Toward Cognitively Grounded AI: Can a Working Memory Interface Bridge Individual Differences and System Design?

The rapid integration of generative AI into language education has
exposed a dual challenge: a “grounding gap” in its cognitive
shallowness, and a profound ethical peril that such technology may
accommodate rather than empower learners. This editorial interrogates
whether a bridge is possible. We argue that working memory (WM)—
empirically central to language aptitude and learning—offers the most
viable, if fraught, interface for such a bridge. We introduce the Cognitive
Load–WM Interaction (CLWM) Matrix not as a solution, but as a
critical heuristic and necessary safeguard. It is designed to enforce a key
distinction: between AI that grounds learning by managing cognitive
load and AI that bypasses cognitive effort. From this, we derive a dualpath risk-aware research agenda, focused on developing WM-aware
pedagogical specifications and probing hybrid AI architectures. The
conclusion is not a blueprint, but a condition: progress in AI must be
subordinated to progress in understanding and protecting the human
cognitive processes it seeks to engage.