Mapping the Emerging Social Science of Large Language Models

URL SCAN: Mapping the Emerging Social Science of Large Language Models
FIRST LINE: # Computer Science > Computers and Society

The Dissection

This paper builds an orderly taxonomy for an unruly technological rupture. Its three domains—LLM as Social Minds, LLM Societies, and LLM-Human Interactions—map how academia is studying model behavior, agent collectives, and human adaptation. The methodological machinery demonstrates classification stability, not social stability.

The paper is really institutionalizing a research field: turning a civilization-scale power shift into a legible set of subcategories, venues, topics, and citation patterns. It catalogs the surface manifestations of LLM expansion while leaving ownership of compute, energy, infrastructure, and automated production largely outside the frame.

The Core Fallacy

The central error, under Discontinuity Thesis mechanics, is confusing taxonomy with causal explanation. The paper organizes research about LLMs but does not test the decisive chain:

P1: durable superiority in cognitive work.
P2: inability of human institutions to preserve human-only economic domains.
P3: collapse of mass access to economically necessary labor.

Its focus on trust, work, creativity, education, games, and interaction treats these as continuing social domains. Under DT, they may instead be transitional interfaces around the severing of the employment–wage–consumption circuit. The paper studies what humans and models do together without confronting who controls the system once models can perform the economically necessary work more cheaply and at scale.

Its impressive adjusted Rand index and clustering agreement validate the map’s internal consistency. They do not validate the map’s completeness, predictive power, or relevance to systemic survival.

Hidden Assumptions

• That expanding scholarly attention indicates a manageable new field rather than the early administrative language of displacement.
• That model behavior, agent interaction, and human response are sufficient units of analysis without class ownership and control.
• That “institutional context” can be studied without directly analyzing capital concentration, labor substitution, and infrastructure monopolies.
• That work, creativity, and education remain stable categories instead of domains being progressively automated and economically devalued.
• That a taxonomy stable under resampling reflects a stable underlying reality. It only reflects stable coding and corpus structure.
• That bibliographic databases and venue citation patterns provide a representative picture of the social consequences rather than a picture of what academia can currently measure and reward.
• That coordination problems are research variables rather than evidence for P2: human institutions losing the ability to preserve human-only economic space.

Social Function

Primary classification: transition management and prestige signaling.

Secondary classification: partial truth and ideological anesthetic.

The paper is not fraudulent. Its empirical contribution may be useful for locating the academic field and identifying recurring research objects. But its abstraction performs a familiar service: it makes disruption discussable without making the power transfer explicit. “Social consequences” sounds comprehensive while omitting the ownership structure that determines who receives the gains and who becomes economically redundant.

The 78.02% share attributed to LLM-Human Interactions is especially revealing. Academia is concentrating on how humans perceive, use, and absorb the systems. That is the sociology of adaptation. The conference/journal split shows a prestige hierarchy around model minds and agent societies, but neither pattern establishes that humans retain productive necessity.

The Verdict

This is a competent map of the literature, not a map of the discontinuity. It measures how scholars are arranging the wreckage while leaving the demolition mechanism—cognitive automation, coordination failure, and productive-participation collapse—underexamined.

Under DT logic, the paper is evidence that LLMs have become a broad social operating layer. It is not evidence that the post-WWII order can survive them. The taxonomy may help manage the transition, but it does not alter the terminal mechanics.