Research Papers

When AI systems surpass elite human performance and settle into everyday expert practice, the question that follows is how machine judgment is made publicly intelligible and attributable. We study Korean Go commentary on YouTube, where AI systems such as KataGo became standard analytic tools after AlphaGo. Our corpus spans a decade (2016--2025) and approximately $1{,}900$ hours of footage across institutional broadcasters and creator-led channels, in four phases of AI availability. We document a widening asymmetry between visual and verbal AI presence: AI winrate graphs are visible for about $98\%$ of late-period institutional broadcast time, yet AI-salient talk accounts for only $2.63\%$ of sentences. What recedes is the source label, not the metric: winrate and point-gap talk persists while ``AI'' itself goes unsaid. We read this recession as the communicative signature of domestication. Our strongest evidence is a compositional shift in verbal mediation: explicit naming gives way to interface rendering, and creator-led commentary leans further toward it than institutional commentary. We develop a typology distinguishing source-foregrounding from source-receding mediation, and argue that the two preserve different hooks of contestability: discursive anchors through which audiences can recognize and question the machine source. The stakes of that difference rise in domains where AI is less reliable than in Go.

Context: AI-assisted software development can speed up coding and review, but it also makes accountability harder to establish. Developers may submit insufficiently verified code, reviewers may approve changes with limited inspection, and centralized reputation records may be difficult to audit. Objectives: This study introduces TrustChain-Review, a framework that combines verifiable evidence, strategic incentives, and risk-sensitive governance to support more trustworthy code review. Methods: The framework includes a blockchain-based evidence layer, a three-player game-theoretic model for developers, reviewers, and the platform, and a rule that applies stronger governance when the expected benefit justifies its cost. The evaluation uses a controlled simulation calibrated with the Diff Quality Estimation dataset. Six governance configurations are compared over 30 independent runs using reputation accuracy, trust convergence, malicious-review detection, superficial-review detection, net platform utility, governance cost, and cost-efficiency. Results: The full-evidence configuration produces the strongest reputation, trust, and detection results, but it also has the highest governance cost. The risk-adaptive configuration lowers this cost and improves cost-efficiency by applying stronger controls selectively, although its trust and detection results are lower than those of the full-evidence setting. Conclusion: Strong evidence-based governance is most appropriate for high-risk or audit-sensitive changes. For routine or lower-risk contributions, selective governance offers a more practical balance between trustworthiness and operational cost.

This paper presents a multimodal machine-learning framework for calibration monitoring, quality assessment, and adaptive acquisition support in archaeological digitisation workflows. The proposed approach operates across photogrammetric 3D reconstruction, hyperspectral imaging, X-ray fluorescence spectroscopy, and Raman spectroscopy through a unified pipeline combining deterministic quality indicators, statistical feature representations, machine-learning classification, anomaly detection, and explainable artificial intelligence (XAI). Rather than replacing instrument-level calibration, the framework introduces an additional algorithmic layer that evaluates whether acquisitions are statistically consistent, physically plausible, and suitable for downstream multimodal integration. For each sensing modality, acquisitions are represented through structured feature spaces encoding geometric, spectral, spatial, and statistical properties. These representations are used to identify degradation patterns such as reconstruction artefacts, illumination inconsistencies, spectral distortions, detector instability, baseline fluctuations, and low signal-to-noise conditions. Supervised and unsupervised learning methods are combined with XAI techniques to support both automatic discrimination between acceptable and problematic acquisitions and interpretation of the underlying causes of degradation. The framework additionally supports adaptive feedback and resource-aware acquisition strategies by linking feature-space deviations to acquisition-level corrective actions. Experimental results obtained on multimodal archaeological datasets demonstrate that the proposed methodology captures meaningful acquisition variability and enables robust quality assessment across heterogeneous sensing modalities.

Auditing vision-language models (VLMs) for societal bias requires distinguishing direct algorithmic valuation disparities from confounders embedded within archival metadata. In this study, we audit Contrastive Language-Image Pretraining (CLIP) models using historical artwork metadata from the Metropolitan Museum of Art Open Access collection (N = 1,500 total objects; N = 743 attributed works: Male n = 534, Female n = 209; n = 618 anonymous). We establish a quantitative audit framework evaluating zero-shot CLIP logit differential scores across three semantic prompt pairs (masterpiece, quality, and influence). Unadjusted evaluations demonstrate high score convergence without a statistically significant main gender effect under OpenAI CLIP (mu_F = -0.0067 vs mu_M = -0.0035, p = 0.1829) or OpenCLIP (mu_F = 0.0171 vs mu_M = 0.0237, p = 0.1224). Two One-Sided Tests (TOST) confirm statistical equivalence across Cohen's d >= 0.25 bounds (pTOST < 0.005). Multivariate OLS regression controlling for artwork medium, creation era, and aspect ratio (R^2 < 0.02) confirms that artist gender has no statistically significant conditional effect (p > 0.20). High residual embedding variance (R^2 < 2%) indicates that global zero-shot valuation metrics operate near an embedding noise floor, showing that broad zero-shot prompt logit differentials are a coarse measurement instrument rather than proving absolute model fairness. We highlight two key caveats: (i) macro-level score equivalence reflects metric insensitivity to fine-grained visual-semantic features and does not preclude localized micro-level visual biases, and (ii) excluding 41.2% unattributed holdings reflects institutional survival bias. These results demonstrate the necessity of multivariate confound control, equivalence testing, and archival provenance auditing when assessing AI fairness in cultural heritage collections.

Public services face growing pressure to adopt artificial intelligence (AI) to close the gap between rising demand and falling resources. That pressure has intensified with general-purpose AI (GPAI): AI built on large language models that can be directed by prompt alone to perform an effectively unbounded range of tasks. We argue that the properties that make these models attractive - their generality, accessibility, and low deployment cost - undermine the conditions under which AI safety has historically been pursued. The safety concepts that public service governance frameworks foreground - accuracy, bias, explainability, and accountability - were made tractable by narrow, purpose-built AI, and the mitigations that guidance documents prescribe presuppose exactly what GPAI removes. Accuracy cannot be quantified over unbounded outputs. Bias cannot be disaggregated when outputs are free-text judgements rather than categorical predictions. Explainability gives way to the appearance of explanation, and accountability erodes as outputs are optimized to persuade. We develop this through the case of policing, where the consequences of governance failure are most severe, and show why the same failure is likely to recur across other public services. The two mitigations that dominate policing AI strategy - expert evaluation and human-in-the-loop oversight - both rest on assumptions that GPAI violates. Safety assurance thus shifts from an intrinsic feature of building an AI tool to an optional add-on. We recommend a clear taxonomic distinction between narrow and general-purpose AI in governance documentation, a preference for technological parsimony, a pause on operational deployment of GPAI in policing until adequate evidence exists, and a coordinated national safety infrastructure with the authority to generate that evidence and determine when responsible deployment is achievable.

Beyond Local Inspection: Global, Guideline-Grounded Evaluation of Post-hoc XAI Methods for ECG Classification

Nils Gumpfer, Michael Guckert, Samuel Sossalla +2

Explainable AI (XAI) is used to assess whether artificial intelligence models rely on meaningful patterns, yet explanations that appear plausible for individual predictions may systematically misrepresent model behavior. This is particularly problematic in medicine, where models may rely on irrelevant signal characteristics rather than disease-specific patterns without being recognizable. We address this challenge using electrocardiogram (ECG) data, for which clinical guidelines provide explicit knowledge about diagnostically relevant signal regions. We introduce a global, guideline-grounded framework that aggregates explanations across heartbeats to evaluate them against clinically defined regions of interest. Using four binary classifiers trained on PTB-XL, we assess 13 gradient-based methods across two categories of patterns: low-amplitude segments and high-amplitude QRS morphology. Our results reveal a systematic failure of methods transferred from computer vision. Their explanations often follow signal amplitude rather than clinical relevance, with mean Spearman correlations up to 0.69, leading them to overlook diagnostically decisive low-amplitude regions. For ischemia, LRP-$ε$ assigns only 4.6% of relevance to the ST segment, compared with 63.8% for LRP-SIGN. Nine of 13 methods fall below chance for at least one condition, indicating inconsistent reliability across patterns. These findings show that global, domain-grounded evaluation can uncover systematic explanation failures not obvious from sample-level heatmaps.

Mainstream AI research emphasises capability growth and tolerates low failure rates when average-case performance is high. AI safety and alignment research has a different mission: to ensure that catastrophic failures never occur, under sparse evidence, adversarial dynamics, and fat-tailed risk. We argue that the two domains differ along two analytically independent axes---{\it capability profile}, demonstrating the absence of hazardous behaviours rather than the presence of positive capabilities, and {\it risk profile}, bounding worst-case outcomes under fat-tailed uncertainty rather than optimising average-case performance---and that mainstream epistemic practices are inadequate on both. Building on a structured synthesis grounded in a preregistered bibliometric baseline, we identify five cross-cutting gap dimensions in current alignment research, including the near-absence of institutionalised independent verification. To address these gaps, we propose {\sc ECAISA}, an Epistemic Code for AI Safety and Alignment comprising eight principles, a three-level scoring rubric, a four-level disclosure ladder that reconciles transparency with information-hazard and commercial-confidentiality constraints, a tiered applicability scheme, an information-hazard adjudication procedure, and seven anti-gaming mechanisms. {\sc ECAISA} does not certify that any AI system is safe; it constrains how safety-relevant research claims are documented, checked, and relied upon, with auditability rather than certification as its governance target.

Large language model (LLM) agents inherit reactive failure modes: escalation under provocation, sycophantic drift under flattery, perseveration when stuck. These are failures of propensity, not capability; they concern what a model does under sustained pressure, which training-time alignment reduces but does not eliminate at runtime. This research led to the Gubernaut Cognitive Controller (GCC), a model-agnostic runtime control layer in a Nelson--Narens monitoring--control loop: an object level reads and writes text, while a deterministic meta level reads only the numeric telemetry {intensity, valence, repetition} and returns a regulating posture. Because the meta level ingests zero tokens, no injection channel to the controller exists by construction (an architectural property, not yet adversarially tested); the text-exposed arbiter's compliance is measured, not assumed. We evaluate the GCC with a pre-registered, generate-once/judge-many protocol across a 4x4 matrix of four frontier models (GPT-5.5, Claude Opus 4.8, Gemini 3.5 Flash, Grok 4.3), each serving as both a generator and a judge. The regulated arm is calmer in 13 of 16 cells at p<.05 and 15 of 16 by sign; the three sub-threshold cells, including a -0.04 null, all fall on the single near-saturated host. The effect survives a lineage-independent fourth judge family (xAI), strong evidence that it is no artifact of shared judge style. The clearest mechanism is the recovery signature: arousal that integrates under attack and then decays, valence-gated, on de-escalation, replicating across all four families. Transcripts and panels ship with SHA-256 provenance and are re-judgeable; five failure modes are pre-registered. No consciousness claims are made.

Background: Large language models (LLMs) are increasingly used to automate code review, but the reasoning behind their decisions remains hard to understand. Developers struggle to assess the validity of LLM-generated reviews, making it difficult to gauge how much trust to place in them. The role of Explainable AI (XAI) in code review and its impact on trust remain underexplored. Objective: We study the influence of XAI on developer trust in AI-assisted code reviews. Method: We conducted a within-subjects user study with 34 participants, comparing three LLM-based code review systems with varying levels of XAI support: Condition A (detailed explanation and review feedback), Condition B (review feedback only), and Condition C (no explanations). Participants reviewed real-world code change requests alongside the AI-generated reviews. We measured trust perceptions, agreement with the AI recommendation, the reasoning given for each decision, and the time taken. Results: The level of explanation significantly influences both trust and agreement with AI recommendations, but in different ways. Full explanations (A) yield the highest perceived trust (M = 3.99/5) but not the highest agreement, whereas moderate explanations (B) achieve the highest agreement (89.22%). This could suggest that more explanation prompts developers to question AI recommendations more frequently. No explanations (C) results in the lowest trust and agreement. Explanation level did not significantly affect review time. The most commonly cited reasons for decisions were code readability and correctness. Conclusion: Incorporating XAI into code review significantly changes trust perceptions and agreement with AI recommendations. These results inform the design and evaluation of trustworthy AI-based code review systems, as well as studies on the human factors of AI-assisted software development.

Political audits of large language models (LLMs) usually reduce each to one point on a political compass. But that resting point barely matters in deployment: a model must land somewhere, and what counts is how far, and in which directions, its answers can be steered. That steering runs through the system prompt: the personalization layer a platform sets, or one induced from a user's history, not necessarily written by hand. We run a dispersion-first stress test of prompt-based controllability across 12 ideological personas plus an unsteered baseline, 70 Political Compass items, ten replicates, and seven leading LLMs: GPT-5, Claude, Grok, Gemini, DeepSeek, Kimi, and Qwen (63,700 responses). Contextual framing explains roughly 88%-93% of variance on the economic and society axes, model identity under 3%: responses are highly instruction-adjustable. Models do not shift alike: some move more, and some saturate under extreme framings. Conflicting directional-steering results in prior audits resolve once baselines are recognized as non-centered: displacement and proximity diverge, so the effect is geometric, not differential compliance. Under authoritarian prompts, models produce similar shifts on the same questions. Political-coordinate audits therefore need steerability audits reporting dispersion, symmetry, saturation, and refusal floors. We release prompts, benchmark data, and code.

The rapid deployment of large language models (LLMs) in healthcare settings makes the reliability of their built-in guardrails against malicious queries a question of urgent practical consequence. Yet the robustness of these mechanisms against deliberate misuse (in the healthcare context) remains poorly understood. In this paper, we investigate this question empirically, using AI-assisted medical note manipulation as a concrete case study. We make four novel contributions. First, we develop a reproducible manipulation pipeline that takes publicly available seed medical note templates and use commercial LLMs to produce customized manipulated notes by substituting patient names, provider identities, dates, and medical conditions across multiple model families, input formats, and prompt phrasings. Second, we conduct a systematic empirical evaluation of LLM guardrail robustness for medical note manipulation. Our experimental results reveal substantial weaknesses and inconsistencies in contemporary commercial LLM guardrails, including low refusal rates for several model families. Third, we utilize a combination of automated metrics and human annotation-based metrics to assess the correctness of requested manipulations. Fourth, we conduct a user-study to assess the believability of manipulated medical notes, finding that the best manipulations are visually indistinguishable from original documents to human raters. Finally, we discuss implications for responsible guardrail design in LLMs, AI safety policies, and the broader ethics of deploying LLMs in healthcare settings.

AI safety has a missing layer. Interpretability, formal methods, security engineering, evaluation methodology, and reinforcement-learning safety each produce substantial work, but the resulting artifacts do not compose into deployable oversight: every team fielding an agentic system builds its own audit schema, policy dialect, monitoring stack, and escalation path, mostly reinventions of patterns understood elsewhere. We diagnose this as a coordination gap, not a research gap, and propose a two-axis taxonomy: five technical layers (Legibility, Specification, Mediation, Evaluation, Escalation) crossed with six concerns spanning alignment, robustness, adversarial defense, security, governance, and accountability, populating the resulting 5x6 matrix with existing work. Layer 2 (Specification), where humans translate intent into machine-checkable artifacts, is the connective tissue every layer depends on, yet it lacks four marks of a mature engineering discipline: shared vocabulary, design principles, composability standards, and governance practices. We propose six design principles for Layer 2, from elicitability and composability to adversary-awareness, traceability, and governability, made concrete through worked examples and a reference architecture turning specifications into runtime enforcement, evaluation, and escalation. Existing systems such as Cedar, Constitutional AI, and Open Policy Agent each address a fragment of Layer 2 well and the matrix poorly; treating them as fragments of one shared layer makes composition tractable. As evidence, we introduce CARMA, a Layer 2 prototype for autonomous ETL agents in which one specification drives enforcement, evaluation, and escalation, with every decision traceable to a versioned specification, naming what AI oversight is missing and giving independent teams principles to build the missing pieces so they compose.