AI evidence extraction

Benefits and Risks of Using AI Agents in Research

Mohammad Hosseini, Maya Murad, David B. Resnik · The Hastings Center Report · 2026

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

10/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

14.46

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1002/hast.70025

Methodology & findings

Study design

Narrative review with critical analysis of existing AI applications and theoretical discussion of ethical and practical implications.

Sample

N = 2070, 2 groups

Primary method

Survey analysis; generative agent matching analysis (85% matching rate reported for AI agent responses compared to human responses in general social survey)

Main result

The paper identifies that "AI agents might offer tremendous benefits for scientific research and its administration. They can improve efficiency by performing routine tasks without the need for rest" and can "assist with writing and reviewing protocols, journal submissions, grant proposals, and other documents and with coding, record keeping, note taking, mixed-methods analyses, and email correspondence." However, the authors also found significant risks: "AI agents are prone to various errors, inaccuracies, and biases, including factual mistakes, citational errors, and reasoning known as 'hallucinations.'" Additionally, a survey revealed that "when AI is present, these skills are underused, leading to reduced work satisfaction for 82 percent of respondents."

Reports effect sizes.

Research paradigm

Philosophical inquiry with critical analysis

Author conclusions

The authors conclude that "automation of research tasks poses significant risks for science and society, and these need to be managed responsibly." They recommend that "a key step to move the debate forward and minimize risks is to reflect on which research tasks should (and should not) be automated." The authors argue for proactive measures including: "identifying individuals who are responsible for different AI-related products and systems, developing procedures or processes for reviewing and verifying AI-assisted research, and implementing changes needed to protect the integrity of science." They emphasize the urgent need for "researchers to receive education in algorithmic and AI literacy, including education in how to assess biases, identify errors and mistakes, and validate AI results."

Limitations

The authors acknowledge that "there is no straightforward solution to the deskilling problem, and its long-term effects on science and society remain largely unknown." Additionally, they note that "while it may sound exciting to imagine a future in which researchers no longer need to conduct certain tasks that are laborious and tedious, such as performing systematic reviews, delegating such responsibilities to AI agents may have unforeseen adverse impacts on the education, training, and development of the scientific labor force." They also state that "empirical studies focused on the impacts of delegating research tasks to machines are sorely needed."

Open questions raised

The authors identify several gaps: (1) The long-term effects of deskilling on science and society remain largely unknown; (2) Empirical studies focused on the impacts of delegating research tasks to machines are sorely needed; (3) It remains unclear what the impacts of AI agents will be on researchers' practical skillset and critical thinking abilities; (4) Existing ethical and legal frameworks governing research seem ill-suited to address challenges and risks posed by AI agents.
Long-term effects of deskilling on science and society remain largely unknown
Empirical studies focused on the impacts of delegating research tasks to machines are sorely needed
Unclear how the research enterprise should deal with workforce disruptions caused by AI automation
No straightforward solution exists to the deskilling problem
Lack of understanding about impacts of AI agents on researchers' practical skillset and critical thinking abilities in the long run

Extracted from: pdfAgreement 68%

Explore related topics

Responsible AI Use Research Integrity Position Paper