Research Papers

Decision processes of computer vision models - especially deep neural networks - are opaque in nature, meaning that these decisions cannot be understood by humans. Thus, over the last years, many methods to provide human-understandable explanations have been proposed. For image classification, the most common group are saliency methods, which provide (super-)pixelwise feature attribution scores for input images. But their evaluation still poses a problem, as their results cannot be simply compared to the unknown ground truth. To overcome this, a slew of different proxy metrics have been defined, which are - as the explainability methods themselves - often built on intuition and thus, are possibly unreliable. In this paper, new evaluation metrics for saliency methods are developed and common saliency methods are benchmarked on ImageNet. In addition, a scheme for reliability evaluation of such metrics is proposed that is based on concepts from psychometric testing. The used code can be found at https://github.com/lelo204/ClassificationMetricsForImageExplanations .

We present a method for quantitative, in-depth analyses of fairness issues in AI systems with an application to credit scoring. To this aim we use BRIO, a tool for the evaluation of AI systems with respect to social unfairness and, more in general, ethically undesirable behaviours. It features a model-agnostic bias detection module, presented in \cite{DBLP:conf/beware/CoragliaDGGPPQ23}, to which a full-fledged unfairness risk evaluation module is added. As a case study, we focus on the context of credit scoring, analysing the UCI German Credit Dataset \cite{misc_statlog_(german_credit_data)_144}. We apply the BRIO fairness metrics to several, socially sensitive attributes featured in the German Credit Dataset, quantifying fairness across various demographic segments, with the aim of identifying potential sources of bias and discrimination in a credit scoring model. We conclude by combining our results with a revenue analysis.

Artificial intelligence (AI) is interacting with people at an unprecedented scale, offering new avenues for immense positive impact, but also raising widespread concerns around the potential for individual and societal harm. Today, the predominant paradigm for human--AI safety focuses on fine-tuning the generative model's outputs to better agree with human-provided examples or feedback. In reality, however, the consequences of an AI model's outputs cannot be determined in isolation: they are tightly entangled with the responses and behavior of human users over time. In this paper, we distill key complementary lessons from AI safety and control systems safety, highlighting open challenges as well as key synergies between both fields. We then argue that meaningful safety assurances for advanced AI technologies require reasoning about how the feedback loop formed by AI outputs and human behavior may drive the interaction towards different outcomes. To this end, we introduce a unifying formalism to capture dynamic, safety-critical human--AI interactions and propose a concrete technical roadmap towards next-generation human-centered AI safety.

Researchers urge technology practitioners such as data scientists to consider the impacts and ethical implications of algorithmic decisions. However, unlike programming, statistics, and data management, discussion of ethical implications is rarely included in standard data science training. To begin to address this gap, we designed and tested a toolbox called the data ethics emergency drill (DEED) to help data science teams discuss and reflect on the ethical implications of their work. The DEED is a roleplay of a fictional ethical emergency scenario that is contextually situated in the team's specific workplace and applications. This paper outlines the DEED toolbox and describes three studies carried out with two different data science teams that iteratively shaped its design. Our findings show that practitioners can apply lessons learnt from the roleplay to real-life situations, and how the DEED opened up conversations around ethics and values.

LIME (Local Interpretable Model-agnostic Explanations) is a popular XAI framework for unraveling decision-making processes in vision machine-learning models. The technique utilizes image segmentation methods to identify fixed regions for calculating feature importance scores as explanations. Therefore, poor segmentation can weaken the explanation and reduce the importance of segments, ultimately affecting the overall clarity of interpretation. To address these challenges, we introduce the DSEG-LIME (Data-Driven Segmentation LIME) framework, featuring: i) a data-driven segmentation for human-recognized feature generation by foundation model integration, and ii) a user-steered granularity in the hierarchical segmentation procedure through composition. Our findings demonstrate that DSEG outperforms on several XAI metrics on pre-trained ImageNet models and improves the alignment of explanations with human-recognized concepts. The code is available under: https://github. com/patrick-knab/DSEG-LIME

Over the last 70 years, we, humans, have created an economic market where attention is being captured and turned into money thanks to advertising. During the last two decades, leveraging research in psychology, sociology, neuroscience and other domains, Web platforms have brought the process of capturing attention to an unprecedented scale. With the initial commonplace goal of making targeted advertising more effective, the generalization of attention-capturing techniques and their use of cognitive biases and emotions have multiple detrimental side effects such as polarizing opinions, spreading false information and threatening public health, economies and democracies. This is clearly a case where the Web is not used for the common good and where, in fact, all its users become a vulnerable population. This paper brings together contributions from a wide range of disciplines to analyze current practices and consequences thereof. Through a set of propositions and principles that could be used do drive further works, it calls for actions against these practices competing to capture our attention on the Web, as it would be unsustainable for a civilization to allow attention to be wasted with impunity on a world-wide scale.

When users perceive AI systems as mindful, independent agents, they hold them responsible instead of the AI experts who created and designed these systems. So far, it has not been studied whether explanations support this shift in responsibility through the use of mind-attributing verbs like "to think". To better understand the prevalence of mind-attributing explanations we analyse AI explanations in 3,533 explainable AI (XAI) research articles from the Semantic Scholar Open Research Corpus (S2ORC). Using methods from semantic shift detection, we identify three dominant types of mind attribution: (1) metaphorical (e.g. "to learn" or "to predict"), (2) awareness (e.g. "to consider"), and (3) agency (e.g. "to make decisions"). We then analyse the impact of mind-attributing explanations on awareness and responsibility in a vignette-based experiment with 199 participants. We find that participants who were given a mind-attributing explanation were more likely to rate the AI system as aware of the harm it caused. Moreover, the mind-attributing explanation had a responsibility-concealing effect: Considering the AI experts' involvement lead to reduced ratings of AI responsibility for participants who were given a non-mind-attributing or no explanation. In contrast, participants who read the mind-attributing explanation still held the AI system responsible despite considering the AI experts' involvement. Taken together, our work underlines the need to carefully phrase explanations about AI systems in scientific writing to reduce mind attribution and clearly communicate human responsibility.

The evaluation of explainable artificial intelligence is challenging, because automated and human-centred metrics of explanation quality may diverge. To clarify their relationship, we investigated whether human and artificial image classification will benefit from the same visual explanations. In three experiments, we analysed human reaction times, errors, and subjective ratings while participants classified image segments. These segments either reflected human attention (eye movements, manual selections) or the outputs of two attribution methods explaining a ResNet (Grad-CAM, XRAI). We also had this model classify the same segments. Humans and the model largely agreed on the interpretability of attribution methods: Grad-CAM was easily interpretable for indoor scenes and landscapes, but not for objects, while the reverse pattern was observed for XRAI. Conversely, human and model performance diverged for human-generated segments. Our results caution against general statements about interpretability, as it varies with the explanation method, the explained images, and the agent interpreting them.

The field of explainable artificial intelligence (XAI) aims to explain how black-box machine learning models work. Much of the work centers around the holy grail of providing post-hoc feature attributions to any model architecture. While the pace of innovation around novel methods has slowed down, the question remains of how to choose a method, and how to make it fit for purpose. Recently, efforts around benchmarking XAI methods have suggested metrics for that purpose -- but there are many choices. That bounty of choice still leaves an end user unclear on how to proceed. This paper focuses on comparing metrics with the aim of measuring faithfulness of local explanations on tabular classification problems -- and shows that the current metrics don't agree; leaving users unsure how to choose the most faithful explanations.

Regulation and NLP (RegNLP): Taming Large Language Models

Catalina Goanta, Nikolaos Aletras, Ilias Chalkidis +2

The scientific innovation in Natural Language Processing (NLP) and more broadly in artificial intelligence (AI) is at its fastest pace to date. As large language models (LLMs) unleash a new era of automation, important debates emerge regarding the benefits and risks of their development, deployment and use. Currently, these debates have been dominated by often polarized narratives mainly led by the AI Safety and AI Ethics movements. This polarization, often amplified by social media, is swaying political agendas on AI regulation and governance and posing issues of regulatory capture. Capture occurs when the regulator advances the interests of the industry it is supposed to regulate, or of special interest groups rather than pursuing the general public interest. Meanwhile in NLP research, attention has been increasingly paid to the discussion of regulating risks and harms. This often happens without systematic methodologies or sufficient rooting in the disciplines that inspire an extended scope of NLP research, jeopardizing the scientific integrity of these endeavors. Regulation studies are a rich source of knowledge on how to systematically deal with risk and uncertainty, as well as with scientific evidence, to evaluate and compare regulatory options. This resource has largely remained untapped so far. In this paper, we argue how NLP research on these topics can benefit from proximity to regulatory studies and adjacent fields. We do so by discussing basic tenets of regulation, and risk and uncertainty, and by highlighting the shortcomings of current NLP discussions dealing with risk assessment. Finally, we advocate for the development of a new multidisciplinary research space on regulation and NLP (RegNLP), focused on connecting scientific knowledge to regulatory processes based on systematic methodologies.

Model fairness (a.k.a., bias) has become one of the most critical problems in a wide range of AI applications. An unfair model in autonomous driving may cause a traffic accident if corner cases (e.g., extreme weather) cannot be fairly regarded; or it will incur healthcare disparities if the AI model misdiagnoses a certain group of people (e.g., brown and black skin). In recent years, there have been emerging research works on addressing unfairness, and they mainly focus on a single unfair attribute, like skin tone; however, real-world data commonly have multiple attributes, among which unfairness can exist in more than one attribute, called 'multi-dimensional fairness'. In this paper, we first reveal a strong correlation between the different unfair attributes, i.e., optimizing fairness on one attribute will lead to the collapse of others. Then, we propose a novel Multi-Dimension Fairness framework, namely Muffin, which includes an automatic tool to unite off-the-shelf models to improve the fairness on multiple attributes simultaneously. Case studies on dermatology datasets with two unfair attributes show that the existing approach can achieve 21.05% fairness improvement on the first attribute while it makes the second attribute unfair by 1.85%. On the other hand, the proposed Muffin can unite multiple models to achieve simultaneously 26.32% and 20.37% fairness improvement on both attributes; meanwhile, it obtains 5.58% accuracy gain.

The future of the arts and artificial intelligence (AI) is promising as technology advances. As the use of AI in design becomes more widespread, art practice may not be a human-only art form and could instead become a digitally integrated experience. With enhanced creativity and collaboration, arts and AI could work together towards creating artistic outputs that are visually appealing and meet the needs of the artist and viewer. While it is uncertain how far the integration will go, arts and AI will likely influence one another. This workshop pictorial puts forward first-person research that shares interactions between an HCI researcher and AI as they try to escape the creative block. The pictorial paper explores two questions: How can AI support artists' creativity, and what does it mean to be explainable in this context? HIs, ChatGPT and Midjourney were engaged; the result was a series of reflections that require further discussion and explorations in the XAIxArts community: Transparency of attribution, the creation process, ethics of asking, and inspiration vs copying.