Research Papers

When we consult with a doctor, lawyer, or financial advisor, we generally assume that they are acting in our best interests. But what should we assume when it is an artificial intelligence (AI) system that is acting on our behalf? Early examples of AI assistants like Alexa, Siri, Google, and Cortana already serve as a key interface between consumers and information on the web, and users routinely rely upon AI-driven systems like these to take automated actions or provide information. Superficially, such systems may appear to be acting according to user interests. However, many AI systems are designed with embedded conflicts of interests, acting in ways that subtly benefit their creators (or funders) at the expense of users. To address this problem, in this paper we introduce the concept of AI loyalty. AI systems are loyal to the degree that they are designed to minimize, and make transparent, conflicts of interest, and to act in ways that prioritize the interests of users. Properly designed, such systems could have considerable functional and competitive - not to mention ethical - advantages relative to those that do not. Loyal AI products hold an obvious appeal for the end-user and could serve to promote the alignment of the long-term interests of AI developers and customers. To this end, we suggest criteria for assessing whether an AI system is sufficiently transparent about conflicts of interest, and acting in a manner that is loyal to the user, and argue that AI loyalty should be considered during the technological design process alongside other important values in AI ethics such as fairness, accountability privacy, and equity. We discuss a range of mechanisms, from pure market forces to strong regulatory frameworks, that could support incorporation of AI loyalty into a variety of future AI systems.

As freelancing work keeps on growing almost everywhere due to a sharp decrease in communication costs and to the widespread of Internet-based labour marketplaces (e.g., guru.com, feelancer.com, mturk.com, upwork.com), many researchers and practitioners have started exploring the benefits of outsourcing and crowdsourcing. Since employers often use these platforms to find a group of workers to complete a specific task, researchers have focused their efforts on the study of team formation and matching algorithms and on the design of effective incentive schemes. Nevertheless, just recently, several concerns have been raised on possibly unfair biases introduced through the algorithms used to carry out these selection and matching procedures. For this reason, researchers have started studying the fairness of algorithms related to these online marketplaces, looking for intelligent ways to overcome the algorithmic bias that frequently arises. Broadly speaking, the aim is to guarantee that, for example, the process of hiring workers through the use of machine learning and algorithmic data analysis tools does not discriminate, even unintentionally, on grounds of nationality or gender. In this short paper, we define the Fair Team Formation problem in the following way: given an online labour marketplace where each worker possesses one or more skills, and where all workers are divided into two or more not overlapping classes (for examples, men and women), we want to design an algorithm that is able to find a team with all the skills needed to complete a given task, and that has the same number of people from all classes. We provide inapproximability results for the Fair Team Formation problem together with four algorithms for the problem itself. We also tested the effectiveness of our algorithmic solutions by performing experiments using real data from an online labor marketplace.

Modern AI image classifiers have made impressive advances in recent years, but their performance often appears strange or violates expectations of users. This suggests humans engage in cognitive anthropomorphism: expecting AI to have the same nature as human intelligence. This mismatch presents an obstacle to appropriate human-AI interaction. To delineate this mismatch, I examine known properties of human classification, in comparison to image classifier systems. Based on this examination, I offer three strategies for system design that can address the mismatch between human and AI classification: explainable AI, novel methods for training users, and new algorithms that match human cognition.

Always Lurking: Understanding and Mitigating Bias in Online Human Trafficking Detection

Kyle Hundman, Thamme Gowda, Mayank Kejriwal +1

Web-based human trafficking activity has increased in recent years but it remains sparsely dispersed among escort advertisements and difficult to identify due to its often-latent nature. The use of intelligent systems to detect trafficking can thus have a direct impact on investigative resource allocation and decision-making, and, more broadly, help curb a widespread social problem. Trafficking detection involves assigning a normalized score to a set of escort advertisements crawled from the Web -- a higher score indicates a greater risk of trafficking-related (involuntary) activities. In this paper, we define and study the problem of trafficking detection and present a trafficking detection pipeline architecture developed over three years of research within the DARPA Memex program. Drawing on multi-institutional data, systems, and experiences collected during this time, we also conduct post hoc bias analyses and present a bias mitigation plan. Our findings show that, while automatic trafficking detection is an important application of AI for social good, it also provides cautionary lessons for deploying predictive machine learning algorithms without appropriate de-biasing. This ultimately led to integration of an interpretable solution into a search system that contains over 100 million advertisements and is used by over 200 law enforcement agencies to investigate leads.

Artificial Intelligence (AI) is an effective science which employs strong enough approaches, methods, and techniques to solve unsolvable real world based problems. Because of its unstoppable rise towards the future, there are also some discussions about its ethics and safety. Shaping an AI friendly environment for people and a people friendly environment for AI can be a possible answer for finding a shared context of values for both humans and robots. In this context, objective of this paper is to address the ethical issues of AI and explore the moral dilemmas that arise from ethical algorithms, from pre set or acquired values. In addition, the paper will also focus on the subject of AI safety. As general, the paper will briefly analyze the concerns and potential solutions to solving the ethical issues presented and increase readers awareness on AI safety as another related research interest.

Skyra: AI-Generated Video Detection via Grounded Artifact Reasoning

Yifei Li, Wenzhao Zheng, Yanran Zhang +5

The misuse of AI-driven video generation technologies has raised serious social concerns, highlighting the urgent need for reliable AI-generated video detectors. However, most existing methods are limited to binary classification and lack the necessary explanations for human interpretation. In this paper, we present Skyra, a specialized multimodal large language model (MLLM) that identifies human-perceivable visual artifacts in AI-generated videos and leverages them as grounded evidence for both detection and explanation. To support this objective, we construct ViF-CoT-4K for Supervised Fine-Tuning (SFT), which represents the first large-scale AI-generated video artifact dataset with fine-grained human annotations. We then develop a two-stage training strategy that systematically enhances our model's spatio-temporal artifact perception, explanation capability, and detection accuracy. To comprehensively evaluate Skyra, we introduce ViF-Bench, a benchmark comprising 3K high-quality samples generated by over ten state-of-the-art video generators. Extensive experiments demonstrate that Skyra surpasses existing methods across multiple benchmarks, while our evaluation yields valuable insights for advancing explainable AI-generated video detection.

The development of Long-CoT reasoning has advanced LLM performance across various tasks, including language understanding, complex problem solving, and code generation. This paradigm enables models to generate intermediate reasoning steps, thereby improving both accuracy and interpretability. However, despite these advancements, a comprehensive understanding of how CoT-based reasoning affects the trustworthiness of language models remains underdeveloped. In this paper, we survey recent work on reasoning models and CoT techniques, focusing on five core dimensions of trustworthy reasoning: truthfulness, safety, robustness, fairness, and privacy. For each aspect, we provide a clear and structured overview of recent studies in chronological order, along with detailed analyses of their methodologies, findings, and limitations. Future research directions are also appended at the end for reference and discussion. Overall, while reasoning techniques hold promise for enhancing model trustworthiness through hallucination mitigation, harmful content detection, and robustness improvement, cutting-edge reasoning models themselves often suffer from comparable or even greater vulnerabilities in safety, robustness, and privacy. By synthesizing these insights, we hope this work serves as a valuable and timely resource for the AI safety community to stay informed on the latest progress in reasoning trustworthiness. A full list of related papers can be found at \href{https://github.com/ybwang119/Awesome-reasoning-safety}{https://github.com/ybwang119/Awesome-reasoning-safety}.

Addressing contextual privacy concerns remains challenging in interactive settings where large language models (LLMs) process information from multiple sources (e.g., summarizing meetings with private and public information). We introduce a multi-agent framework that decomposes privacy reasoning into specialized subtasks (extraction, classification), reducing the information load on any single agent while enabling iterative validation and more reliable adherence to contextual privacy norms. To understand how privacy errors emerge and propagate, we conduct a systematic ablation over information-flow topologies, revealing when and why upstream detection mistakes cascade into downstream leakage. Experiments on the ConfAIde and PrivacyLens benchmark with several open-source and closed-sourced LLMs demonstrate that our best multi-agent configuration substantially reduces private information leakage (\textbf{18\%} on ConfAIde and \textbf{19\%} on PrivacyLens with GPT-4o) while preserving the fidelity of public content, outperforming single-agent baselines. These results highlight the promise of principled information-flow design in multi-agent systems for contextual privacy with LLMs.

In line with the AI-native 6G vision, explainability and robustness are crucial for building trust and ensuring reliable performance in millimeter-wave (mmWave) systems. Efficient beam alignment is essential for initial access, but deep learning (DL) solutions face challenges, including high data collection overhead, hardware constraints, lack of explainability, and susceptibility to adversarial attacks. This paper proposes a robust and explainable DL-based beam alignment engine (BAE) for mmWave multiple-input multiple output (MIMO) systems. The BAE uses received signal strength indicator (RSSI) measurements from wide beams to predict the best narrow beam, reducing the overhead of exhaustive beam sweeping. To overcome the challenge of real-world data collection, this work leverages a site-specific digital twin (DT) to generate synthetic channel data closely resembling real-world environments. A model refinement via transfer learning is proposed to fine-tune the pre-trained model residing in the DT with minimal real-world data, effectively bridging mismatches between the digital replica and real-world environments. To reduce beam training overhead and enhance transparency, the framework uses deep Shapley additive explanations (SHAP) to rank input features by importance, prioritizing key spatial directions and minimizing beam sweeping. It also incorporates the Deep k-nearest neighbors (DkNN) algorithm, providing a credibility metric for detecting out-of-distribution inputs and ensuring robust, transparent decision-making. Experimental results show that the proposed framework reduces real-world data needs by 70%, beam training overhead by 62%, and improves outlier detection robustness by up to 8.5x, achieving near-optimal spectral efficiency and transparent decision making compared to traditional softmax based DL models.

As robots and digital assistants are deployed in the real world, these agents must be able to communicate their decision-making criteria to build trust, improve human-robot teaming, and enable collaboration. While the field of explainable artificial intelligence (xAI) has made great strides to enable such communication, these advances often assume that one xAI approach is ideally suited to each problem (e.g., decision trees to explain how to triage patients in an emergency or feature-importance maps to explain radiology reports). This fails to recognize that users have diverse experiences or preferences for interaction modalities. In this work, we present two user-studies set in a simulated autonomous vehicle (AV) domain. We investigate (1) population-level preferences for xAI and (2) personalization strategies for providing robot explanations. We find significant differences between xAI modes (language explanations, feature-importance maps, and decision trees) in both preference (p < 0.01) and performance (p < 0.05). We also observe that a participant's preferences do not always align with their performance, motivating our development of an adaptive personalization strategy to balance the two. We show that this strategy yields significant performance gains (p < 0.05), and we conclude with a discussion of our findings and implications for xAI in human-robot interactions.

We present our Balanced, Integrated and Grounded (BIG) argument for assuring the safety of AI systems. The BIG argument adopts a whole-system approach to constructing a safety case for AI systems of varying capability, autonomy and criticality. Firstly, it is balanced by addressing safety alongside other critical ethical issues such as privacy and equity, acknowledging complexities and trade-offs in the broader societal impact of AI. Secondly, it is integrated by bringing together the social, ethical and technical aspects of safety assurance in a way that is traceable and accountable. Thirdly, it is grounded in long-established safety norms and practices, such as being sensitive to context and maintaining risk proportionality. Whether the AI capability is narrow and constrained or general-purpose and powered by a frontier or foundational model, the BIG argument insists on a systematic treatment of safety. Further, it places a particular focus on the novel hazardous behaviours emerging from the advanced capabilities of frontier AI models and the open contexts in which they are rapidly being deployed. These complex issues are considered within a wider AI safety case, approaching assurance from both technical and sociotechnical perspectives. Examples illustrating the use of the BIG argument are provided throughout the paper.

Practitioners and researchers trying to strike a balance between accuracy and transparency center Explainable Artificial Intelligence (XAI) at the junction of finance. This paper offers a thorough overview of the changing scene of XAI applications in finance together with domain-specific implementations, methodological developments, and trend mapping of research. Using bibliometric and content analysis, we find topic clusters, significant research, and most often used explainability strategies used in financial industries. Our results show a substantial dependence on post-hoc interpretability techniques; attention mechanisms, feature importance analysis and SHAP are the most often used techniques among them. This review stresses the need of multidisciplinary approaches combining financial knowledge with improved explainability paradigms and exposes important shortcomings in present XAI systems.