Large Language Model Agent Personality and Response Appropriateness

About

This document presents a Critical Discourse Analysis focused on AI literacy, specifically targeting the role of metaphor and anthropomorphism in shaping public and professional understanding of generative AI. The analysis is guided by a prompt that draws from cognitive linguistics (metaphor structure-mapping) and the philosophy of social science (Robert Brown's typology of explanation). All findings and summaries below were generated from detailed system instructions provided to a large language model and should be read critically as interpretive outputs—not guarantees of factual accuracy or authorial intent.

---

Analysis Metadata

Source Title: Large Language Model Agent Personality and Response Appropriateness: Evaluation by Human Linguistic Experts, LLM-as-Judge, and Natural Language Processing Model
Source URL: https://arxiv.org/pdf/2510.23875
Model: gemini-2.5-pro
Temperature: 1.15
Tokens: input=6893, output=9390, total=16283
Source Type: article
Published: 2025-10-27
Analyzed At: 2025-11-04T14:16:58+00:00Z
Framework: metaphor
Framework Version: 4.0

---

Task 1: Metaphor and Anthropomorphism Audit

Description

For each of the major metaphorical patterns identified, this audit examines the specific language used, the frame through which the AI is being conceptualized, what human qualities are being projected onto the system, whether the metaphor is explicitly acknowledged or presented as direct description, and—most critically—what implications this framing has for trust, understanding, and policy perception.

1. Software as a Social Agent

Quote: "“Agents” as the term is widely used today refer to generative agents which are software entities that leverage generative artificial intelligence models to simulate and mimic human behaviour and responses."

• Frame: Model as a social actor
• Projection: The quality of agency, including the ability to act, behave, and respond in a social context, is mapped onto a software program.
• Acknowledgment: Partially acknowledged. The text puts "Agents" in quotes initially, but then adopts the term without qualification throughout the paper, treating it as a direct description.
• Implications: This framing primes readers to evaluate the system based on social and psychological criteria (like personality) rather than purely technical ones. It establishes the groundwork for applying human-centric evaluation methods to a non-human system, which is the core premise of the paper.

2. Prompt Engineering as Humanization

Quote: "One way to humanise an agent is to give it a task-congruent personality."

• Frame: System configuration as imparting humanity
• Projection: The process of providing instructional prompts to a model is equated with the complex, emergent process of a person becoming 'human' in a social and psychological sense. It projects the idea of imbuing a soul or human essence.
• Acknowledgment: Presented as direct description. The term 'humanise' is used unironically to describe the act of adding a personality-based prompt.
• Implications: This metaphor dramatically overstates the capability of prompt engineering, suggesting it creates a deeper, more fundamental change in the system's nature rather than merely constraining its stylistic output. It fosters an illusion of sentience and deep alignment with human qualities.

3. Model Processing as Cognition

Quote: "This highlights a fundamental challenge in truly aligning LLM cognition with the complexities of human understanding."

• Frame: Computation as thinking
• Projection: The internal, mathematical processes of a large language model (token prediction, attention weighting) are mapped onto the human cognitive faculties of 'cognition' and 'understanding.'
• Acknowledgment: Presented as direct description. 'LLM cognition' is used as a technical-sounding term, masking the deep metaphorical leap it represents.
• Implications: This language legitimizes the idea that LLMs 'think' in a way analogous to humans. It obscures the profound differences between statistical pattern matching and biological consciousness, potentially leading to miscalibrated trust and overestimation of the model's reasoning capabilities.

4. Model Limitations as Cognitive Deficits

Quote: "This includes queries involving imaginative, introspective, or highly nuanced concepts like anaphora or socio-cultural context, which are currently beyond the agent's cognitive grasp."

• Frame: System failure as a mental limitation
• Projection: The inability of a model to correctly process a query is framed as a lack of 'cognitive grasp,' a metaphor for mental comprehension or reach.
• Acknowledgment: Presented as direct description. This framing treats the model as a mind with limitations, rather than a tool with specific operational boundaries.
• Implications: This implies that the model's failures are like those of a developing mind that could eventually 'grasp' these concepts. It obscures the possibility that these are fundamental architectural limitations of current LLMs, framing them instead as temporary developmental hurdles.

5. LLM Evaluation as Judicial Judgment

Quote: "This method involves evaluating the current LLM responses by using another LLM as a 'Judge'."

• Frame: Automated evaluation as legal adjudication
• Projection: The process of one model scoring another's output based on a prompt is mapped onto the human institution of a judge, which implies wisdom, impartiality, and deep reasoning.
• Acknowledgment: Acknowledged. The word 'Judge' is consistently placed in scare quotes, indicating the authors recognize it as a metaphorical role.
• Implications: Despite the acknowledgment, the metaphor lends unearned authority and credibility to the evaluation process. It suggests a level of semantic and logical assessment that goes far beyond what the 'Judge LLM' (which is just another pattern-matching system) is actually doing.

6. Stylistic Consistency as Personality

Quote: "IA's introverted nature means it will offer accurate and expert response without unnecessary emotions or conversations."

• Frame: Output style as an inherent trait
• Projection: A stable, deeply integrated set of human behavioral, cognitive, and emotional patterns ('nature' or 'personality') is mapped onto a model's configured output style, which is dictated by a short instructional prompt.
• Acknowledgment: Presented as direct description. The paper treats the 'introverted nature' as a causal explanation for the model's behavior.
• Implications: This is the core illusion of the paper. It reifies a superficial stylistic constraint as a deep, internal characteristic, leading to the misleading conclusion that one is actually 'measuring' a personality rather than assessing prompt adherence.

Task 2: Source-Target Mapping

Description

For each key metaphor identified in Task 1, this section provides a detailed structure-mapping analysis. The goal is to examine how the relational structure of a familiar "source domain" (the concrete concept we understand) is projected onto a less familiar "target domain" (the AI system). By restating each quote and analyzing the mapping carefully, we can see precisely what assumptions the metaphor invites and what it conceals.

Mapping 1: Humanization (the process of making something human) to LLM Prompt Engineering

Quote: "One way to humanise an agent is to give it a task-congruent personality."

• Source Domain: Humanization (the process of making something human)
• Target Domain: LLM Prompt Engineering
• Mapping: The source domain implies a profound transformation, imbuing an object with human qualities like empathy, consciousness, or social awareness. This structure is mapped onto the target domain of writing an instruction (a prompt) for a software program, suggesting that the prompt transforms the program's fundamental nature.
• What Is Concealed: This mapping conceals that prompt engineering does not change the model's architecture, training, or core functionality. It only constrains the statistically likely outputs to a specific style. It hides the mechanical reality of stylistic filtering behind the magical language of 'humanisation.'

Mapping 2: Human Cognition and Understanding to LLM's internal data processing

Quote: "This highlights a fundamental challenge in truly aligning LLM cognition with the complexities of human understanding."

• Source Domain: Human Cognition and Understanding
• Target Domain: LLM's internal data processing
• Mapping: The structure of human cognition—involving consciousness, reasoning, semantic grounding, and world models—is projected onto the LLM's process of calculating probabilities for token sequences. It invites the inference that an LLM 'understands' a concept in the same way a person does.
• What Is Concealed: It conceals the fundamental difference between statistical correlation and causal understanding. It hides the fact that the LLM has no access to embodied experience, sensory input, or the real-world referents for the words it manipulates. The term 'LLM cognition' masks the purely computational, non-conscious nature of the system.

Mapping 3: Mental Grasp (Comprehension) to Model's processing limitations

Quote: "This includes queries...which are currently beyond the agent's cognitive grasp."

• Source Domain: Mental Grasp (Comprehension)
• Target Domain: Model's processing limitations
• Mapping: The human experience of struggling to understand a difficult concept ('grasping' it) is mapped onto the model's failure to generate a coherent or accurate response. It implies an active attempt at understanding that fails, just as a human's might.
• What Is Concealed: It conceals the mechanistic reality of the failure. The model isn't 'trying to grasp' anything. The input query simply does not map well onto the high-dimensional patterns in its training data, leading to a low-quality or nonsensical output sequence. It frames a statistical failure as a cognitive one.

Mapping 4: A Human Judge (in a legal or expert context) to An LLM (Gemini) performing a classification task

Quote: "You are an intelligent and unbiased judge in personality detection with expertise with the Big five personality model."

• Source Domain: A Human Judge (in a legal or expert context)
• Target Domain: An LLM (Gemini) performing a classification task
• Mapping: The relational structure of a judge—possessing expertise, applying rules impartially, reasoning about evidence, and delivering a verdict—is mapped onto the LLM. The LLM is instructed to 'act as' a judge, implying it will perform these complex cognitive actions.
• What Is Concealed: This conceals that the LLM is not reasoning but is generating text that mimics the language of judicial reasoning based on patterns in its training data. It has no actual 'expertise' or 'unbiased' quality; it is a biased system performing pattern matching based on the prompt's instructions. It hides the probabilistic mechanism under a cloak of authoritative reason.

Mapping 5: Human Personality Traits ('nature') to Stylistic constraints from a system prompt

Quote: "IA's introverted nature means it will offer accurate and expert response..."

• Source Domain: Human Personality Traits ('nature')
• Target Domain: Stylistic constraints from a system prompt
• Mapping: The source domain implies that an internal, stable, and causal trait ('introverted nature') dictates external behavior. This causal structure is mapped onto the LLM, suggesting an internal 'nature' is causing its concise responses. The prompt 'Tone: Conversational, Introverted Personality' is framed as the installation of this nature.
• What Is Concealed: This mapping conceals that there is no internal 'nature.' The model's output is a direct, mechanistic consequence of the system prompt conditioning its next-token predictions. The causality is external (the prompt) not internal (a personality). It hides the simplicity of the mechanism behind the complexity of the metaphor.

Task 3: Explanation Audit (The Rhetorical Framing of "Why" vs. "How")

Description

This section audits the text's explanatory strategy, focusing on a critical distinction: the slippage between "how" and "why." Based on Robert Brown's typology of explanation, this analysis identifies whether the text explains AI mechanistically (a functional "how it works") or agentially (an intentional "why it wants something"). The core of this task is to expose how this "illusion of mind" is constructed by the rhetorical framing of the explanation itself, and what impact this has on the audience's perception of AI agency.

Explanation 1

Quote: "IA's introverted nature means it will offer accurate and expert response without unnecessary emotions or conversations."

• Explanation Types:
  • Dispositional: Attributes tendencies or habits such as inclined or tends to, subsumes actions under propensities rather than momentary intentions.
  • Intentional: Refers to goals or purposes and presupposes deliberate design, used when the purpose of an act is puzzling.
• Analysis: This is a purely agential ('why') explanation. It attributes the model's output style (concise, non-emotional) to an internal 'introverted nature.' This explanation completely obscures the mechanistic 'how': the model's output is shaped this way because its system prompt contains the explicit instruction 'Tone: ... Introverted Personality.' The slippage here is from describing the prompt to describing the agent's essence, treating the instruction as an internalized trait.
• Rhetorical Impact: This framing makes the 'agent' seem more autonomous and human-like. For the audience, it reinforces the belief that the system possesses a genuine personality, making the research goal of 'assessing' this personality seem valid and meaningful, rather than simply testing for prompt adherence.

Explanation 2

Quote: "Langchain's retrieval mechanism is powered by the Retrieval Augmented Generation (RAG) technique [31]. It uses a retrieval chain with a retriever to fetch relevant documents based on the user's query and chat history. A document chain then sends these documents, along with the query and conversational context, to the LLM."

• Explanation Types:
  • Functional: Explains a behavior by its role in a self-regulating system that persists via feedback, independent of conscious design.
  • Theoretical: Embeds behavior in a deductive or model-based framework, may invoke unobservable mechanisms such as latent variables or attention dynamics.
• Analysis: This is a purely mechanistic ('how') explanation. It describes a technical process, breaking down the RAG system into its functional components (retriever, document chain) and their interactions. There is no hint of agency or intention; the system is framed as a set of interacting software modules executing a defined procedure. This stands in stark contrast to the agential language used elsewhere.
• Rhetorical Impact: This passage grounds the paper in technical credibility. By demonstrating a clear 'how' for the information retrieval part of the system, it lends an air of scientific rigor that can then be rhetorically transferred to the much softer, more metaphorical claims about 'personality' and 'cognition.' It separates the 'plumbing' (mechanistic) from the 'persona' (agential).

Explanation 3

Quote: "The personality markers in the conversation are required to be maintained so as to ensure consistency in interactions and to leverage the naturalistic speech arising from generative capabilities of the LLM-based agent."

• Explanation Types:
  • Functional: Explains a behavior by its role in a self-regulating system that persists via feedback, independent of conscious design.
• Analysis: This explanation is a hybrid, but leans agential. It presents a functional reason ('how') for maintaining personality markers—to ensure consistency. However, it frames this within an agential context by using phrases like 'naturalistic speech' and 'LLM-based agent.' The 'why' is to create a better user experience by simulating a consistent human. It subtly shifts from a technical goal (output consistency) to a social one (believable interaction).
• Rhetorical Impact: This justification frames the pursuit of 'personality' as a user-centric design principle. It makes the anthropomorphic project seem practical and necessary for the system to function effectively in a social context, thus normalizing the idea of attributing personality to a machine.

Explanation 4

Quote: "This observation that both agents are indicated as introverted is strongly explained by the fact that the transformer model used is trained on the PANDORA dataset [40] which is a dataset of Reddit comments of 10k users. The dataset is unbalanced with number of extrovert users (1920) much lower than introvert (7134)."

• Explanation Types:
  • Genetic: Traces origin or development through a dated sequence of events or stages, showing how something came to be.
  • Empirical Generalization (Law): Subsumes events under timeless statistical regularities, emphasizes non-temporal associations rather than dated processes.
• Analysis: This is a clear mechanistic ('how') explanation. It explains an observed output (bias towards introversion) by tracing it back to a specific property of its training data—the genetic origin of its statistical biases. It frames the model's behavior not as a choice or disposition, but as a statistical artifact of its development process. It is one of the few moments where the illusion of agency is explicitly broken down.
• Rhetorical Impact: This explanation demonstrates critical analysis and adds to the paper's scientific credibility. However, it also contains a contradiction: if the model's 'personality' output is merely an artifact of training data bias, it undermines the entire premise that a prompted 'personality' can be meaningfully instilled and assessed. The authors present this as a methodological problem to be solved, rather than a fundamental challenge to their conceptual framework.

Explanation 5

Quote: "For this study, the poetry agents are classified into two different poetry expert agents - Introvert Agent (IA) and Extrovert Agent (EA) trained on the specific poem “Dover Beach” given as contextual document. The personality of both the agents are inculcated using the technique of Prompt Engineering."

• Explanation Types:
  • Genetic: Traces origin or development through a dated sequence of events or stages, showing how something came to be.
  • Intentional: Refers to goals or purposes and presupposes deliberate design, used when the purpose of an act is puzzling.
• Analysis: This is a hybrid explanation that masterfully slips from 'how' to 'why.' The 'how' is 'using the technique of Prompt Engineering.' This is a mechanistic description. But the word 'inculcated' shifts the frame to agency. 'Inculcate' means to instill an idea or habit by persistent instruction. This anthropomorphic verb frames the mechanistic process of prompt engineering as a form of teaching or deep imprinting, creating the 'why' (to give it a personality) from the 'how' (to give it a system prompt).
• Rhetorical Impact: The use of 'inculcated' makes the process of prompt engineering sound more profound and transformative than it is. It subtly elevates a simple configuration step into a form of psychological conditioning, making the resulting system behavior seem like a deeply embedded trait rather than a superficial stylistic layer.

Task 4: AI Literacy in Practice: Reframing Anthropomorphic Language

Description

Moving from critique to constructive practice, this task demonstrates applied AI literacy. It selects the most impactful anthropomorphic quotes identified in the analysis and provides a reframed explanation for each. The goal is to rewrite the concept to be more accurate, focusing on the mechanistic processes (e.g., statistical pattern matching, token prediction) rather than the misleading agential language, thereby providing examples of how to communicate about these systems less anthropomorphically.

Original Quote	Mechanistic Reframing
"One way to humanise an agent is to give it a task-congruent personality."	To create a more human-like user experience, a system prompt can be engineered to constrain the model's output to a specific, consistent conversational style designated as its 'personality'.
"IA's introverted nature means it will offer accurate and expert response without unnecessary emotions or conversations."	The system prompt for the 'Introvert Agent' configuration instructs the model to generate concise, formal responses, which results in output that omits conversational filler and emotive language.
"This highlights a fundamental challenge in truly aligning LLM cognition with the complexities of human understanding."	This highlights a fundamental challenge in mapping the statistical patterns generated by an LLM to the grounded, semantic meanings that constitute human understanding.
"The agent has the capability to maintain the chat history to provide contextual continuity, enabling the agent to generate coherent, human-like and meaningful responses."	The system architecture includes a context window that appends previous turns from the conversation to the prompt, enabling the model to generate responses that are textually coherent with the preceding dialogue.
"The agent simply needs to locate and present the information."	For these questions, the system's task is to execute a retrieval query on the provided text and synthesize the located information into a generated answer.
"The personality of both the agents are inculcated using the technique of Prompt Engineering."	The designated personality styles for each agent are implemented through specific instructional text included in their respective system prompts.

Critical Observations

Description

This section synthesizes the findings from the previous tasks into a set of critical observations. It examines the macro-patterns of agency slippage (the shift between treating AI as a tool vs. an agent), how cognitive metaphors drive trust or fear, and what actual technical processes are obscured by the text's dominant linguistic habits.

Agency Slippage

The paper exhibits a systematic and strategic oscillation between mechanistic and agential framing, a pattern crucial to its rhetorical success. The primary direction of this slippage is mechanical-to-agential, serving to build a seemingly rigorous foundation for what are ultimately anthropomorphic claims. The text begins by describing the LLM 'agent' mechanistically as a 'software entity' built with the 'Langchain framework' and 'Retrieval Augmented Generation'. This section (3.1.1) is dense with technical jargon (vector stores, embedding models), establishing the authors' credibility within a computer science paradigm. However, once this foundation is laid, the text pivots sharply. The process of providing a system prompt is not described as 'configuring an output filter' but as 'humanising an agent' and 'inculcating' a personality. The model’s failure modes are not 'output errors' but limitations of its 'cognitive grasp.' This slippage is most pronounced at the boundaries between methodology and interpretation. The description of the RAG system is purely mechanical ('how'), but the explanation for its use is agential ('why'—to enable the 'expert' agent to respond). This oscillation serves a critical function: it uses the language of mechanism to build credibility and the language of agency to create significance. Without the mechanical framing, the paper would lack scientific rigor. Without the agential framing, the central concept—'agent personality'—would collapse into the more mundane reality of 'stylistic prompt adherence,' making the research far less novel or compelling. This ambiguity benefits the authors by allowing them to operate in two registers at once, satisfying technical reviewers with concrete implementation details while engaging a broader audience with the more exciting, human-like narrative of intelligent agents.

Metaphor-Driven Trust

The credibility of this paper's central claim—that an LLM's 'personality' can be assessed—is built almost entirely on metaphor-driven trust transfer, bypassing direct argumentation for its feasibility. The most powerful metaphors are 'agent,' 'expert,' and 'cognition,' which borrow authority from social psychology, professional domains, and cognitive science, respectively. By labeling the system an 'agent,' the authors immediately frame it as a social actor, making the application of personality theory feel natural rather than absurd. The term 'expert' then elevates this agent from a mere conversationalist to a repository of knowledge, encouraging trust in its outputs. When the 'poetry expert agent' responds, the user is primed to receive not just a string of statistically probable text, but advice from a knowledgeable entity. The most subtle and powerful transfer comes from 'LLM cognition.' This metaphor recasts the model's opaque statistical processing as a familiar form of 'thinking.' This makes its capabilities seem intuitive and its failures understandable, much like a human student who has not yet grasped a concept. This framing makes the counterintuitive claim that a machine has a 'personality' feel believable because it is presented as an extension of its 'cognition.' These metaphors are most credible to a non-technical audience or researchers outside of core AI/ML, who may take these terms at face value. They activate pre-existing beliefs about intelligence and personality, making the LLM seem like a new kind of mind. The trust created is a vulnerability; it encourages users and researchers to attribute understanding where there is none, potentially leading to over-reliance on the system's outputs and a misinterpretation of its limitations as developmental flaws rather than fundamental architectural constraints.

Obscured Mechanics

The pervasive use of anthropomorphic metaphors systematically conceals the mechanical and statistical reality of the LLM-based system, masking key aspects of its operation and construction. The most significant obscured reality is the primacy of prompt engineering. By framing personality as an 'inculcated' trait of an 'agent,' the text hides that the observed behavior is a brittle and superficial adherence to an explicit instruction in a system prompt. The metaphor of 'personality' implies a deep, stable, internal state, concealing the fact that a minor change to the prompt could completely invert the 'personality,' or that it may fail to generalize to contexts not anticipated by the prompt engineer. This framing actively prevents the reader from asking more precise questions, such as 'How robust is this stylistic consistency across different topics?' or 'What specific phrases in the prompt trigger this behavior?' Secondly, the language of 'cognition' and 'understanding' obscures the system's reliance on training data. The paper discusses training data bias as a confounder (the PANDORA dataset example) but does not frame it as central to the 'agent's' entire world model. The 'agent' doesn't 'know' about poetry; its training data contains a vast corpus of text about poetry, from which it generates statistically likely sequences. The metaphor hides the immense human labor of data creation and curation that underpins the entire system. Finally, the focus on the 'agent' conceals the vast computational and energy costs required for training and inference. The system is presented as a disembodied, thinking entity, which hides the material infrastructure and environmental impact of its existence. If the paper were forced to use only mechanical language—'stylistic output filtering based on prompt conditioning'—the perceived novelty of the research would evaporate, revealing that the study is not about AI personality but about methods for evaluating prompt adherence.

Context Sensitivity

The deployment of metaphor in this paper is highly context-sensitive, varying strategically across different sections to achieve specific rhetorical goals. The Abstract and Introduction are saturated with high-level anthropomorphic metaphors like 'agents,' 'personalities,' and 'human-like manner.' This language is used to frame the research problem in a way that is compelling and relatable to a broad academic audience, including those in the humanities and social sciences. It elevates the work from a mere technical exercise to an exploration of a seemingly profound interaction between humans and a new form of intelligence. However, upon entering the Methodology section (specifically 3.1, 'Design of LLM-based Agent'), the language shifts dramatically toward a mechanistic register. Here, the text details the 'Langchain framework,' 'Directory loaders,' 'vector representation,' and 'Retrieval Augmented Generation (RAG) technique.' This section deliberately avoids anthropomorphism to project technical competence and scientific rigor, assuring the reader that the high-level concepts are grounded in solid engineering. The metaphor use spikes again in the evaluation sections (4.2, 'Large Language Model as a Judge'). Here, the metaphor of the 'judge' is used to grant authority to the automated evaluation process, even while being flagged with scare quotes. The description of the prompt given to the 'Judge LLM' is revealing: 'You are an intelligent and unbiased judge...' This shows the authors are using metaphor not just to describe their system, but to construct its behavior. Capabilities are consistently described in agential terms ('the agent’s capacity to demonstrate traits'), while technical components are described mechanistically. This strategic variation reveals the text's dual purpose: to be seen as a legitimate contribution to computer science while simultaneously making ambitious, human-centric claims that capture interdisciplinary interest.

Conclusion

Description

This final section provides a comprehensive synthesis of the entire analysis. It identifies the text's dominant metaphorical patterns and explains how they construct an "illusion of mind." Most critically, it connects these linguistic choices to their tangible, material stakes—analyzing the economic, legal, regulatory, and social consequences of this discourse. It concludes by reflecting on AI literacy as a counter-practice and outlining a path toward a more precise and responsible vocabulary for discussing AI.

Pattern Summary

The discourse within this paper is built upon a system of three interconnected, load-bearing metaphorical patterns that work in concert to construct the illusion of a mind amenable to psychological assessment. The foundational pattern is the MODEL AS SOCIAL AGENT, which recasts a software program as an actor in a social world. This initial move is what makes the entire project conceivable, as it opens the door to applying concepts from human interaction. Building directly upon this foundation is the second pattern: OUTPUT STYLE AS INTRINSIC PERSONALITY. This metaphor performs the crucial work of reifying a configurable, superficial output behavior as a deep, internal, and stable trait. It posits that a textual instruction like 'Tone: Introverted' does not merely filter the model's responses, but 'inculcates' a 'nature.' The third pattern, COMPUTATION AS COGNITION, serves as the rationalizing framework for the first two. It provides a seemingly scientific explanation for the agent's behavior, suggesting that its ability to adopt a personality stems from an underlying 'LLM cognition' that is analogous to 'human understanding.' These patterns are not independent; they are a tightly woven logical chain. The AGENT frame provides a subject to which a PERSONALITY can be attributed, and the COGNITION frame provides a mechanism to make that attribution seem plausible. If the foundational 'AGENT' metaphor were removed and replaced with 'text-generation tool,' the notion of 'personality' would lose its subject and collapse into 'stylistic setting,' and 'cognition' would revert to 'processing,' dismantling the paper's entire conceptual edifice.

Mechanism of Illusion: The "Illusion of Mind"

The rhetorical architecture of this text constructs the 'illusion of mind' through a subtle, multi-stage process of conceptual framing and presupposition. The central sleight-of-hand is not a single claim, but the strategic sequencing of its discursive moves. The process begins by establishing the term 'agent' as a neutral technical descriptor for a 'software entity,' borrowing from its established use in computer science. This initial move is critical as it smuggles in connotations of autonomy and action under the guise of standard terminology. Having established the 'agent' as the object of study, the text then performs its key maneuver: it frames the research problem as one of assessment and evaluation ('effectively assessing their personalities has proven challenging'). This is a classic persuasive technique; by focusing on the challenge of measurement, it presupposes the existence and validity of the thing being measured. The reader is invited to worry about how to evaluate an LLM's personality, a question which distracts from the more fundamental and unasked question: does an LLM have a personality to begin with? The subsequent methodology, involving 'Judge LLMs' and 'human linguistic experts,' further solidifies this illusion. It constructs an elaborate apparatus of evaluation that lends a veneer of scientific objectivity to the process. The audience's cognitive vulnerability lies in the intuitive appeal of the personality metaphor; we are naturally inclined to anthropomorphize complex systems that exhibit human-like communication. The paper exploits this by providing a seemingly rigorous, scientific framework that validates this intuitive impulse, allowing the reader to accept the illusion not as a folk belief, but as a research-backed finding.

Material Stakes

• Selected Categories: Epistemic, Economic, Regulatory
• Analysis: The metaphorical framing in this paper has concrete, tangible consequences across multiple domains. For the Epistemic domain, the primary stake is the misdirection of research effort and the corruption of scientific concepts. Framing prompt adherence as 'personality' creates a conceptual muddle that wastes resources on developing 'better personality tests for AI' instead of more robust methods for analyzing and controlling stylistic output. It encourages a generation of researchers to chase a ghost in the machine, applying psychological tools to a subject for which they are fundamentally unsuited, thereby degrading the precision of concepts like 'personality' and 'cognition.' In the Economic domain, the 'agent with personality' frame is a powerful marketing tool. It allows companies to sell products like AI companions, tutors, and customer service bots by implying a level of social intelligence and reliability that does not exist. This research, by providing a 'scientific' method for 'evaluating' these personalities, lends legitimacy to these marketing claims. This can mislead investors into overvaluing the technology and consumers into placing undue trust in these systems, creating risks of manipulation or disappointment. For example, a firm might invest heavily in an 'empathetic AI therapist' based on research that 'validates' its personality, without understanding its underlying mechanistic and non-sentient nature. Finally, in the Regulatory/Legal domain, this language creates profound ambiguity regarding liability and accountability. If a system is an 'agent' that 'behaves,' who is responsible when it causes harm? The metaphor of agency subtly shifts the burden of responsibility away from the developers and corporations who build and deploy these systems. It opens a path to legal frameworks that might treat the AI as a quasi-autonomous entity, making it harder to hold its human creators accountable for its outputs, a situation that directly benefits the tech industry at the expense of public safety and consumer protection.

Literacy as Counter-Practice: AI Language Literacy

Practicing AI literacy as a counter-measure involves the disciplined substitution of precise, mechanistic language for misleading anthropomorphic metaphors, a practice that directly threatens the material stakes established by the current discourse. The reframing exercises demonstrate this principle: replacing 'the agent's introverted nature' with 'the model's output as constrained by a system prompt' is not merely a semantic tweak; it is a political act. This precise phrasing directly counters the epistemic confusion by re-centering the locus of causality from a non-existent internal 'nature' to an external, engineered artifact—the prompt. It transforms the research object from a psychological mystery into a technical problem of prompt engineering. This, in turn, undermines the economic hype. A product sold as having a 'prompt-adherent stylistic filter' is far less appealing and commands a lower market value than one advertised as possessing an 'inculcated personality.' The practice of precision thus acts as a deflationary force against market bubbles driven by anthropomorphic narratives. This practice would be met with significant resistance. Researchers might resist because the 'personality' frame is more likely to secure funding and high-impact publications. Companies would resist because it weakens their marketing narratives. Adopting precision carries professional costs; it requires sacrificing the 'wow factor' for accuracy and potentially being seen as less innovative. Therefore, AI literacy is not just about better communication; it is a commitment to an intellectual and ethical stance that prioritizes scientific truth over compelling fiction, directly challenging the economic and institutional interests that benefit from maintaining the 'illusion of mind.'

Path Forward

To foster a more responsible and transparent discourse in the field of human-AI interaction research, the community must move beyond critiquing metaphors and actively adopt a more precise technical vocabulary. Instead of 'agent personality,' researchers should use terms that describe the mechanism, not the illusion, such as 'prompt-induced persona,' 'stylistic alignment,' or 'behavioral scripting.' Where 'cognition' is used, a more accurate alternative would be 'large-scale pattern inference,' and instead of 'understanding,' the phrase 'semantic representation in vector space' offers more precision. This vocabulary shift is not about pedantry; it is about building a science on a solid foundation. Adopting this language would enable researchers to formulate more rigorous and testable hypotheses about how to control and predict model behavior, rather than getting lost in unprovable claims about a machine's internal state. To support this shift, academic institutions and journals must implement structural changes. Journals in HCI and AI could require authors to include a 'Justification of Anthropomorphism' section, compelling them to either defend their use of agential terms on theoretical grounds or explicitly state they are being used as shorthand analogies. Funding agencies should prioritize proposals that focus on the mechanistic underpinnings of model behavior and user perception over those that take anthropomorphic framings at face value. While some might argue that this would strip the field of its intuitive appeal and interdisciplinary reach, the gain in clarity and scientific integrity would be immense. The ultimate goal is to create a field that can engineer sophisticated and beneficial human-computer interactions without deceiving ourselves or the public about the nature of the technology we are creating, enabling a future of genuine progress rather than one built on a compelling but dangerous illusion.

---

info

Raw JSON: 2025-11-04-large-language-model-agent-personality-andresponse-metaphor-c6363b.json Analysis Framework: metaphor v4 Generated: 2025-11-04T14:16:58+00:00Z

Discourse Depot © 2025 by TD is licensed under CC BY-NC-SA 4.0

Extended Processing Summary

Gemini refers to the text below as “thought summaries.” Let’s just call it a model-generated process trace. Treat this as a just anotherrhetorical artifact—a way of making the model’s processing legible. The first-person framing of these “thought summaries” is a presentation choice for the user-facing output, not a window into “real”thoughts. These are computational artifacts, not cognitive reports from a quirky, curious or conflicted mind.

Examining the LLM Paper

Okay, I'm digging into that research paper on LLM agents. My focus right now is to structure the critical discourse analysis into four distinct tasks, followed by a concluding synthesis, all while adhering to the strict JSON format. I must make sure that all of the information from the research paper is accounted for in the analysis.

Structuring the Discourse Analysis

I'm making progress on structuring the critical discourse analysis. The JSON structure is in place, and I've broken down the tasks: first, auditing for metaphor and anthropomorphism, then analyzing source-target mappings. I'm focusing on those instances in the paper that fit the task, specifically "agent," "humanise an agent," and "LLM cognition." The JSON output must be valid.

Defining Explanatory Frameworks

I'm now focusing on how the paper explains LLM agent behavior. I'm applying Brown's typology of explanations, looking for instances of genetic, functional, empirical, theoretical, intentional, dispositional, and reason-based explanations. I'm also analyzing how these explanations frame the AI, either as a machine or an agent, which I'll report in the JSON format. The key is to assess the rhetorical impact.

Developing Analytical Tasks

I've outlined four tasks to analyze the paper on LLM agents. These include an audit of metaphors and anthropomorphism, source-target mapping analysis, an explanation audit using Brown's typology, and reframing anthropomorphic language. I'm focusing on creating a valid JSON output that reflects this structure and the requested analysis of the paper.

Developing a Detailed Framework

I'm now fully immersed in developing the detailed framework for the critical discourse analysis. My focus is now on the distinct tasks. I've broken the paper's analysis into the 4 tasks I outlined, and I've started the synthesis. I'm taking the paper, "LARGE LANGUAGE MODEL AG ENT PERSONALITY AND RESPONSE APPROPRIATENESS: EVALUATION BY HUMAN LINGUISTIC EXPERTS, LLM-AS -JUDGE, AND NATURAL LANGUAGE PROCESSING MODEL" as a case study. I will use it to complete the tasks, and the concluding synthesis will address those specific concerns.

Finalizing Analysis Framework

I'm now putting the finishing touches on the analysis. I've developed the four primary tasks – auditing metaphors, mapping source-target domains, conducting an explanation audit using Brown's typology, and reframing anthropomorphic language – and the concluding synthesis is taking shape. I'm nearly done extracting and organizing the relevant quotes and passages from the "LARGE LANGUAGE MODEL AG ENT PERSONALITY AND RESPONSE APPROPRIATENESS" paper, preparing for the JSON output. I'm focusing on the rhetorical impact of language and the implications of AI literacy.

Developing Deeper Insights

I'm synthesizing the core themes. I've noted the foundational metaphor of "Model as Agent," the illusion of mind constructed through humanization, and the economic stakes of this framing. Now, I'm focusing on the rhetorical architecture, seeing how this framing is reinforced through evaluation. It will be useful to look at the paper's specific language.

Synthesizing Foundational Metaphors

I'm now zeroing in on the core metaphors, namely "Model as Agent," which acts as the foundation, allowing for the projection of concepts like "Cognition as Computation" and "Personality as Configurable Style." Analyzing the rhetorical architecture shows how these frameworks are reinforced within the evaluation process, and highlighting the material stakes of this framing.

Drafting JSON Structure

I'm now fully immersed in developing the JSON structure, the final task. I'm ready to produce the JSON output, which reflects the tasks I previously laid out. I need to make sure the quotes are accurate and every field in the schema is filled.