AI evidence extraction

Classification of artificial intelligence tools for educational research by the criterion of research autonomy

T A Vakaliuk, Serhiy O. Semerikov, Oleh M. Spirin, Viacheslav Osadchyi, Vasyl Oleksiuk · CTE Workshop Proceedings · 2026

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

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.55056/cte.1357

Methodology & findings

Study design

Conceptual synthesis grounded in actor-network theory and automation theory.

Main result

The paper proposes a five-cluster classification system organized by research autonomy as the primary criterion. The central finding is that "delegation in the second case is not computational but epistemic -the tool decides what counts as a meaningful connection, a function that previously belonged to the researcher and that carries theoretical commitments shaping the status of the resulting knowledge claim." The framework maps Molenaar's six automation levels onto five functional clusters specific to educational research, where "Research autonomy is the degree to which a researcher retains control over the cognitive operations constitutive of scientific knowledge production: observation and classification of phenomena, detection of patterns in data, interpretation of pattern significance, synthesis of heterogeneous information into a conceptual frame, and generation of new concepts and theoretical propositions."

Research paradigm

Hermeneutic/interpretivist; grounded in actor-network theory and automation theory

Author conclusions

The authors conclude that "The absence of a systematic framework connecting tool choice to epistemic consequence leaves educational researchers without principled guidance for the most consequential decisions they face when using AI: not how to operate a tool, but what they are surrendering when they do." They propose that the classification "organizes five functional clusters in descending order of research autonomy" and that "The practical output of the framework is a reporting standard requiring three elements per AI tool used: (1) the cluster assignment, (2) the cognitive operation delegated, and (3) the verification procedure." They further argue that "Research whose conceptual framework is formed primarily by model output rather than by the researcher's own engagement with primary sources may constrain originality, since the model's outputs tend toward what has already accumulated mass in the training corpus."

Risk of bias

Selection bias in literature survey: non-systematic source identification may exclude relevant publications; Authorial positionality bias: Ukrainian-context development may not generalize to other regulatory environments; Tool selection bias: framework reflects tools indexed in academic databases; proprietary or regional tools may be under-represented; Classification bias: cluster boundary determinations made through interpretive judgment rather than empirical validation; Analytical derivation without empirical validation of cluster boundaries; Limited source corpus (21 sources after exclusion); Potential English-language bias in literature identified (primarily peer-reviewed publications 2019-2026); Author positionality in Ukrainian research context may limit generalizability to other regulatory environments; Framework focuses on individual researcher autonomy, potentially missing team-based research configurations; Authorial positionality: framework developed in Ukrainian institutional context with heightened research integrity concerns; applicability in other regulatory environments untested; Literature selection bias: 21 sources retained from 47 candidates; non-systematic search strategy may not capture full scope of AI tools in educational research; Tool selection bias: framework focuses on tools cited in reviewed literature; emerging tools may be underrepresented; Temporal bias: sources drawn from 2019-2026; earlier AI applications in educational research may be underrepresented

Limitations

The authors identify five key limitations: (1) "Cluster boundaries are analytically derived from the intersection of automation theory and the AI tools literature
They have not been validated through surveys of researcher behaviour or observational studies of tool use." (2) "The framework was developed by researchers affiliated with Ukrainian institutions in a context of heightened concern for research integrity following Russia's full-scale invasion
Its applicability in other regulatory environments -the EU AI Act framework, US institutional review contexts, low-resource settings without consistent tool access -has not been empirically tested." (3) "Products cross cluster boundaries as capabilities evolve." (4) "The framework assumes a single researcher making autonomous decisions
Team-based research with distributed cognitive labour -where hypothesis generation, coding, and synthesis are assigned to different members -involves a different configuration of autonomy that the current framework does not address." (5) "The methodological obligations identified for Cluster IV-V tools do not incorporate the environmental costs (energy, carbon) of inference at scale, nor do they address the unequal access to proprietary Cluster III-V tools across institutions and countries."

Open questions raised

Empirical validation of cluster boundaries through surveys of researcher behaviour and observational studies of tool use
Testing framework applicability across different regulatory environments (EU AI Act, US institutional review, low-resource settings)
Extension of framework to team-based research with distributed cognitive labour
Multi-dimensional modeling of research autonomy (current framework treats it as single-axis simplification)
Integration of environmental and access costs in methodological obligations
Lack of systematic framework connecting tool choice to epistemic consequence in existing literature

Extracted from: pdfAgreement 46%

Explore related topics

Responsible AI Use AI Governance Conceptual