Research Papers

Representational harms in language technologies often occur in short spans within otherwise neutral text, where phrases may simultaneously convey generalizations, unfairness, or stereotypes. Framing bias detection as sentence-level classification obscures which words carry bias and what type is present, limiting both auditability and targeted mitigation. We introduce the GUS-Net Framework, comprising the GUS dataset and a multi-label token-level detector for span-level analysis of social bias. The GUS dataset contains 3,739 unique snippets across multiple domains, with over 69,000 token-level annotations. Each token is labeled using BIO tags (Begin, Inside, Outside) for three pathways of representational harm: Generalizations, Unfairness, and Stereotypes. To ensure reliable data annotation, we employ an automated multi-agent pipeline that proposes candidate spans which are subsequently verified and corrected by human experts. We formulate bias detection as multi-label token-level classification and benchmark both encoder-based models (e.g., BERT family variants) and decoder-based large language models (LLMs). Our evaluations cover token-level identification and span-level entity recognition on our test set, and out-of-distribution generalization. Empirical results show that encoder-based models consistently outperform decoder-based baselines on nuanced and overlapping spans while being more computationally efficient. The framework delivers interpretable, fine-grained diagnostics that enable systematic auditing and mitigation of representational harms in real-world NLP systems.

The rapid scaling of AI has spurred a growing emphasis on ethical considerations in both development and practice. This has led to the formulation of increasingly sophisticated model auditing and reporting requirements, as well as governance frameworks to mitigate potential risks to individuals and society. At this critical juncture, we review the practical challenges of promoting responsible AI and transparency in informal sectors like OSS that support vital infrastructure and see widespread use. We focus on how model performance evaluation may inform or inhibit probing of model limitations, biases, and other risks. Our controlled analysis of 7903 Hugging Face projects found that risk documentation is strongly associated with evaluation practices. Yet, submissions (N=789) from the platform's most popular competitive leaderboard showed less accountability among high performers. Our findings can inform AI providers and legal scholars in designing interventions and policies that preserve open-source innovation while incentivizing ethical uptake.

The Technology of Outrage: Bias in Artificial Intelligence

Will Bridewell, Paul F. Bello, Selmer Bringsjord

Artificial intelligence and machine learning are increasingly used to offload decision making from people. In the past, one of the rationales for this replacement was that machines, unlike people, can be fair and unbiased. Evidence suggests otherwise. We begin by entertaining the ideas that algorithms can replace people and that algorithms cannot be biased. Taken as axioms, these statements quickly lead to absurdity. Spurred on by this result, we investigate the slogans more closely and identify equivocation surrounding the word 'bias.' We diagnose three forms of outrage-intellectual, moral, and political-that are at play when people react emotionally to algorithmic bias. Then we suggest three practical approaches to addressing bias that the AI community could take, which include clarifying the language around bias, developing new auditing methods for intelligent systems, and building certain capabilities into these systems. We conclude by offering a moral regarding the conversations about algorithmic bias that may transfer to other areas of artificial intelligence.

Precision spraying evaluation requires automation primarily in post-spraying imagery. In this paper we propose an eXplainable Artificial Intelligence (XAI) computer vision pipeline to evaluate a precision spraying system post-spraying without the need for traditional agricultural methods. The developed system can semantically segment potential targets such as lettuce, chickweed, and meadowgrass and correctly identify if targets have been sprayed. Furthermore, this pipeline evaluates using a domain-specific Weakly Supervised Deposition Estimation task, allowing for class-specific quantification of spray deposit weights in μL. Estimation of coverage rates of spray deposition in a class-wise manner allows for further understanding of effectiveness of precision spraying systems. Our study evaluates different Class Activation Mapping techniques, namely AblationCAM and ScoreCAM, to determine which is more effective and interpretable for these tasks. In the pipeline, inference-only feature fusion is used to allow for further interpretability and to enable the automation of precision spraying evaluation post-spray. Our findings indicate that a Fully Convolutional Network with an EfficientNet-B0 backbone and inference-only feature fusion achieves an average absolute difference in deposition values of 156.8 μL across three classes in our test set. The dataset curated in this paper is publicly available at https://github.com/Harry-Rogers/PSIE

As large-scale AI systems proliferate, ensuring compliance with data privacy laws such as the General Data Protection Regulation (GDPR) has become critical. This paper introduces Brain Surgery, a transformative methodology for making every local AI model GDPR-ready by enabling real-time privacy management and targeted unlearning. Building on advanced techniques such as Embedding-Corrupted Prompts (ECO Prompts), blockchain-based privacy management, and privacy-aware continual learning, Brain Surgery provides a modular solution that can be deployed across various AI architectures. This tool not only ensures compliance with privacy regulations but also empowers users to define their own privacy limits, creating a new paradigm in AI ethics and governance.

International standards are crucial for ensuring that frontier AI systems are developed and deployed safely around the world. Since the AI Safety Institutes (AISIs) possess in-house technical expertise, mandate for international engagement, and convening power in the national AI ecosystem while being a government institution, we argue that they are particularly well-positioned to contribute to the international standard-setting processes for AI safety. In this paper, we propose and evaluate three models for AISI involvement: 1. Seoul Declaration Signatories, 2. US (and other Seoul Declaration Signatories) and China, and 3. Globally Inclusive. Leveraging their diverse strengths, these models are not mutually exclusive. Rather, they offer a multi-track system solution in which the central role of AISIs guarantees coherence among the different tracks and consistency in their AI safety focus.

We present TRACE-CS, a novel hybrid system that combines symbolic reasoning with large language models (LLMs)to address contrastive queries in course scheduling problems. TRACE-CS leverages logic-based techniques to encode scheduling constraints and generate provably correct explanations, while utilizing an LLM to process natural language queries and refine logical explanations into user friendly responses. This system showcases how combining symbolic KR methods with LLMs creates explainable AI agents that balance logical correctness with natural language accessibility, addressing a fundamental challenge in deployed scheduling systems.

In the field of eXplainable AI (XAI) in language models, the progression from local explanations of individual decisions to global explanations with high-level concepts has laid the groundwork for mechanistic interpretability, which aims to decode the exact operations. However, this paradigm has not been adequately explored in image models, where existing methods have primarily focused on class-specific interpretations. This paper introduces a novel approach to systematically trace the entire pathway from input through all intermediate layers to the final output within the whole dataset. We utilize Pointwise Feature Vectors (PFVs) and Effective Receptive Fields (ERFs) to decompose model embeddings into interpretable Concept Vectors. Then, we calculate the relevance between concept vectors with our Generalized Integrated Gradients (GIG), enabling a comprehensive, dataset-wide analysis of model behavior. We validate our method of concept extraction and concept attribution in both qualitative and quantitative evaluations. Our approach advances the understanding of semantic significance within image models, offering a holistic view of their operational mechanics.

Bringing more transparency to the judicial system for the purposes of increasing accountability often demands extensive effort from auditors who must meticulously sift through numerous disorganized legal case files to detect patterns of bias and errors. For example, the high-profile investigation into the Curtis Flowers case took seven reporters a full year to assemble evidence about the prosecutor's history of selecting racially biased juries. LLMs have the potential to automate and scale these transparency pipelines, especially given their demonstrated capabilities to extract information from unstructured documents. We discuss the opportunities and challenges of using LLMs to provide transparency in two important court processes: jury selection in criminal trials and housing eviction cases.

This pictorial aims to critically consider the nature of text-to-audio and text-to-music generative tools in the context of explainable AI. As a group of experimental musicians and researchers, we are enthusiastic about the creative potential of these tools and have sought to understand and evaluate them from perspectives of prompt creation, control, usability, understandability, explainability of the AI process, and overall aesthetic effectiveness of the results. One of the challenges we have identified that is not explicitly addressed by these tools is the inherent semantic gap in using text-based tools to describe something as abstract as music. Other gaps include explainability vs. useability, and user control and input vs. the human creative process. The aim of this pictorial is to raise questions for discussion and make a few general suggestions on the kinds of improvements we would like to see in generative AI music tools.

Complex sequential decision-making planning problems, covering infinite states' space have been shown to be solvable by AlphaZero type of algorithms. Such an approach that trains a neural model while simulating projection of futures with a Monte Carlo Tree Search algorithm were shown to be applicable to real life planning problems. As such, engineers and users interacting with the resulting policy of behavior might benefit from obtaining automated explanations about these planners' decisions offline or online. This paper focuses on the information within the Monte Carlo Tree Search data structure. Given its construction, this information contains much of the reasoning of the sequential decision-making algorithm and is essential for its explainability. We show novel methods using information theoretic tools for the simplification and reduction of the Monte Carlo Tree Search and the extraction of information. Such information can be directly used for the construction of human understandable explanations. We show that basic explainability quantities can be calculated with limited additional computational cost, as an integrated part of the Monte Carlo Tree Search construction process. We focus on the theoretical and algorithmic aspects and provide examples of how the methods presented here can be used in the construction of human understandable explanations.

Systematic literature reviews are the highest quality of evidence in research. However, the review process is hindered by significant resource and data constraints. The Literature Review Network (LRN) is the first of its kind explainable AI platform adhering to PRISMA 2020 standards, designed to automate the entire literature review process. LRN was evaluated in the domain of surgical glove practices using 3 search strings developed by experts to query PubMed. A non-expert trained all LRN models. Performance was benchmarked against an expert manual review. Explainability and performance metrics assessed LRN's ability to replicate the experts' review. Concordance was measured with the Jaccard index and confusion matrices. Researchers were blinded to the other's results until study completion. Overlapping studies were integrated into an LRN-generated systematic review. LRN models demonstrated superior classification accuracy without expert training, achieving 84.78% and 85.71% accuracy. The highest performance model achieved high interrater reliability (k = 0.4953) and explainability metrics, linking 'reduce', 'accident', and 'sharp' with 'double-gloving'. Another LRN model covered 91.51% of the relevant literature despite diverging from the non-expert's judgments (k = 0.2174), with the terms 'latex', 'double' (gloves), and 'indication'. LRN outperformed the manual review (19,920 minutes over 11 months), reducing the entire process to 288.6 minutes over 5 days. This study demonstrates that explainable AI does not require expert training to successfully conduct PRISMA-compliant systematic literature reviews like an expert. LRN summarized the results of surgical glove studies and identified themes that were nearly identical to the clinical researchers' findings. Explainable AI can accurately expedite our understanding of clinical practices, potentially revolutionizing healthcare research.