joint maximization of the discrepancy between the enhancers and the diversity among the selected input images. Careful visual inspection on the resulting enhanced images provides a debiased ranking of the enhancement algorithms. We demonstrate our subjective assessment method using three popular and practically demanding image enhancement tasks: dehazing, super-resolution, and low-light enhancement.