AI evidence extraction

Are Researchers Being Replaced by Artificial Intelligence?

Angelo A. Salatino, Ansgar Scherp, Christin Katharina Kreutz, Sahar Vahdati · ArXiv.org · 2026

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

9/10

Relevance

3/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

Methodology & findings

Study design

Hermeneutic analysis and conceptual argumentation.

Main result

The paper identifies a fundamental paradox: "integrating AI into the scientific lifecycle accelerates productivity and is a welcome support in alleviating repetitive, time-consuming tasks. At the same time, it also threatens to decouple human researchers from the intellectual heavy lifting that defines scientific rigor." The authors argue that replacement represents not job loss but a shift from "researcher-as-creator to researcher-as-curator, where the intellectual work is outsourced but the responsibility remains with the human."

Reports effect sizes.

Research paradigm

Critical interpretive analysis

Author conclusions

The authors conclude: "Bottom line, there will be a silver lining, or a path of mutual benefit for agentic AI and humans doing research. Surely, there has to be a willingness to adapt, to change one's skill set, work routines, and scientific procedures. By embracing affordable agentic AI as a 'Swiss army knife' of research, our technology stack, learning methods, and ethical frameworks will evolve." They also note that "AI agents might become co-authors, assume responsibility for their actions, or even participate as humanoid robots at future conferences."

Risk of bias

Confirmation bias risk: The authors self-identify as "AI practitioners" advocating for field evolution while simultaneously providing critical examination, creating potential cognitive conflict. The intentionally exaggerated position on risks may overstate harms. No systematic search strategy was employed, introducing selection bias in the examples cited.; Confirmation bias: authors advocate for the field's evolution while examining risks, potentially selecting examples that highlight concerns; Lack of empirical grounding: conclusions drawn from hypothetical scenarios rather than measured outcomes; Incomplete representation: focuses on risks without systematic comparison of benefits across research domains; Selection bias in tool examples: curated list of AI tools may not represent comprehensive landscape; Speculative framing: analysis projects future scenarios rather than measuring current displacement; Selection bias in tool examples: curated list of AI research tools presented without systematic inventory methodology; Confirmation bias risk: framing emphasizes risks and challenges without balanced evidence of benefits

Limitations

The authors explicitly state these challenges "represent an intentionally exaggerated position and thus reflect the upper bound of potential risks and challenges of using AI in research." The paper lacks empirical data on actual impacts of AI replacement and relies on hypothetical scenarios and anecdotal examples rather than systematic measurement of researcher displacement or competency erosion.

Open questions raised

The authors identify gaps in governance and accountability mechanisms, including unclear frameworks for measuring and safeguarding distinct human researcher value in AI-augmented landscapes, lack of tangible consequences for AI-generated errors in peer review, unclear scalability of scholarly validation mechanisms, and unresolved questions about who will maintain authority and oversight for validity of AI-generated research.
The authors identify gaps in understanding: (1) how to measure and safeguard distinct value of human researchers in AI-augmented landscape; (2) how to establish clear accountability when AI output scales faster than scrutiny; (3) how to prevent erosion of cognitive competencies in next generation of scholars; (4) how to ensure validity and novelty beyond throughput expansion.
How to measure and safeguard the distinct value of human researchers in AI-augmented landscapes
Mechanisms to preserve accountability when AI output scales faster than human scrutiny
Whether current AI systems can reliably propose genuinely novel hypotheses beyond recombination
How to prevent model collapse and synthetic self-reinforcement from narrowing scientific diversity

Extracted from: pdfAgreement 73%

Explore related topics

Epistemic Risk Human-AI Collaboration Position Paper