AI evidence extraction

Sustainable AI-Driven Assessment in Higher Education: A Systematic Review of Fairness, Transparency, Pedagogical Innovation, and Governance

Maha Alfaleh · Sustainability · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

7/10

Relevance

0/4

Quality (LMQS)

Evidence

Citations

30.92

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.3390/su18020785

Methodology & findings

Study design

Systematic review employing the PRISMA 2020 framework, analyzing 47 studies published between 2019 and 2025 across Western, Gulf, South Asian, and East Asian contexts.

Sample

N = 47, 5 groups

Primary method

Systematic review methodology with narrative synthesis following PRISMA 2020 framework. No specific statistical pooling or meta-analytical methods are described in the abstract.

Main result

The systematic review found that "AI achieved greater scoring consistency than human graders in over two-thirds of fairness-focused studies." Additionally, "more than half of the transparency studies identified inadequate or partial disclosure of AI decision processes." Furthermore, "AI-enhanced feedback frequency and revision opportunities in approximately 70% of cases, although teacher mediation was necessary to mitigate over-reliance."

Reports effect sizes.

Research paradigm

Mixed methods (quantitative evidence synthesis with qualitative thematic analysis)

Author conclusions

The authors conclude that "the review proposes a governance-anchored model that integrates fairness and transparency with pedagogical design, providing a coherent framework for institutions aiming to implement AI-based assessment responsibly and equitably."

Risk of bias

Selection bias (publication language restrictions not specified), potential geographic bias (focusing on specific regions only), heterogeneity in study designs and contexts across the 47 included studies, potential outcome reporting bias in transparency studies.; Geographic limitation to Western, Gulf, South Asian, and East Asian contexts (potential geographic bias); Time period restriction to 2019-2025 (temporal bias); Potential publication bias (only published studies included in systematic reviews); Selection bias: Limited to studies published between 2019-2025, potentially excluding earlier relevant work; Geographic bias: Limited to specific regional contexts (Western, Gulf, South Asian, East Asian); Publication bias: Systematic reviews inherently prone to publication bias favoring positive results; Study heterogeneity: Significant variation in methodologies across 47 included studies may limit synthesis validity

Open questions raised

The review identifies fragmented existing research with limited synthesis regarding the interplay of fairness, transparency, pedagogy, and governance in AI-driven assessment. The authors note that fewer than one-third of institutions had established policies or audit mechanisms for ethical AI use, indicating a significant gap in governance practices.
The authors identified that existing research on AI in higher-education assessment "remains fragmented, with limited synthesis regarding the interplay of fairness, transparency, pedagogy, and governance." They also highlight that "fewer than one-third of institutions had established policies or audit mechanisms for ethical AI use," indicating a governance implementation gap.
The abstract indicates existing research is fragmented, with limited synthesis regarding the interplay of fairness, transparency, pedagogy, and governance. The review addresses this gap but implies further work is needed on institutional implementation of governance frameworks and ethical AI audit mechanisms.

Data: not_statedCode: not_statedExtracted from: pdfAgreement 60%

Explore related topics

Institutional Policy AI Governance