Revolutionizing Scientific Citations: How LLMs Alter the Rhetoric and Social Reach of Academic References

The landscape of academic writing is evolving with the emergence of large language models (LLMs) that assist researchers in crafting their citations. A recent study conducted by Yixuan Liu and colleagues at Northeastern University investigates how LLMs transform scientific citation behaviors, revealing significant shifts in rhetorical intent and social connections among cited authors.

The Role of Citations in Academic Writing

Citations are more than just references; they convey the author's stance towards previous work, whether it's supportive, contrasting, or neutral. Traditionally, these decisions reflect the author's insights and social connections within their field. However, the integration of LLMs in writing raises questions about the preservation of these rhetorical intents. Can an LLM replicate the nuanced decision-making process of human authors?

Key Findings: LLMs Cite "Less Critically"

The study developed a masked-citation task to analyze how LLMs compare to human citation practices. The findings are revealing:

• LLMs cited significantly fewer contrasting works, showing a tendency to generate more supportive citations instead. This "warming" of citation tone means critical voices in the literature may be softened or erased.
• The models displayed a bias towards citing older, well-known works over newer, possibly more relevant research. This tendency creates a bias toward visibility that may marginalize emerging voices in academia.
• Humans often cite from within their immediate social networks, whereas LLMs tended to reference more distantly related authors, highlighting a disconnect in academic collaboration patterns.

Implications for the Future of Academic Citations

The implications of this research are profound. By generating citations with less criticism and favoring established works, LLMs could inadvertently promote a homogenized perspective in scientific writing. While they may broaden the reach of citations—introducing researchers to works outside their typical circles—they also risk diluting the critical engagement that is so vital in academia.

Continued Oversight and Future Research

As LLMs become increasingly embedded in academic writing, researchers must remain vigilant about the biases these models introduce. Continuous auditing of LLM-assisted citation practices is essential to preserve the integrity and richness of scholarly discourse. Future studies are necessary to probe deeper into the mechanisms behind these observed biases, ensuring that the benefits of LLMs do not come at the cost of critical scientific dialogue.

In conclusion, while LLMs offer promising tools for enhancing productivity in academic writing, their influence on citation practices calls for careful consideration. The need for critical engagement in research must not be overshadowed by the convenience of automated writing assistance.

Authors: Yixuan Liu, Lin Chen, Zhuoqi Liu, Jianglin Lu, Dakota Murray