Research Papers

With the powerful performance of Artificial Intelligence (AI) also comes prevalent ethical issues. Though governments and corporations have curated multiple AI ethics guidelines to curb unethical behavior of AI, the effect has been limited, probably due to the vagueness of the guidelines. In this paper, we take a closer look at how AI ethics issues take place in real world, in order to have a more in-depth and nuanced understanding of different ethical issues as well as their social impact. With a content analysis of AI Incident Database, which is an effort to prevent repeated real world AI failures by cataloging incidents, we identified 13 application areas which often see unethical use of AI, with intelligent service robots, language/vision models and autonomous driving taking the lead. Ethical issues appear in 8 different forms, from inappropriate use and racial discrimination, to physical safety and unfair algorithm. With this taxonomy of AI ethics issues, we aim to provide AI practitioners with a practical guideline when trying to deploy AI applications ethically.

This chapter formulates seven lessons for preventing harm in artificial intelligence (AI) systems based on insights from the field of system safety for software-based automation in safety-critical domains. New applications of AI across societal domains and public organizations and infrastructures come with new hazards, which lead to new forms of harm, both grave and pernicious. The text addresses the lack of consensus for diagnosing and eliminating new AI system hazards. For decades, the field of system safety has dealt with accidents and harm in safety-critical systems governed by varying degrees of software-based automation and decision-making. This field embraces the core assumption of systems and control that AI systems cannot be safeguarded by technical design choices on the model or algorithm alone, instead requiring an end-to-end hazard analysis and design frame that includes the context of use, impacted stakeholders and the formal and informal institutional environment in which the system operates. Safety and other values are then inherently socio-technical and emergent system properties that require design and control measures to instantiate these across the technical, social and institutional components of a system. This chapter honors system safety pioneer Nancy Leveson, by situating her core lessons for today's AI system safety challenges. For every lesson, concrete tools are offered for rethinking and reorganizing the safety management of AI systems, both in design and governance. This history tells us that effective AI safety management requires transdisciplinary approaches and a shared language that allows involvement of all levels of society.

Order in the Court: Explainable AI Methods Prone to Disagreement

Michael Neely, Stefan F. Schouten, Maurits J. R. Bleeker +1

By computing the rank correlation between attention weights and feature-additive explanation methods, previous analyses either invalidate or support the role of attention-based explanations as a faithful and plausible measure of salience. To investigate whether this approach is appropriate, we compare LIME, Integrated Gradients, DeepLIFT, Grad-SHAP, Deep-SHAP, and attention-based explanations, applied to two neural architectures trained on single- and pair-sequence language tasks. In most cases, we find that none of our chosen methods agree. Based on our empirical observations and theoretical objections, we conclude that rank correlation does not measure the quality of feature-additive methods. Practitioners should instead use the numerous and rigorous diagnostic methods proposed by the community.

Combating fake news and misinformation propagation is a challenging task in the post-truth era. News feed and search algorithms could potentially lead to unintentional large-scale propagation of false and fabricated information with users being exposed to algorithmically selected false content. Our research investigates the effects of an Explainable AI assistant embedded in news review platforms for combating the propagation of fake news. We design a news reviewing and sharing interface, create a dataset of news stories, and train four interpretable fake news detection algorithms to study the effects of algorithmic transparency on end-users. We present evaluation results and analysis from multiple controlled crowdsourced studies. For a deeper understanding of Explainable AI systems, we discuss interactions between user engagement, mental model, trust, and performance measures in the process of explaining. The study results indicate that explanations helped participants to build appropriate mental models of the intelligent assistants in different conditions and adjust their trust accordingly for model limitations.

How to Democratise and Protect AI: Fair and Differentially Private Decentralised Deep Learning

Lingjuan Lyu, Yitong Li, Karthik Nandakumar +2

This paper firstly considers the research problem of fairness in collaborative deep learning, while ensuring privacy. A novel reputation system is proposed through digital tokens and local credibility to ensure fairness, in combination with differential privacy to guarantee privacy. In particular, we build a fair and differentially private decentralised deep learning framework called FDPDDL, which enables parties to derive more accurate local models in a fair and private manner by using our developed two-stage scheme: during the initialisation stage, artificial samples generated by Differentially Private Generative Adversarial Network (DPGAN) are used to mutually benchmark the local credibility of each party and generate initial tokens; during the update stage, Differentially Private SGD (DPSGD) is used to facilitate collaborative privacy-preserving deep learning, and local credibility and tokens of each party are updated according to the quality and quantity of individually released gradients. Experimental results on benchmark datasets under three realistic settings demonstrate that FDPDDL achieves high fairness, yields comparable accuracy to the centralised and distributed frameworks, and delivers better accuracy than the standalone framework.

In this paper, we cast fair machine learning as invariant machine learning. We first formulate a version of individual fairness that enforces invariance on certain sensitive sets. We then design a transport-based regularizer that enforces this version of individual fairness and develop an algorithm to minimize the regularizer efficiently. Our theoretical results guarantee the proposed approach trains certifiably fair ML models. Finally, in the experimental studies we demonstrate improved fairness metrics in comparison to several recent fair training procedures on three ML tasks that are susceptible to algorithmic bias.

Deep Verifier Networks: Verification of Deep Discriminative Models with Deep Generative Models

Tong Che, Xiaofeng Liu, Site Li +4

AI Safety is a major concern in many deep learning applications such as autonomous driving. Given a trained deep learning model, an important natural problem is how to reliably verify the model's prediction. In this paper, we propose a novel framework -- deep verifier networks (DVN) to verify the inputs and outputs of deep discriminative models with deep generative models. Our proposed model is based on conditional variational auto-encoders with disentanglement constraints. We give both intuitive and theoretical justifications of the model. Our verifier network is trained independently with the prediction model, which eliminates the need of retraining the verifier network for a new model. We test the verifier network on out-of-distribution detection and adversarial example detection problems, as well as anomaly detection problems in structured prediction tasks such as image caption generation. We achieve state-of-the-art results in all of these problems.

Agentic AI systems powered by large language models (LLMs) and endowed with planning, tool use, memory, and autonomy, are emerging as powerful, flexible platforms for automation. Their ability to autonomously execute tasks across web, software, and physical environments creates new and amplified security risks, distinct from both traditional AI safety and conventional software security. This survey outlines a taxonomy of threats specific to agentic AI, reviews recent benchmarks and evaluation methodologies, and discusses defense strategies from both technical and governance perspectives. We synthesize current research and highlight open challenges, aiming to support the development of secure-by-design agent systems.

BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models

Yige Li, Hanxun Huang, Yunhan Zhao +2

Generative large language models (LLMs) have achieved state-of-the-art results on a wide range of tasks, yet they remain susceptible to backdoor attacks: carefully crafted triggers in the input can manipulate the model to produce adversary-specified outputs. While prior research has predominantly focused on backdoor risks in vision and classification settings, the vulnerability of LLMs in open-ended text generation remains underexplored. To fill this gap, we introduce BackdoorLLM (Our BackdoorLLM benchmark was awarded First Prize in the SafetyBench competition, https://www.mlsafety.org/safebench/winners, organized by the Center for AI Safety, https://safe.ai/.), the first comprehensive benchmark for systematically evaluating backdoor threats in text-generation LLMs. BackdoorLLM provides: (i) a unified repository of benchmarks with a standardized training and evaluation pipeline; (ii) a diverse suite of attack modalities, including data poisoning, weight poisoning, hidden-state manipulation, and chain-of-thought hijacking; (iii) over 200 experiments spanning 8 distinct attack strategies, 7 real-world scenarios, and 6 model architectures; (iv) key insights into the factors that govern backdoor effectiveness and failure modes in LLMs; and (v) a defense toolkit encompassing 7 representative mitigation techniques. Our code and datasets are available at https://github.com/bboylyg/BackdoorLLM. We will continuously incorporate emerging attack and defense methodologies to support the research in advancing the safety and reliability of LLMs.

The success of Graph Neural Networks (GNNs) has led to a need for understanding their decision-making process and providing explanations for their predictions, which has given rise to explainable AI (XAI) that offers transparent explanations for black-box models. Recently, the use of prototypes has successfully improved the explainability of models by learning prototypes to imply training graphs that affect the prediction. However, these approaches tend to provide prototypes with excessive information from the entire graph, leading to the exclusion of key substructures or the inclusion of irrelevant substructures, which can limit both the interpretability and the performance of the model in downstream tasks. In this work, we propose a novel framework of explainable GNNs, called interpretable Prototype-based Graph Information Bottleneck (PGIB) that incorporates prototype learning within the information bottleneck framework to provide prototypes with the key subgraph from the input graph that is important for the model prediction. This is the first work that incorporates prototype learning into the process of identifying the key subgraphs that have a critical impact on the prediction performance. Extensive experiments, including qualitative analysis, demonstrate that PGIB outperforms state-of-the-art methods in terms of both prediction performance and explainability.

Towards Building AI-CPS with NVIDIA Isaac Sim: An Industrial Benchmark and Case Study for Robotics Manipulation

Zhehua Zhou, Jiayang Song, Xuan Xie +5

As a representative cyber-physical system (CPS), robotic manipulator has been widely adopted in various academic research and industrial processes, indicating its potential to act as a universal interface between the cyber and the physical worlds. Recent studies in robotics manipulation have started employing artificial intelligence (AI) approaches as controllers to achieve better adaptability and performance. However, the inherent challenge of explaining AI components introduces uncertainty and unreliability to these AI-enabled robotics systems, necessitating a reliable development platform for system design and performance assessment. As a foundational step towards building reliable AI-enabled robotics systems, we propose a public industrial benchmark for robotics manipulation in this paper. It leverages NVIDIA Omniverse Isaac Sim as the simulation platform, encompassing eight representative manipulation tasks and multiple AI software controllers. An extensive evaluation is conducted to analyze the performance of AI controllers in solving robotics manipulation tasks, enabling a thorough understanding of their effectiveness. To further demonstrate the applicability of our benchmark, we develop a falsification framework that is compatible with physical simulators and OpenAI Gym environments. This framework bridges the gap between traditional testing methods and modern physics engine-based simulations. The effectiveness of different optimization methods in falsifying AI-enabled robotics manipulation with physical simulators is examined via a falsification test. Our work not only establishes a foundation for the design and development of AI-enabled robotics systems but also provides practical experience and guidance to practitioners in this field, promoting further research in this critical academic and industrial domain.

Software Engineering for Responsible AI: An Empirical Study and Operationalised Patterns

Qinghua Lu, Liming Zhu, Xiwei Xu +3

Although artificial intelligence (AI) is solving real-world challenges and transforming industries, there are serious concerns about its ability to behave and make decisions in a responsible way. Many AI ethics principles and guidelines for responsible AI have been recently issued by governments, organisations, and enterprises. However, these AI ethics principles and guidelines are typically high-level and do not provide concrete guidance on how to design and develop responsible AI systems. To address this shortcoming, we first present an empirical study where we interviewed 21 scientists and engineers to understand the practitioners' perceptions on AI ethics principles and their implementation. We then propose a template that enables AI ethics principles to be operationalised in the form of concrete patterns and suggest a list of patterns using the newly created template. These patterns provide concrete, operationalised guidance that facilitate the development of responsible AI systems.