AI evidence extraction

Scientific Artificial Intelligence: From a Procedural Toolkit to Cognitive Coauthorship

Adilbek K. Bisenbaev · Philosophies · 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

7.11

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.3390/philosophies11010012

Methodology & findings

Study design

Analytic-design approach combining philosophical grounding, structural formalization, and normative-ethical synthesis.

Primary method

Design science methodology integrating philosophical grounding with operational design. The approach proceeds through three stages: (1) philosophical grounding in instrumentality ontology; (2) structural formalization via the design of TraceAuth protocol and AIEIS metric; (3) normative-ethical synthesis integrating with existing editorial standards (COPE, ICMJE).

Main result

The paper establishes that "AI does not merely assist the researcher; it can reshape the problem space and the semantic horizons of inquiry" and proposes that "the boundary between 'instrument' and 'coauthor,' therefore, ceases to be ontological and becomes a question of participation regimes and degrees of cognitive impact." The core contribution is operationalizing AI's cognitive participation through "TraceAuth, a protocol for tracing the cognitive chain of participation (from ideation through interpretation to text), and AIEIS (AI epidemic impact score), a metric of epistemic impact that assesses contributions along the axes of procedural, semantic, and generative participation."

Research paradigm

Design science with philosophical grounding in epistemology and science and technology studies

Author conclusions

"In this configuration, we can sustain both the innovative potential of algorithmic coagency and the fundamental requirements of scientific integrity, reproducibility, and accountability." The authors conclude that "science has entered a phase of hybrid meaning-making, in which epistemic objects are jointly constructed by humans, algorithms, and infrastructures. To keep such science reproducible and accountable, protocols for documented participation and metrics of explanatory impact are needed." They assert that "The AI-AUTHORSHIP framework-together with TraceAuth and AIEIS-provides a practical standard for that transparency: AI's contribution is recognized, localized, and measured; explanations are interpretable and testable; and responsibility is human."

Risk of bias

This is a philosophical/theoretical paper with no empirical data collection, so traditional bias risks do not apply; Potential bias in the selection of philosophical frameworks and precedents; The proposed AIEIS weighting scheme may contain disciplinary bias if calibration is not sufficiently inclusive; The framework reflects the authors' normative commitments to transparency and anthropocentric responsibility; No empirical validation of proposed frameworks; Absence of user testing or pilot implementation data; Reliance on philosophical argumentation without quantitative validation of AIEIS scoring reliability; Potential confirmation bias in selecting historical cases supporting the cognitive-turn thesis

Limitations

The authors acknowledge that "the tripartite taxonomy (P/S/G) is heuristic rather than exhaustive" and that "the boundaries of good-faith disclosure remain blurry." They also note that "TraceAuth and AIEIS cannot rely solely on philosophical persuasiveness
it requires demonstrable, discipline-level benefits" and that implementation would require "pilot clusters..
where the issues of reproducibility and algorithmic mediation are already pressing." Additionally, they recognize that "the implementation of TraceAuth/AIEIS cannot rely solely on philosophical persuasiveness
it requires demonstrable, discipline-level benefits."

Open questions raised

Authors identify the need for: (1) pilot implementations in high-stakes disciplines (computational linguistics, biomedicine, scientometrics); (2) discipline-specific calibration of AIEIS weighting through Delphi panels; (3) development of uncertainty zones for S+G components to minimize underdisclosure; (4) mechanisms for automated metadata generation by AI systems to reduce TraceAuth burden; (5) integration with existing scientific readiness frameworks and citation/impact systems.
The authors identify the following gaps: (1) absence of mechanisms for measuring and attributing AI's cognitive contribution that distinguish between 'augmentation' and 'creation'; (2) methodological linkage needed to mitigate risks of stochastic eloquence while acknowledging factual contribution; (3) lack of unified verification protocols implementing ICMJE/COPE recommendations; (4) need for discipline-specific calibration of AI contribution thresholds; (5) requirement for pilot implementation across disciplines to demonstrate practical benefits.
Authors identify the need for: (1) empirical pilots in computational linguistics, biomedicine, and scientometrics to demonstrate practical benefits; (2) discipline-specific calibration of AIEIS weights via Delphi-style expert panels; (3) integration with scientific readiness levels (SRL) and CRediT taxonomies; (4) future iterations incorporating uncertainty zones for S+G components; (5) temporal reproducibility mechanisms for deprecated AI model versions.

Extracted from: pdfAgreement 53%

Explore related topics

Authorship & Attribution Research Integrity Conceptual