Explainability Evaluation

Innovations in digital intelligence are transforming robotic surgery with more informed decision-making. Real-time awareness of surgical instrument presence and actions (e.g., cutting tissue) is essential for such systems. Yet, despite decades of research, most machine learning models for this task are trained on small datasets and still struggle to generalize. Recently, vision-Language Models (VLMs) have brought transformative advances in reasoning across visual and textual modalities. Their unprecedented generalization capabilities suggest great potential for advancing intelligent robotic surgery. However, surgical VLMs remain under-explored, and existing models show limited performance, highlighting the need for benchmark studies to assess their capabilities and limitations and to inform future development. To this end, we benchmark the zero-shot performance of several advanced VLMs on two public robotic-assisted laparoscopic datasets for instrument and action classification. Beyond standard evaluation, we integrate explainable AI to visualize VLM attention and uncover causal explanations behind their predictions. This provides a previously underexplored perspective in this field for evaluating the reliability of model predictions. We also propose several explainability analysis-based metrics to complement standard evaluations. Our analysis reveals that surgical VLMs, despite domain-specific training, often rely on weak contextual cues rather than clinically relevant visual evidence, highlighting the need for stronger visual and reasoning supervision in surgical applications.

The evaluation of explainable AI (XAI) approaches often relies on three families of methods: subjective measures (e.g., questionnaires on trust or satisfaction), objective measures (e.g., task performance metrics), and mathematical metrics (e.g., for faithfulness). Yet, it remains unclear how these families align or diverge in practice. In a{preregistered} between-subjects study (N=166), we use three established saliency map techniques (LIME, Grad-CAM, Guided Backpropagation) as a testbed to examine this issue. We find that each family of methods leads to different conclusions: participants reported no differences in trust or satisfaction, Grad-CAM improved user performance, while mathematical metrics favored Guided Backpropagation. At the same time, mathematical metrics were only partially related to user performance, and these relationships were sometimes counterintuitive. Our findings highlight the methodological importance of comparing subjective, objective, and mathematical approaches when evaluating XAI, illustrating both tensions and aspects that are aligned. We discuss implications for XAI evaluation frameworks.

When asked to explain their decisions, LLMs can often give explanations which sound plausible to humans. But are these explanations faithful, i.e. do they convey the factors actually responsible for the decision? In this work, we analyse counterfactual faithfulness across 75 models from 13 families. We analyze the tradeoff between conciseness and comprehensiveness, how correlational faithfulness metrics assess this tradeoff, and the extent to which metrics can be gamed. This analysis motivates two new metrics: the phi-CCT, a simplified variant of the Correlational Counterfactual Test (CCT) which avoids the need for token probabilities while explaining most of the variance of the original test; and F-AUROC, which eliminates sensitivity to imbalanced intervention distributions and captures a model's ability to produce explanations with different levels of detail. Our findings reveal a clear scaling trend: larger and more capable models are consistently more faithful on all metrics we consider. Our code is available at https://github.com/google-deepmind/corr_faith.

Concept-based explanations help users understand vision models through recognizable visual patterns. However, existing methods often rely on correlational signals without directly validating which image cues support prediction-relevant internal features. To this end, we introduce VisionLogic, a post-hoc framework that grounds these features in visual concepts through intervention-based validation. VisionLogic first identifies compact sets of features whose contributions reproduce the model's original prediction. It then represents their activation states as predicates using class-specific thresholds. An iterative refinement procedure grounds these predicates in visual regions through ablation tests. A region is accepted when its removal deactivates the corresponding predicate, linking the feature's numerical role to visual evidence in the input. The same predicates allow us to examine how features are activated, selected, and reused across images and classes. Across CNNs and vision transformers on ImageNet-1k, we find that only a few features are selected to explain each prediction, and frequently active features are not always selected. In a large-scale human evaluation with 465 participants, VisionLogic significantly improves participants' understanding of model behavior over established methods ACE and CRAFT. Code is available at https://github.com/allengeng123/VisionLogic.

Existing algorithms for explaining the output of image classifiers use different definitions of explanations and a variety of techniques to find them. However, none of the existing tools use a principled approach based on formal definitions of cause and explanation. In this paper we present a novel black-box approach to computing explanations grounded in the theory of actual causality. We prove relevant theoretical results and present an algorithm for computing approximate explanations based on these definitions. We prove termination of our algorithm and discuss its complexity and the amount of approximation compared to the precise definition. We implemented the framework in a tool ReX and we present experimental results and a comparison with state-of-the-art tools. We demonstrate that ReX is the most efficient black-box tool and produces the smallest explanations, in addition to outperforming other black-box tools on standard quality measures.

Post-hoc attribution is widely used to identify important model inputs, but evaluating whether these attributions identify truly informative features is difficult because real datasets rarely provide reliable ground truth. We introduce a multimodal benchmark containing symbolic tabular, vision, and audio tasks with known informative feature sets. In the main benchmark conditions, informative variables, spatial regions, or channels occupy fixed input coordinates while the remaining inputs provide irrelevant or distracting information. We use the benchmark to study how attribution quality depends on predictive performance, irrelevant-feature burden and structure, model configuration, and attribution mechanism, while separately measuring identification, stability, and computational cost. In most symbolic-data sweeps, informative-set identification co-varies with predictive performance, while the relative ordering of attribution methods remains largely stable. Across modalities, no method dominates all architectures and conditions, and greater attribution complexity does not consistently improve identification. Simple gradient attribution is often competitive at lower computational cost, while the vision and audio results show that architecture and the form of irrelevant content materially affect attribution quality.

Grad-ECLIP is published at ICML 2024 and represents a new Transformer interpretation technical route (intermediate features-based). First, this paper demonstrates that the intermediate features-based technical route is not a novel one. Based on the existing attention-based route, we have developed Attention-ECLIP, which is completely equivalent to Grad-ECLIP but with simpler computation. Both through formal derivation and experimental validation, we prove that the intermediate feature-based route represented by Grad-ECLIP is actually an equivalent variant of the attention-based route. Next, this paper demonstrates that the Grad-ECLIP method is flawed. The model interpretation results obtained by Grad-ECLIP are not those of the original model, and the interpretation results are misaligned with the model's performance. We analyze the causes of Grad-ECLIP's flaws and propose, or rather, explicitly emphasize two fundamental principles that model interpretation should adhere to in order to avoid similar errors.

Simulatability is an evaluation protocol for explanations that quantifies their usefulness by how well they help a user predict a task model's outputs. Since human evaluation is costly, automated simulatability replaces human explainees with LLM simulators, as proposed in ConSim (Poch'e et al., 2025) for large-scale experiments. We qualitatively replicate and extend ConSim's ranking of explanation methods across the tested datasets, explanation families, and simulator LLMs, and identify two limitations. First, when class names are meaningful, simulators can obtain high simulatability by solving the classification task directly, without relying on the explanations. Second, class anonymization can reward explanations for leaking the hidden label mapping, a limitation we expose with a new classes-as-concepts baseline. These results are consistent with a shortcut hypothesis: in the tested settings, simulator predictions mainly rely on task priors, while explanations produce small changes. We derive recommendations for more robust automated simulatability evaluations.