In recent years, the academic landscape has undergone a profound transformation in the ways knowledge is produced and textual engagement occurs—driven largely by the rise of generative artificial intelligence models, particularly large language models (LLMs) such as ChatGPT. These models go beyond mere linguistic processing; they generate complex outputs ranging from literature reviews and hypothesis generation to the simulation of analytical reasoning patterns found within specialized disciplines. This new reality calls for a reassessment of the epistemic roles these models can occupy—whether they should be seen merely as assistive tools or as participants in the production of knowledge.
Within this context, a philosophical and methodological question arises—one that is increasingly relevant at the intersection of technology and critical thought, demanding reflection on the boundaries of agency, subjectivity, and academic accountability.
AI language models develop social norms like groups of people
When LLMs are grouped together, they exhibit similar characteristics to human societies...
"Groups of large language models (LLMs) playing simple games can develop social norms. After several rounds of a ‘game’ that paired copies of the LLM Claude together and asked them to choose a letter from the alphabet, pairs began selecting the same letter. This suggests that the LLMs developed a collective bias — a type of social norm. The formation of collective biases could result in harmful biases, says complexity scientist and study co-author Andrea Baronchelli, even if individual agents seem unbiased..."
https://www.nature.com/articles/d41586-025-01500-6
Thank you, Dr. Ljubomir Jacić, for this valuable contribution.
Your observation highlights how large language models (LLMs) may develop emergent collective behaviors resembling social norms. This suggests that they transcend their traditional role as mere linguistic processors and begin to participate—albeit unintentionally—in forms of distributed cognition.
This raises a profound epistemological question regarding the legitimacy of knowledge production in the presence of such models: Can the outputs generated by non-conscious systems be considered valid academic knowledge? And to what extent can we trust the reasoning and meaning-making processes of systems that lack intentionality?
The emergence of collective bias within these models urges us, as scholars, to reconsider our epistemic responsibilities and critically assess the mechanisms we increasingly rely on. We are entering a new phase that demands a redefinition of the relationship between artificial intelligence and the epistemic foundations of scientific inquiry.
With appreciation and respect, Dr. Foudil Rais University of Tebessa – Algeria
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology