Framework learns causal models directly from human explanations
A new framework called xWhyL bridges causal inference and explainable AI by treating human explanations as training signals for causal discovery. Rather than treating XAI and causality as separate domains, the work formalizes how explanations can overcome observational data limitations in learning causal models. This inverts the typical XAI pipeline: instead of using causal models to explain predictions, the framework learns causal structure from explanations themselves. For practitioners building interpretable systems, this offers a path to stronger causal reasoning without requiring exhaustive interventional experiments.
Modelwire context
ExplainerThe key move is treating human explanations as a substitute for interventional data. Most causal discovery requires expensive experiments; xWhyL proposes that qualitative explanations from domain experts can encode causal constraints strong enough to learn structure from observational data alone.
This connects directly to the fluctuation-supervised pretraining work from earlier today. Both papers are trying to extract causal signal from limited data, but from opposite angles. FSP optimizes how foundation models learn causal effects once structure is known; xWhyL focuses on learning the structure itself when interventions are infeasible. Together they suggest a two-stage pipeline: use explanations to discover causal graphs, then use those graphs to train models that generalize effect estimation. The Learning to Defer paper also shares a core tension: both xWhyL and that work rely on human judgment (explanations vs. deferral decisions) to constrain what models can learn, bridging the gap between pure ML and human-in-the-loop reasoning.
If xWhyL is validated on a benchmark where ground-truth causal graphs are known but explanations are collected from practitioners who don't see the graph, watch whether the recovered structure matches the ground truth at >80% edge accuracy. If it does, that confirms explanations encode genuine causal knowledge rather than just correlation patterns.
Coverage we drew on
This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.
MentionsxWhyL
Modelwire Editorial
This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.
Modelwire summarizes, we don’t republish. arXiv cs.LG originally reported this story as “xWhyL: Causal Interactive Learning”. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.