AI evidence extraction

Beyond replacement anxiety: a psychological framework for understanding AI in natural science research

Xinyi Qi, Yuchen Gao, Peiqing Sun · Frontiers in Psychology · 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/10.3389/fpsyg.2026.1824256

Methodology & findings

Study design

Theoretical framework development and conceptual analysis.

Main result

The paper identifies four interconnected psychological dimensions through which AI reorganizes the experience of natural science research: "labor visibility concerns whether human effort remains recognizable in AI-mediated workflows; identity stability concerns the continuity of professional self-concept and developmental learning; accountability under delegated cognition concerns the asymmetry between delegated cognitive operations and retained human responsibility; and institutional climate concerns the local norms and governance structures that shape how the other three dimensions are interpreted and managed." The authors argue that "AI can feel simultaneously empowering and destabilizing in research environments."

Reports effect sizes.

Research paradigm

Interpretivist/critical realist perspective on organizational psychology and human-AI collaboration

Author conclusions

The authors conclude that "The challenge is therefore not to defend a pre-AI past or to surrender to automation as an unquestioned future. It is to design research cultures in which human agency remains meaningful inside AI-rich environments." They further state that "A psychological perspective does not slow innovation; it makes innovation more sustainable. With such a perspective, we can ask not only what kinds of discoveries AI may enable, but also what kinds of researchers-and what kinds of research lives-scientific institutions will continue to make possible."

Limitations

The authors acknowledge that "the present framework should be read not as a claim that all of these dynamics have already been fully established across natural science settings, but as a theoretically reasoned account of pressures that are becoming plausible, visible, and testable in AI-rich research environments." Additionally, they note that "AI adoption differs across laboratory sciences, computational fields, interdisciplinary teams, and publication-centered tasks, so the framework proposed here is intended as a cross-context lens rather than as a claim of identical effects across all natural science domains." They further state that "at present, these identity implications should be read as plausible extensions requiring direct empirical examination in research-training environments."

Open questions raised

Future work should "examine these dynamics at multiple levels, including individual researchers, research groups, doctoral and postdoctoral training environments, and institutional policy regimes." Candidate variables include "perceived labor recognition, invisibility of validation work, identity insecurity, developmental role ambiguity, delegation burden, trust calibration, disclosure uncertainty, and perceived institutional permission to use AI cautiously or selectively." The authors recommend "mixed-methods designs that combine surveys, interviews, comparative institutional analysis, and field-based qualitative inquiry" to test the framework's applicability.
The authors identify multiple future research directions: "Future work could examine these dynamics at multiple levels, including individual researchers, research groups, doctoral and postdoctoral training environments, and institutional policy regimes. Candidate variables include perceived labor recognition, invisibility of validation work, identity insecurity, developmental role ambiguity, delegation burden, trust calibration, disclosure uncertainty, and perceived institutional permission to use AI cautiously or selectively. Mixed-methods designs that combine surveys, interviews, comparative institutional analysis, and field-based qualitative inquiry may be especially useful for testing where the framework travels well and where its boundary conditions become visible."
The authors identify several research gaps: (1) The need for direct empirical examination of identity implications in research-training environments; (2) An empirical agenda examining dynamics at multiple levels including individual researchers, research groups, doctoral/postdoctoral training, and institutional policy regimes; (3) Future work examining variables such as perceived labor recognition, invisibility of validation work, identity insecurity, developmental role ambiguity, delegation burden, trust calibration, and disclosure uncertainty; (4) Mixed-methods designs combining surveys, interviews, comparative institutional analysis, and field-based qualitative inquiry to test where the framework applies and its boundary conditions.

Extracted from: pdfAgreement 80%

Explore related topics

Responsible AI Use Human-AI Collaboration Conceptual