Research Papers

Artificial intelligence (AI) systems operate in increasingly diverse areas, from healthcare to facial recognition, the stock market, autonomous vehicles, and so on. While the underlying digital infrastructure of AI systems is developing rapidly, each area of implementation is subject to different degrees and processes of legitimization. By combining elements from institutional theory and information systems-theory, this paper presents a conceptual framework to analyze and understand AI-induced field-change. The introduction of novel AI-agents into new or existing fields creates a dynamic in which algorithms (re)shape organizations and institutions while existing institutional infrastructures determine the scope and speed at which organizational change is allowed to occur. Where institutional infrastructure and governance arrangements, such as standards, rules, and regulations, still are unelaborate, the field can move fast but is also more likely to be contested. The institutional infrastructure surrounding AI-induced fields is generally little elaborated, which could be an obstacle to the broader institutionalization of AI-systems going forward.

Among daily tasks of database administrators (DBAs), the analysis of query workloads to identify schema issues and improving performances is crucial. Although DBAs can easily pinpoint queries repeatedly causing performance issues, it remains challenging to automatically identify subsets of queries that share some properties only (a pattern) and simultaneously foster some target measures, such as execution time. Patterns are defined on combinations of query clauses, environment variables, database alerts and metrics and help answer questions like what makes SQL queries slow? What makes I/O communications high? Automatically discovering these patterns in a huge search space and providing them as hypotheses for helping to localize issues and root-causes is important in the context of explainable AI. To tackle it, we introduce an original approach rooted on Subgroup Discovery. We show how to instantiate and develop this generic data-mining framework to identify potential causes of SQL workloads issues. We believe that such data-mining technique is not trivial to apply for DBAs. As such, we also provide a visualization tool for interactive knowledge discovery. We analyse a one week workload from hundreds of databases from our company, make both the dataset and source code available, and experimentally show that insightful hypotheses can be discovered.

In today's society, AI systems are increasingly used to make critical decisions such as credit scoring and patient triage. However, great convenience brought by AI systems comes with troubling prevalence of bias against underrepresented groups. Mitigating bias in AI systems to increase overall fairness has emerged as an important challenge. Existing studies on mitigating bias in AI systems focus on eliminating sensitive demographic information embedded in data. Given the temporal and contextual complexity of conceptualizing fairness, lossy treatment of demographic information may contribute to an unnecessary trade-off between accuracy and fairness, especially when demographic attributes and class labels are correlated. In this paper, we present an information-lossless de-biasing technique that targets the scarcity of data in the disadvantaged group. Unlike the existing work, we demonstrate, both theoretically and empirically, that oversampling underrepresented groups can not only mitigate algorithmic bias in AI systems that consistently predict a favorable outcome for a certain group, but improve overall accuracy by mitigating class imbalance within data that leads to a bias towards the majority class. We demonstrate the effectiveness of our technique on real datasets using a variety of fairness metrics.

Altruist: Argumentative Explanations through Local Interpretations of Predictive Models

Ioannis Mollas, Nick Bassiliades, Grigorios Tsoumakas

Explainable AI is an emerging field providing solutions for acquiring insights into automated systems' rationale. It has been put on the AI map by suggesting ways to tackle key ethical and societal issues. Existing explanation techniques are often not comprehensible to the end user. Lack of evaluation and selection criteria also makes it difficult for the end user to choose the most suitable technique. In this study, we combine logic-based argumentation with Interpretable Machine Learning, introducing a preliminary meta-explanation methodology that identifies the truthful parts of feature importance oriented interpretations. This approach, in addition to being used as a meta-explanation technique, can be used as an evaluation or selection tool for multiple feature importance techniques. Experimentation strongly indicates that an ensemble of multiple interpretation techniques yields considerably more truthful explanations.

Trustworthiness is a central requirement for the acceptance and success of human-centered artificial intelligence (AI). To deem an AI system as trustworthy, it is crucial to assess its behaviour and characteristics against a gold standard of Trustworthy AI, consisting of guidelines, requirements, or only expectations. While AI systems are highly complex, their implementations are still based on software. The software engineering community has a long-established toolbox for the assessment of software systems, especially in the context of software testing. In this paper, we argue for the application of software engineering and testing practices for the assessment of trustworthy AI. We make the connection between the seven key requirements as defined by the European Commission's AI high-level expert group and established procedures from software engineering and raise questions for future work.

Modern face recognition systems leverage datasets containing images of hundreds of thousands of specific individuals' faces to train deep convolutional neural networks to learn an embedding space that maps an arbitrary individual's face to a vector representation of their identity. The performance of a face recognition system in face verification (1:1) and face identification (1:N) tasks is directly related to the ability of an embedding space to discriminate between identities. Recently, there has been significant public scrutiny into the source and privacy implications of large-scale face recognition training datasets such as MS-Celeb-1M and MegaFace, as many people are uncomfortable with their face being used to train dual-use technologies that can enable mass surveillance. However, the impact of an individual's inclusion in training data on a derived system's ability to recognize them has not previously been studied. In this work, we audit ArcFace, a state-of-the-art, open source face recognition system, in a large-scale face identification experiment with more than one million distractor images. We find a Rank-1 face identification accuracy of 79.71% for individuals present in the model's training data and an accuracy of 75.73% for those not present. This modest difference in accuracy demonstrates that face recognition systems using deep learning work better for individuals they are trained on, which has serious privacy implications when one considers all major open source face recognition training datasets do not obtain informed consent from individuals during their collection.

The integration of Large Language Models (LLMs) into the financial domain is driving a paradigm shift from passive information retrieval to dynamic, agentic interaction. While general-purpose tool learning has witnessed a surge in benchmarks, the financial sector, characterized by high stakes, strict compliance, and rapid data volatility, remains critically underserved. Existing financial evaluations predominantly focus on static textual analysis or document-based QA, ignoring the complex reality of tool execution. Conversely, general tool benchmarks lack the domain-specific rigor required for finance, often relying on toy environments or a negligible number of financial APIs. To bridge this gap, we introduce FinToolBench, the first real-world, runnable benchmark dedicated to evaluating financial tool learning agents. Unlike prior works limited to a handful of mock tools, FinToolBench establishes a realistic ecosystem coupling 760 executable financial tools with 295 rigorous, tool-required queries. We propose a novel evaluation framework that goes beyond binary execution success, assessing agents on finance-critical dimensions: timeliness, intent type, and regulatory domain alignment. Furthermore, we present FATR, a finance-aware tool retrieval and reasoning baseline that enhances stability and compliance. By providing the first testbed for auditable, agentic financial execution, FinToolBench sets a new standard for trustworthy AI in finance. The tool manifest, execution environment, and evaluation code will be open-sourced to facilitate future research.

VLSU: Mapping the Limits of Joint Multimodal Understanding for AI Safety

Shruti Palaskar, Leon Gatys, Mona Abdelrahman +9

Safety evaluation of multimodal foundation models often treats vision and language inputs separately, missing risks from joint interpretation where benign content becomes harmful in combination. Existing approaches also fail to distinguish clearly unsafe content from borderline cases, leading to problematic over-blocking or under-refusal of genuinely harmful content. We present Vision Language Safety Understanding (VLSU), a comprehensive framework to systematically evaluate multimodal safety through fine-grained severity classification and combinatorial analysis across 17 distinct safety patterns. Using a multi-stage pipeline with real-world images and human annotation, we construct a large-scale benchmark of 8,187 samples spanning 15 harm categories. Our evaluation of eleven state-of-the-art models reveals systematic joint understanding failures: while models achieve 90%-plus accuracy on clear unimodal safety signals, performance degrades substantially to 20-55% when joint image-text reasoning is required to determine the safety label. Most critically, 34% of errors in joint image-text safety classification occur despite correct classification of the individual modalities, further demonstrating absent compositional reasoning capabilities. Additionally, we find that models struggle to balance refusing unsafe content while still responding to borderline cases that deserve engagement. For example, we find that instruction framing can reduce the over-blocking rate on borderline content from 62.4% to 10.4% in Gemini-1.5, but only at the cost of under-refusing on unsafe content with refusal rate dropping from 90.8% to 53.9%. Overall, our framework exposes weaknesses in joint image-text understanding and alignment gaps in current models, and provides a critical test bed to enable the next milestones in research on robust vision-language safety.

Tree-based Dialogue Reinforced Policy Optimization for Red-Teaming Attacks

Ruohao Guo, Afshin Oroojlooy, Roshan Sridhar +3

Despite recent rapid progress in AI safety, current large language models remain vulnerable to adversarial attacks in multi-turn interaction settings, where attackers strategically adapt their prompts across conversation turns and pose a more critical yet realistic challenge. Existing approaches that discover safety vulnerabilities either rely on manual red-teaming with human experts or employ automated methods using pre-defined templates and human-curated attack data, with most focusing on single-turn attacks. However, these methods did not explore the vast space of possible multi-turn attacks, failing to consider novel attack trajectories that emerge from complex dialogue dynamics and strategic conversation planning. This gap is particularly critical given recent findings that LLMs exhibit significantly higher vulnerability to multi-turn attacks compared to single-turn attacks. We propose DialTree, an on-policy reinforcement learning framework integrated with tree search that autonomously discovers diverse multi-turn attack strategies by treating the dialogue as a sequential decision-making problem, enabling systematic exploration without manually curated data. Through extensive experiments, our approach not only achieves more than 44.2% higher ASR across 12 target models compared to previous state-of-the-art approaches, but also effectively uncovers new attack strategies by learning optimal dialogue policies that maximize attack success across multiple turns.

Recent advancements in machine learning have spurred growing interests in automated interpreting quality assessment. Nevertheless, existing research suffers from insufficient examination of language use quality, unsatisfactory modeling effectiveness due to data scarcity and imbalance, and a lack of efforts to explain model predictions. To address these gaps, we propose a multi-dimensional modeling framework that integrates feature engineering, data augmentation, and explainable machine learning. This approach prioritizes explainability over ``black box'' predictions by utilizing only construct-relevant, transparent features and conducting Shapley Value (SHAP) analysis. Our results demonstrate strong predictive performance on a novel English-Chinese consecutive interpreting dataset, identifying BLEURT and CometKiwi scores to be the strongest predictive features for fidelity, pause-related features for fluency, and Chinese-specific phraseological diversity metrics for language use. Overall, by placing particular emphasis on explainability, we present a scalable, reliable, and transparent alternative to traditional human evaluation, facilitating the provision of detailed diagnostic feedback for learners and supporting self-regulated learning advantages not afforded by automated scores in isolation.

AI Agents and the Law

Mark O. Riedl, Deven R. Desai

As AI becomes more "agentic," it faces technical and socio-legal issues it must address if it is to fulfill its promise of increased economic productivity and efficiency. This paper uses technical and legal perspectives to explain how things change when AI systems start being able to directly execute tasks on behalf of a user. We show how technical conceptions of agents track some, but not all, socio-legal conceptions of agency. That is, both computer science and the law recognize the problems of under-specification for an agent, and both disciplines have robust conceptions of how to address ensuring an agent does what the programmer, or in the law, the principal desires and no more. However, to date, computer science has under-theorized issues related to questions of loyalty and to third parties that interact with an agent, both of which are central parts of the law of agency. First, we examine the correlations between implied authority in agency law and the principle of value-alignment in AI, wherein AI systems must operate under imperfect objective specification. Second, we reveal gaps in the current computer science view of agents pertaining to the legal concepts of disclosure and loyalty, and how failure to account for them can result in unintended effects in AI ecommerce agents. In surfacing these gaps, we show a path forward for responsible AI agent development and deployment.

Monitoring AIs at runtime can help us detect and stop harmful actions. In this paper, we study how to efficiently combine multiple runtime monitors into a single monitoring protocol. The protocol's objective is to maximize the probability of applying a safety intervention on misaligned outputs (i.e., maximize recall). Since running monitors and applying safety interventions are costly, the protocol also needs to adhere to an average-case budget constraint. Taking the monitors' performance and cost as given, we develop an algorithm to find the best protocol. The algorithm exhaustively searches over when and which monitors to call, and allocates safety interventions based on the Neyman-Pearson lemma. By focusing on likelihood ratios and strategically trading off spending on monitors against spending on interventions, we more than double our recall rate compared to a naive baseline in a code review setting. We also show that combining two monitors can Pareto dominate using either monitor alone. Our framework provides a principled methodology for combining existing monitors to detect undesirable behavior in cost-sensitive settings.