The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2708
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T17:12:19.640359Z
status: published
lens: g_space
sector: education
headline: Emerging Technology-Supported Intercultural Learning for Cognitive, Attitudinal, and Behavioral Outcomes: A Systematic Review and Three-Level Meta-Analysis
dek: Emerging technologies are increasingly used to support intercultural learning, yet evidence on the distinct effects of artificial intelligence (AI), augmented reality (AR), and virtual reality (VR) remains limited. This systematic review and three-level meta-analysis synthesized 17 studies published from 2020 to July 2026 (77 effect sizes) to examine cognitive, attitudinal, and behavioral outcomes. The overall effect was large and significant (Hedges’ g = 0.965, 95% CI [0.580, 1.350], p < .001), although heterog…
gain_title: In a meta-analysis of nine controlled studies (43 effect sizes), learners taught intercultural skills with AI tools such as chatbots and adaptive systems scored higher than comparison groups, with a pooled effect of g = 1.150 (95% CI 0.744 to 1.555).
problem_title: (none)
trace_subject: (none)
gain_reading: In a meta-analysis of nine controlled studies (43 effect sizes), learners taught intercultural skills with AI tools such as chatbots and adaptive systems scored higher than comparison groups, with a pooled effect of g = 1.150 (95% CI 0.744 to 1.555).
gain_evidence: AI-supported interventions showed a significant positive effect, g = 1.150, 95% CI [0.744, 1.555], p < .001.
problem_reading: (none)
problem_evidence: (none)
quick_read: A systematic review and three-level meta-analysis pooled 17 studies published from January 2020 to July 2026 (77 effect sizes) on technology-supported intercultural learning. Across AI, augmented reality and virtual reality, learners beat comparison groups with a pooled Hedges' g of 0.965. The nine AI-supported studies, which used chatbots, generative AI, intelligent agents and adaptive systems, gave a pooled g of 1.150 across 43 effect sizes.

The AI studies covered different tools and tasks, including translation and feedback in a 12-week online exchange between Chinese and Malaysian students, an adaptive system that flagged culturally inappropriate expressions, and AI responses with cultural bias that learners were asked to critique. The authors say gains often relied on teacher-designed tasks, structured reflection and human partners, so the effect cannot be credited to the AI alone.

Results varied widely between studies. The prediction interval for the AI studies ran from about -0.29 to 2.59, meaning a new comparable study could find little or no benefit. All randomized studies in the review had some risk-of-bias concerns, and the evidence base is only 17 studies.
limitation: Effects varied widely between studies (the AI prediction interval was about -0.29 to 2.59), the review covers only 17 studies, all randomized studies had some risk-of-bias concerns, and the moderator comparisons were exploratory with uneven groups.
tag: Evidence-backed gain
key_points: The review pooled 17 studies (77 effect sizes) published between January 2020 and July 2026; eight were published in 2026. | The overall pooled effect of AI, AR and VR on intercultural learning was g = 0.965 (95% CI 0.580 to 1.350), with a prediction interval of about -0.781 to 2.711. | Nine AI-supported studies contributed 43 effect sizes and a pooled g of 1.150; dropping the one preprint raised it slightly to g = 1.217 across eight studies. | Heterogeneity was very high (total I-squared 93.88%), and most of it came from differences between studies rather than within them. | Outcome type (cognitive, attitudinal or behavioral) did not significantly change the effect; cognitive outcomes had g = 0.892. | AR looked stronger than VR (two AR studies versus six VR studies), but the gap was no longer significant after excluding one AR study. | K-12 and larger-sample studies showed bigger effects, but only two studies involved K-12 learners and the authors call these comparisons exploratory. | Three of the 17 studies compared against technology-based alternatives; the other 14 used conventional or non-technology instruction.
rundown: The authors searched Web of Science, Scopus and ProQuest through 31 July 2026 and included only randomized or quasi-experimental studies with a comparison group and enough data to calculate effect sizes. Two reviewers screened, coded and assessed risk of bias independently, with high agreement on coding (Cohen's kappa 0.96). Outcomes were sorted into cognitive, attitudinal and behavioral domains.

For AI, the studies used tools for different jobs: translation and feedback in a 12-week Chinese-Malaysian exchange, an adaptive system that detected and explained culturally inappropriate wording, an AI agent that responded during pragmatic conflicts, and deliberately biased AI answers used for critique. The largest effects in two studies appeared in critical cultural awareness and metacognitive cultural intelligence rather than factual knowledge.

Quality checks gave a mixed picture. All randomized studies were rated as having some concerns, five of seven non-randomized studies had moderate risk, and two were rated serious. Excluding those two serious-risk studies raised the overall effect to g = 1.110. Funnel-plot, Egger's and Begg's tests showed no strong sign of publication bias, though 17 studies may be too few to detect it.

Moderator findings were fragile. Academic level and sample size were significant, but K-12 was represented by two studies, sample size was unevenly spread across technology types, and the covariance-based sensitivity checks covered only the overall effect, not the moderators.
sources:
- peer_reviewed | Journal of Intelligence | https://doi.org/10.3390/jintelligence14100265 | 2026-10-16
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

ffbeb92ed4eda0fdd08a021c985c39bdb7678b243e81e762bd505e88c51a8d6e

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000