Research Papers

AI Risk Atlas: Taxonomy and Tooling for Navigating AI Risks and Resources

Frank Bagehorn, Kristina Brimijoin, Elizabeth M. Daly +17

The rapid evolution of generative AI has expanded the breadth of risks associated with AI systems. While various taxonomies and frameworks exist to classify these risks, the lack of interoperability between them creates challenges for researchers, practitioners, and policymakers seeking to operationalise AI governance. To address this gap, we introduce the AI Risk Atlas, a structured taxonomy that consolidates AI risks from diverse sources and aligns them with governance frameworks. Additionally, we present the Risk Atlas Nexus, a collection of open-source tools designed to bridge the divide between risk definitions, benchmarks, datasets, and mitigation strategies. This knowledge-driven approach leverages ontologies and knowledge graphs to facilitate risk identification, prioritization, and mitigation. By integrating AI-assisted compliance workflows and automation strategies, our framework lowers the barrier to responsible AI adoption. We invite the broader research and open-source community to contribute to this evolving initiative, fostering cross-domain collaboration and ensuring AI governance keeps pace with technological advancements.

Ensuring fairness in decentralized multi-agent systems presents significant challenges due to emergent biases, systemic inefficiencies, and conflicting agent incentives. This paper provides a comprehensive survey of fairness in multi-agent AI, introducing a novel framework where fairness is treated as a dynamic, emergent property of agent interactions. The framework integrates fairness constraints, bias mitigation strategies, and incentive mechanisms to align autonomous agent behaviors with societal values while balancing efficiency and robustness. Through empirical validation, we demonstrate that incorporating fairness constraints results in more equitable decision-making. This work bridges the gap between AI ethics and system design, offering a foundation for accountable, transparent, and socially responsible multi-agent AI systems.

Human services organizations and the responsible integration of AI: Considering ethics and contextualizing risk(s)

Brian E. Perron, Lauri Goldkind, Zia Qi +1

This paper examines the responsible integration of artificial intelligence (AI) in human services organizations (HSOs), proposing a nuanced framework for evaluating AI applications across multiple dimensions of risk. The authors argue that ethical concerns about AI deployment -- including professional judgment displacement, environmental impact, model bias, and data laborer exploitation -- vary significantly based on implementation context and specific use cases. They challenge the binary view of AI adoption, demonstrating how different applications present varying levels of risk that can often be effectively managed through careful implementation strategies. The paper highlights promising solutions, such as local large language models, that can facilitate responsible AI integration while addressing common ethical concerns. The authors propose a dimensional risk assessment approach that considers factors like data sensitivity, professional oversight requirements, and potential impact on client wellbeing. They conclude by outlining a path forward that emphasizes empirical evaluation, starting with lower-risk applications and building evidence-based understanding through careful experimentation. This approach enables organizations to maintain high ethical standards while thoughtfully exploring how AI might enhance their capacity to serve clients and communities effectively.

Rerouting LLM Routers

Avital Shafran, Roei Schuster, Thomas Ristenpart +1

LLM routers aim to balance quality and cost of generation by classifying queries and routing them to a cheaper or more expensive LLM depending on their complexity. Routers represent one type of what we call LLM control planes: systems that orchestrate use of one or more LLMs. In this paper, we investigate routers' adversarial robustness. We first define LLM control plane integrity, i.e., robustness of LLM orchestration to adversarial inputs, as a distinct problem in AI safety. Next, we demonstrate that an adversary can generate query-independent token sequences we call ``confounder gadgets'' that, when added to any query, cause LLM routers to send the query to a strong LLM. Our quantitative evaluation shows that this attack is successful both in white-box and black-box settings against a variety of open-source and commercial routers, and that confounding queries do not affect the quality of LLM responses. Finally, we demonstrate that gadgets can be effective while maintaining low perplexity, thus perplexity-based filtering is not an effective defense. We finish by investigating alternative defenses.

Evaluating Explanations Through LLMs: Beyond Traditional User Studies

Francesco Bombassei De Bona, Gabriele Dominici, Tim Miller +2

As AI becomes fundamental in sectors like healthcare, explainable AI (XAI) tools are essential for trust and transparency. However, traditional user studies used to evaluate these tools are often costly, time consuming, and difficult to scale. In this paper, we explore the use of Large Language Models (LLMs) to replicate human participants to help streamline XAI evaluation. We reproduce a user study comparing counterfactual and causal explanations, replicating human participants with seven LLMs under various settings. Our results show that (i) LLMs can replicate most conclusions from the original study, (ii) different LLMs yield varying levels of alignment in the results, and (iii) experimental factors such as LLM memory and output variability affect alignment with human responses. These initial findings suggest that LLMs could provide a scalable and cost-effective way to simplify qualitative XAI evaluation.

Responsible AI (RAI) has emerged as a major focus across industry, policymaking, and academia, aiming to mitigate the risks and maximize the benefits of AI, both on an organizational and societal level. This study explores the global state of RAI through one of the most extensive surveys to date on the topic, surveying 1000 organizations across 20 industries and 19 geographical regions. We define a conceptual RAI maturity model for organizations to map how well they implement organizational and operational RAI measures. Based on this model, the survey assesses the adoption of system-level measures to mitigate identified risks related to, for example, discrimination, reliability, or privacy, and also covers key organizational processes pertaining to governance, risk management, and monitoring and control. The study highlights the expanding AI risk landscape, emphasizing the need for comprehensive risk mitigation strategies. The findings also reveal significant strides towards RAI maturity, but we also identify gaps in RAI implementation that could lead to increased (public) risks from AI systems. This research offers a structured approach to assess and improve RAI practices globally and underscores the critical need for bridging the gap between RAI planning and execution to ensure AI advancement aligns with human welfare and societal benefits.

This paper addresses the critical challenge of building consumer trust in AI-powered customer engagement by emphasising the necessity for transparency and accountability. Despite the potential of AI to revolutionise business operations and enhance customer experiences, widespread concerns about misinformation and the opacity of AI decision-making processes hinder trust. Surveys highlight a significant lack of awareness among consumers regarding their interactions with AI, alongside apprehensions about bias and fairness in AI algorithms. The paper advocates for the development of explainable AI models that are transparent and understandable to both consumers and organisational leaders, thereby mitigating potential biases and ensuring ethical use. It underscores the importance of organisational commitment to transparency practices beyond mere regulatory compliance, including fostering a culture of accountability, prioritising clear data policies and maintaining active engagement with stakeholders. By adopting a holistic approach to transparency and explainability, businesses can cultivate trust in AI technologies, bridging the gap between technological innovation and consumer acceptance, and paving the way for more ethical and effective AI-powered customer engagements. KEYWORDS: artificial intelligence (AI), transparency

Responsible artificial intelligence (RAI) is increasingly recognized as a critical concern. However, the level of corporate RAI prioritization has not kept pace. In this work, we conduct 16 semi-structured interviews with practitioners to investigate what has historically motivated companies to increase the prioritization of RAI. What emerges is a complex story of conflicting and varied factors, but we bring structure to the narrative by highlighting the different strategies available to employ, and point to the actors with access to each. While there are no guaranteed steps for increasing RAI prioritization, we paint the current landscape of motivators so that practitioners can learn from each other, and put forth our own selection of promising directions forward.

On the Value of Labeled Data and Symbolic Methods for Hidden Neuron Activation Analysis

Abhilekha Dalal, Rushrukh Rayan, Adrita Barua +3

A major challenge in Explainable AI is in correctly interpreting activations of hidden neurons: accurate interpretations would help answer the question of what a deep learning system internally detects as relevant in the input, demystifying the otherwise black-box nature of deep learning systems. The state of the art indicates that hidden node activations can, in some cases, be interpretable in a way that makes sense to humans, but systematic automated methods that would be able to hypothesize and verify interpretations of hidden neuron activations are underexplored. This is particularly the case for approaches that can both draw explanations from substantial background knowledge, and that are based on inherently explainable (symbolic) methods. In this paper, we introduce a novel model-agnostic post-hoc Explainable AI method demonstrating that it provides meaningful interpretations. Our approach is based on using a Wikipedia-derived concept hierarchy with approximately 2 million classes as background knowledge, and utilizes OWL-reasoning-based Concept Induction for explanation generation. Additionally, we explore and compare the capabilities of off-the-shelf pre-trained multimodal-based explainable methods. Our results indicate that our approach can automatically attach meaningful class expressions as explanations to individual neurons in the dense layer of a Convolutional Neural Network. Evaluation through statistical analysis and degree of concept activation in the hidden layer show that our method provides a competitive edge in both quantitative and qualitative aspects compared to prior work.

Deep learning models have achieved high performance in medical applications, however, their adoption in clinical practice is hindered due to their black-box nature. Self-explainable models, like prototype-based models, can be especially beneficial as they are interpretable by design. However, if the learnt prototypes are of low quality then the prototype-based models are as good as black-box. Having high quality prototypes is a pre-requisite for a truly interpretable model. In this work, we propose a prototype evaluation framework for coherence (PEF-C) for quantitatively evaluating the quality of the prototypes based on domain knowledge. We show the use of PEF-C in the context of breast cancer prediction using mammography. Existing works on prototype-based models on breast cancer prediction using mammography have focused on improving the classification performance of prototype-based models compared to black-box models and have evaluated prototype quality through anecdotal evidence. We are the first to go beyond anecdotal evidence and evaluate the quality of the mammography prototypes systematically using our PEF-C. Specifically, we apply three state-of-the-art prototype-based models, ProtoPNet, BRAIxProtoPNet++ and PIP-Net on mammography images for breast cancer prediction and evaluate these models w.r.t. i) classification performance, and ii) quality of the prototypes, on three public datasets. Our results show that prototype-based models are competitive with black-box models in terms of classification performance, and achieve a higher score in detecting ROIs. However, the quality of the prototypes are not yet sufficient and can be improved in aspects of relevance, purity and learning a variety of prototypes. We call the XAI community to systematically evaluate the quality of the prototypes to check their true usability in high stake decisions and improve such models further.

Explainable AI (XAI) aids in deciphering 'black-box' models. While several methods have been proposed and evaluated primarily in the image domain, the exploration of explainability in the text domain remains a growing research area. In this paper, we delve into the applicability of XAI methods for the text domain. In this context, the 'Similarity Difference and Uniqueness' (SIDU) XAI method, recognized for its superior capability in localizing entire salient regions in image-based classification is extended to textual data. The extended method, SIDU-TXT, utilizes feature activation maps from 'black-box' models to generate heatmaps at a granular, word-based level, thereby providing explanations that highlight contextually significant textual elements crucial for model predictions. Given the absence of a unified standard for assessing XAI methods, this study applies a holistic three-tiered comprehensive evaluation framework: Functionally-Grounded, Human-Grounded and Application-Grounded, to assess the effectiveness of the proposed SIDU-TXT across various experiments. We find that, in sentiment analysis task of a movie review dataset, SIDU-TXT excels in both functionally and human-grounded evaluations, demonstrating superior performance through quantitative and qualitative analyses compared to benchmarks like Grad-CAM and LIME. In the application-grounded evaluation within the sensitive and complex legal domain of asylum decision-making, SIDU-TXT and Grad-CAM demonstrate comparable performances, each with its own set of strengths and weaknesses. However, both methods fall short of entirely fulfilling the sophisticated criteria of expert expectations, highlighting the imperative need for additional research in XAI methods suitable for such domains.

A structured regression approach for evaluating model performance across intersectional subgroups

Christine Herlihy, Kimberly Truong, Alexandra Chouldechova +1

Disaggregated evaluation is a central task in AI fairness assessment, where the goal is to measure an AI system's performance across different subgroups defined by combinations of demographic or other sensitive attributes. The standard approach is to stratify the evaluation data across subgroups and compute performance metrics separately for each group. However, even for moderately-sized evaluation datasets, sample sizes quickly get small once considering intersectional subgroups, which greatly limits the extent to which intersectional groups are included in analysis. In this work, we introduce a structured regression approach to disaggregated evaluation that we demonstrate can yield reliable system performance estimates even for very small subgroups. We provide corresponding inference strategies for constructing confidence intervals and explore how goodness-of-fit testing can yield insight into the structure of fairness-related harms experienced by intersectional groups. We evaluate our approach on two publicly available datasets, and several variants of semi-synthetic data. The results show that our method is considerably more accurate than the standard approach, especially for small subgroups, and demonstrate how goodness-of-fit testing helps identify the key factors that drive differences in performance.