analiza aiqbench
This research demonstrates a critical gap in current AI reliability metrics, revealing that standard uncertainty quantification methods fail to align with genuine human ambiguity where annotators disagree. Consequently, relying on model confidence scores alone for high-stakes decision-making is insufficient, as models often converge confidently on instances that humans would flag for expert review. This finding necessitates a reevaluation of trust mechanisms in AI-assisted systems, particularly those operating in environments with inherent perceptual variance.
Does Model Uncertainty Track Human Ambiguity? Evidence from Multi-Annotator Vision Benchmarks
arXiv:2609.34506v1 Announce Type: new Abstract: Human-model alignment is critical for trustworthy AI-assisted decision-making systems. Yet, most work evaluates model predictions against single ground-truth labels, overlooking that humans themselves often disagree on labels, a signal of genuine ambiguity. We investigate whether models struggle on the same instances that humans find difficult. We measure this on two vision datasets (FER+ and CIFAR-10H) where multiple human annotations per image capture human disagreement patterns. We evaluate eight pretrained models across three architectures (ResNet, EfficientNet, MobileNetV3) in two parts: first, whether model uncertainty (softmax confidence, entropy) correlates with human disagreement, and second, whether predictive multiplicity measures (inter-model disagreement, Jensen-Shannon divergence) do. We find that it does not: alignment is weak in both dimensions. At the discrete label level, 50.4% of CIFAR-10H images and 33.5% of FER+ images receive multiple valid classifications from humans, while the models converge on only one. These instances represent a critical failure case where humans perceive ambiguity and would request expert review, yet models decide confidently. At the continuous score level, single-model uncertainty correlates weakly with human disagreement ($\rho = 0.24--0.55$), and predictive multiplicity provides only modest improvement. Widely-used uncertainty quantification methods do not reliably identify instances humans find ambiguous. Model uncertainty should not be treated as a trustworthy signal by default for decision-making in high-stakes scenarios.
Title:Does Model Uncertainty Track Human Ambiguity? Evidence from Multi-Annotator Vision Benchmarks
Abstract:Human-model alignment is critical for trustworthy AI-assisted decision-making systems. Yet, most work evaluates model predictions against single ground-truth labels, overlooking that humans themselves often disagree on labels, a signal of genuine ambiguity. We investigate whether models struggle on the same instances that humans find difficult. We measure this on two vision datasets (FER+ and CIFAR-10H) where multiple human annotations per image capture human disagreement patterns. We evaluate eight pretrained models across three architectures (ResNet, EfficientNet, MobileNetV3) in two parts: first, whether model uncertainty (softmax confidence, entropy) correlates with human disagreement, and second, whether predictive multiplicity measures (inter-model disagreement, Jensen-Shannon divergence) do. We find that it does not: alignment is weak in both dimensions. At the discrete label level, 50.4% of CIFAR-10H images and 33.5% of FER+ images receive multiple valid classifications from humans, while the models converge on only one. These instances represent a critical failure case where humans perceive ambiguity and would request expert review, yet models decide confidently. At the continuous score level, single-model uncertainty correlates weakly with human disagreement ($\rho = 0.24--0.55$), and predictive multiplicity provides only modest improvement. Widely-used uncertainty quantification methods do not reliably identify instances humans find ambiguous. Model uncertainty should not be treated as a trustworthy signal by default for decision-making in high-stakes scenarios.
Artificial Intelligence (cs.AI)
arXiv:2609.34506 [cs.AI]
(or arXiv:2609.34506v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.34506 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 316 cuvinte
Subiecte conexe
Debating RSI, the US-China Gap, and Jaggedness with JS Denain of Epoch AI
Interconnects · 22 sept. 2026
Against Export Controls (and China Threat Models)
LessWrong · 24 sept. 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · 2 oct. 2026
How I built agent-based security reviews on Databricks
Databricks · 24 sept. 2026