Early-Signal Detector

method · first seen 4 Oct 2026

explainable AI

Also: model interpretability

Techniques that make the predictions and decisions of AI systems understandable to humans.

Explainable AI covers methods such as feature attribution (SHAP, Shapley values), saliency and post-hoc explanations that make model outputs understandable to people. The sources are a routine spread of applied papers rather than a new technique: XAI for maritime failure prediction, IoT anomaly detection, space weather forecasting and clinical decision support. One recurring thread is using LLMs as an explanation layer, with work on conversational XAI, RAG-based clinical explanations, fairness in LLM-written narratives of SHAP output, and RL-tuned explanations aimed at human decision quality. Other papers ask whether explanation fidelity holds up under data drift.

The field itself is long established, so the rise here is mostly noise. The sub-trend of LLMs turning attribution outputs into natural-language explanations is worth watching, because it raises new questions about faithfulness and bias in regulated domains.

Analyst note (established), 17 Nov 2026.

Mentions per week

0619 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 125 on 14 Nov 2026 · 12 mentions in 28 days · 12 independent groups · papers → code → discussion · first seen 6 weeks ago · acceleration +0.57

Earliest sources

Latest sources

Mentioned together with

concept driftreliance calibrationpredictive maintenancereinforcement learning from human feedbackmultimodal large language model physics reasoningmental modelsfederated learningransomware detectionfraud detectionocclusion-based saliency