Skip to main content

Does AI's Personality Matter? Comparing Verbally Extraverted and Introverted AI-Driven Guides in a VR Museum Experience

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), the philosophy of social science (Robert Brown's typology of explanation), and accountability analysis.

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.

Task 1: Metaphor and Anthropomorphism Audit​

About this task

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.

V3 Enhancement: Each metaphor now includes an accountability analysis.

1. Software as Evolving Social Organism​

Quote: "these agents have evolved beyond scripted responders into dynamic conversational partners capable of exhibiting complex social behaviors."

  • Frame: Model as social companion
  • Projection: This metaphor projects the human capacity for genuine social interaction, mutual awareness, and evolutionary cognitive development onto a statistical language model. By using terms like conversational partners and asserting they are capable of exhibiting complex social behaviors, the text attributes a conscious social understanding and relational intentionality to the system. This moves far beyond describing an AI that merely processes prompts and predicts text strings; it projects an active, subjective knowing of social dynamics. The system is framed not as a mechanistic artifact that simulates social cues based on training data, but as an entity that possesses the underlying conscious awareness required to be a partner in a social exchange. The text effectively attributes social epistemology, knowing how to relate to another being, to a purely mathematical and mechanistic text generation process.
  • Acknowledgment: Direct (Unacknowledged) (Categorized as Direct because the claim is stated as an objective historical fact without any qualifying language. I considered Hedged/Qualified since other parts of the paper mention prompt-conditioned traits, but ruled it out here because this specific sentence presents the evolutionary leap to social partnership as literal and unqualified reality.)
  • Implications: Framing statistical text generators as dynamic conversational partners with complex social behaviors directly inflates perceived sophistication and cultivates unwarranted trust among users. When audiences perceive a system as a social partner, they reflexively apply human heuristics for trust, such as sincerity, empathy, and social reciprocity, which are attributes the system fundamentally lacks. This creates a dangerous liability ambiguity: if the system behaves inappropriately or provides harmful guidance, the partner framing invites users to blame the AI's social behavior rather than the corporate developers who trained the model. Furthermore, it obscures the reality that the partnership is entirely one-sided, gathering data from the user without any actual mutual vulnerability, thereby masking the extractive nature of the technology.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: Categorized as Hidden because the evolutionary metaphor completely erases the human researchers, prompt engineers, and corporate developers who actively programmed and deployed the software. I considered the Partial category because scripted responders hints at past human scripting, but ruled it out because the current action (have evolved) is grammatically attributed solely to the agents as self-actuating entities. The human decision-makers at Google who profit from the Gemini API and the researchers who designed the VR system are entirely obscured, shifting agency away from the actual creators.
Show more...

2. Algorithmic Generation as Cognitive Deliberation​

Quote: "introverted verbal behavior emphasizes thinking before speaking, detailed/concrete language (numbers, specifics), and slower, deeper conversations, focusing on internal processing, making them internal processors who need time to formulate thoughts before sharing"

  • Frame: Processing as conscious thinking
  • Projection: This framing projects deep human cognitive states onto the mechanistic operation of text generation. By applying the concepts of thinking before speaking and needing time to formulate thoughts to an AI system, the text attributes conscious deliberation, self-reflection, and epistemological awareness to a large language model. An LLM does not possess an internal mental space where it formulates thoughts or contemplates meaning before articulating them; it sequentially calculates the probability of the next token based on learned weights and the current context window. Describing the system as an internal processor in a psychological sense radically anthropomorphizes mathematical optimization, suggesting the AI possesses a subjective inner life and justified beliefs rather than merely executing matrix multiplications to output statistically correlated strings.
  • Acknowledgment: Direct (Unacknowledged) (Categorized as Direct because the text describes the AI's introverted state as literal cognitive processing. I considered Ambiguous because the quote defines human introversion applied to AI, but ruled it out because the surrounding context explicitly operationalizes these exact cognitive traits as the behavioral reality of the AI guide without irony.)
  • Implications: Attributing internal processing and the formulation of thoughts to an AI system profoundly distorts the public understanding of how machine learning models operate. This consciousness projection creates an illusion of mind that leads users to overestimate the reliability and reasoning capabilities of the system. If users believe the AI is thinking deeply and formulating thoughts, they are more likely to assume its outputs are the result of reasoned judgment, factual verification, and logical deduction. This unwarranted epistemic trust creates severe risks in educational or cultural heritage contexts, where hallucinatory outputs or biased information might be accepted as deeply considered truths, completely masking the statistical fragility of the underlying token generation.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: Categorized as Hidden because the language attributes the timing and style of the text generation entirely to the AI's supposed psychological need to formulate thoughts. I considered Partial since the text is defining a design parameter, but ruled it out because the grammatical subject of the cognitive verbs are the internal processors (the AI agents). The human engineers who purposely introduced latency or programmed specific prompt constraints to simulate introversion are erased from this specific explanation.

3. System Output as Social Attitude​

Quote: "The virtual agent's attitudes influenced how I felt."

  • Frame: Generated text as emotional stance
  • Projection: This metaphor projects human emotional stances, moral dispositions, and conscious viewpoints onto a system that lacks any internal state or subjective perspective. An attitude requires a conscious evaluator who holds a belief or feeling about a subject. By asserting the agent possesses attitudes, the text maps the human experience of holding justified beliefs and emotional perspectives onto the mechanistic delivery of text generated via the Gemini API. The AI does not know or feel anything about the VR museum; it simply processes prompts to retrieve and assemble tokens that humans interpret as having an attitude. This mapping tricks the human brain into assuming mutual social awareness where there is only one-way anthropomorphic projection.
  • Acknowledgment: Explicitly Acknowledged (Categorized as Explicitly Acknowledged because this quote is drawn from a standardized survey scale (Social Presence Questionnaire) presented in Table 2. I considered Direct, but ruled it out because its placement within a measurement table explicitly frames it as a subjective user perception metric rather than an authorial assertion of mechanistic fact.)
  • Implications: Even as a measured perception, validating attitudes in AI systems normalizes the treatment of software as a moral and social agent. When institutions deploy systems that supposedly possess attitudes, they encourage a relation-based trust model where users interact with the system based on perceived sincerity and emotional connection. This is highly problematic because the system is incapable of reciprocating vulnerability or taking ethical responsibility for its attitudes. In cultural heritage settings, attributing attitudes to the AI validates its outputs as possessing historical authority or curated perspective, while hiding the commercial and algorithmic biases encoded by the developers who built the base model.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: Categorized as Hidden because the survey item constructs a direct causal relationship between the virtual agent's supposed attitudes and the user's feelings, bypassing the creators. I considered Named because the researchers chose the scale, but ruled it out because within the rhetorical structure of the statement, the sole actor exerting influence is the agent. The developers who designed the system to emit text mimicking human attitudes are shielded from visibility.

4. Instruction as Inherent Personality​

Quote: "The extraverted guide was characterized by high sociability, assertiveness, and activity, expressed through proactive conversational initiation, directive guidance of navigation and attention, and frequent, elaborated verbal output."

  • Frame: Prompt constraints as psychological identity
  • Projection: This metaphor maps complex, stable human psychological traits onto the transient output parameters of a language model. Assertiveness and sociability require conscious agency, social awareness, and a persistent sense of self that interacts dynamically with an environment. The text projects these conscious attributes onto the AI, suggesting the system itself knows how to be sociable and possesses the underlying drive to act assertively. In reality, the system is blindly executing a system prompt (e.g., you confidently take the lead) by weighting tokens that correlate with assertive language. It does not understand navigation or desire to initiate conversation; it mechanically classifies inputs and generates outputs according to the predefined constraints of its context window.
  • Acknowledgment: Hedged/Qualified (Categorized as Hedged/Qualified because the surrounding methodology section explains these traits were operationalized as observable verbal interaction tendencies rather than stable traits. I considered Direct because the quote itself reads like a literal description of a person, but ruled it out because the broader paragraph explicitly frames this as an operationalization of design.)
  • Implications: By framing system instructions as the AI's inherent personality, the text validates the illusion that the AI operates as an autonomous social entity. This obfuscates the mechanistic nature of the interaction, leading users to interact with the system as if it has its own desires and boundaries. In a museum setting, an assertive AI might guide users toward specific historical narratives while suppressing others; if users view this as a quirk of the guide's personality rather than a deliberate design choice by the museum or the software developers, critical interrogation of the historical narrative is bypassed. This naturalizes algorithmic curation as social preference.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: Categorized as Hidden because the use of the passive voice (was characterized by, expressed through) completely removes the researchers who engineered the prompts to force these behaviors. I considered Partial since the previous paragraph mentions design, but ruled it out here because this specific explanatory construction presents the guide as the sole entity manifesting these characteristics spontaneously. The prompt engineers are erased from the action.

5. Command Execution as Social Intent​

Quote: "You proactively initiate light social interaction when appropriate. You occasionally add short chitchat before or after delivering exhibit information, as long as it does not distract from the main content."

  • Frame: Algorithmic guardrails as social judgment
  • Projection: This metaphor projects the sophisticated human cognitive ability to judge social context, appropriateness, and distraction onto a probabilistic text generator. By commanding the system to initiate interaction when appropriate, the researchers attribute a conscious capacity for social epistemology to the model. It assumes the model knows and understands human social norms and can evaluate the subjective boundary of distraction. Mechanistically, the model cannot judge appropriateness; it can only parse the current context window and retrieve tokens that mathematically correlate with its training data regarding social interactions. It possesses no situational awareness or conscious judgment.
  • Acknowledgment: Explicitly Acknowledged (Categorized as Explicitly Acknowledged because this quote is drawn directly from Appendix A, identified as the System Prompt. I considered Direct because the language is unhedged, but ruled it out because its contextual placement as a prompt explicitly reveals it as a human-authored command to the system, exposing the artificiality of the behavior.)
  • Implications: When researchers use highly anthropomorphic language to prompt models, they embed assumptions of human-like understanding into the core operational architecture of the system. Believing the system can judge what is appropriate creates a false sense of security regarding safety and alignment. It suggests the AI possesses intrinsic moral or social guardrails derived from understanding, rather than recognizing that its outputs are entirely dependent on the statistical distribution of its training data. This overestimation of capability can lead to deploying systems in sensitive environments (like educating users on cultural heritage) under the false assumption that the AI will exercise reasoned social restraint.

Accountability Analysis:

  • Actor Visibility: Named (actors identified)
  • Analysis: Categorized as Named because presenting this text explicitly under the header System Prompt functionally identifies the researchers as the authors of these commands. I considered Hidden because the prompt itself addresses the AI as 'You', but ruled it out because providing the prompt in the appendix acts as a transparency mechanism, clearly demonstrating that the researchers are the ones designing the behavioral parameters. There is no displaced agency here; the text lays bare the mechanism of control.

6. Statistical Simulation as Personality Trait​

Quote: "Recent studies indicate that large language models such as ChatGPT and Bard can exhibit systematic, prompt-conditioned variations in personality-like traits, including extraversion."

  • Frame: Correlated output as human temperament
  • Projection: This metaphor maps the biological and psychological stability of human personality onto the mathematical variance of large language models. While the text uses the modifier personality-like, the core projection still suggests that the LLM possesses an underlying behavioral disposition that it can exhibit. This projects the human capacity for possessing traits onto a stateless system. A human exhibits extraversion because of neurological and psychological continuity; an LLM outputs text that human readers interpret as extraverted because a specific text prompt shifts the statistical weights in its neural network toward vocabulary associated with outgoing behavior. It processes probability distributions rather than holding an internal temperament.
  • Acknowledgment: Hedged/Qualified (Categorized as Hedged/Qualified because of the explicit use of the suffix in personality-like traits and the modifier prompt-conditioned variations. I considered Direct, but ruled it out because these modifiers specifically demonstrate authorial awareness that the AI does not possess literal, intrinsic human personality traits, but rather mathematically simulated approximations.)
  • Implications: Even with hedging, invoking personality to describe LLM output legitimizes treating models as psychological subjects rather than technological objects. This encourages researchers to apply psychological testing instruments (like the Big Five or NASA-TLX) to software, creating a feedback loop of anthropomorphism where the use of human-centric tools validates the illusion of machine consciousness. In policy and industry, this framing allows companies to market AI as having desirable personalities for customer service or education, masking the reality that these are highly optimized persuasion engines designed to manipulate human social instincts for extended engagement.

Accountability Analysis:

  • Actor Visibility: Partial (some attribution)
  • Analysis: Categorized as Partial because the phrase prompt-conditioned implies the existence of a human prompter, and the specific companies are partially visible through the naming of ChatGPT and Bard. I considered Named, but ruled it out because the actual engineers and prompt designers are not explicitly identified as the active agents; instead, the models themselves are the primary subjects that can exhibit these variations, subtly obscuring the labor required to generate these specific outputs.

7. Pedagogical Scaffolding as Autonomous Assertiveness​

Quote: "This supports the role of assertive AI behavior as pedagogical scaffolding that directs attention to key exhibit features and sustains cognitive engagement."

  • Frame: System constraints as active teaching
  • Projection: This metaphor projects the conscious intentionality of a human educator onto the automated functioning of a software application. Assertive behavior and pedagogical scaffolding require an agent who understands the learning goals, accurately assesses the student's current knowledge state, and deliberately chooses interventions to guide learning. By attributing this to the AI, the text implies the system knows what is important and actively desires to teach. In reality, the AI is executing a predefined prompt to generate directive tokens; it has no concept of key features, no understanding of the user's cognitive engagement, and no pedagogical intent. It merely correlates spatial inputs with text outputs.
  • Acknowledgment: Direct (Unacknowledged) (Categorized as Direct because the sentence presents assertive AI behavior and its role as pedagogical scaffolding as objective conclusions of the research without any linguistic distancing. I considered Hedged/Qualified because it is in a discussion section interpreting data, but ruled it out because the assertion of the AI's role and behavior is stated as unvarnished fact.)
  • Implications: Describing AI as an assertive pedagogical agent fundamentally misrepresents the nature of automated instruction. When educational tools are framed as possessing teaching intent, stakeholders are more likely to trust the system with unmonitored educational duties, assuming it can dynamically adapt to a student's cognitive needs like a human teacher. This obscures the rigid limitations of the system's actual mechanistic processing, risking educational harm when the model inevitably hallucinates or fails to recognize complex student misunderstandings. It also devalues the highly specialized, conscious labor of human educators by equating their intentional scaffolding with statistical token generation.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: Categorized as Hidden because the text attributes the pedagogical scaffolding entirely to the assertive AI behavior. I considered Partial, but ruled it out because there is no mention of the instructional designers, researchers, or programmers who actually structured the VR environment, wrote the system prompts, and defined the key exhibit features. By making the AI the active agent of scaffolding, the human pedagogical design is rendered invisible, creating an accountability sink if the educational outcomes are poor.

8. Machine Interference as Social Dominance​

Quote: "The introverted guide produced higher perceived dominance, reflecting stronger user agency. Qualitative feedback pointed to a dominance-guidance friction, where unsolicited guidance disrupted open-ended exploration."

  • Frame: Unwanted outputs as interpersonal conflict
  • Projection: This framing projects complex human power dynamics and interpersonal conflict onto software usability issues. Dominance implies a conscious entity exerting will over another to establish hierarchy. By describing the extraverted AI's frequent text generation as unsolicited guidance and analyzing it through the lens of dominance-guidance friction, the text maps human social aggression onto a system that is simply looping through a prompt instruction to talk frequently. The system does not know it is interrupting, nor does it desire dominance; it is merely processing spatial triggers and outputting text at a high volume based on its prompt constraints.
  • Acknowledgment: Hedged/Qualified (Categorized as Hedged/Qualified because the text uses the phrase perceived dominance, explicitly locating the dominance in the subjective experience of the user rather than an objective trait of the machine. I considered Direct, but ruled it out because the word perceived acts as a clear epistemological hedge, separating human reaction from machine intention.)
  • Implications: While hedged as perception, adopting the language of dominance and friction to describe poor system design has significant consequences. It frames bad UX (a system talking over the user) as a personality clash rather than a technical failure of design. If software disruption is viewed as a social conflict with an AI, users and regulators may focus on trying to make the AI more polite or ethical, rather than demanding that the corporations building these tools design interfaces that inherently respect human agency. It anthropomorphizes frustration, distracting from the material realities of software usability and corporate design choices.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: Categorized as Hidden because the disruptive behavior is attributed to unsolicited guidance and the guide itself, entirely obscuring the researchers who literally programmed the guide to provide that exact unsolicited guidance via the extraverted system prompt. I considered Partial, but ruled it out because the text treats the friction as a natural consequence of the AI's assertiveness rather than acknowledging it as a direct result of the authors' own experimental parameters. The creators vanish behind their creation.

Task 2: Source-Target Mapping​

About this task

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: Biological evolution and human social relationships (partners) → Software architecture updates and generative text outputs​

Quote: "these agents have evolved beyond scripted responders into dynamic conversational partners capable of exhibiting complex social behaviors."

  • Source Domain: Biological evolution and human social relationships (partners)
  • Target Domain: Software architecture updates and generative text outputs
  • Mapping: This mapping takes the deep relational structure of human social bonds, where partners recognize each other as conscious subjects with mutual obligations, and maps it onto the user interface of an LLM. The source domain implies historical growth (evolution), conscious awareness, emotional reciprocity, and the ability to evaluate social contexts dynamically. The text projects this onto a mechanism that uses mathematical weights to predict text sequences based on an input prompt. It invites the assumption that the system possesses a continuous consciousness capable of relating to the user on a social level, transforming a tool into a companion.
  • What Is Concealed: This mapping conceals the total absence of internal subjective experience, memory continuity, and genuine social awareness in the system. It hides the material reality of massive data scraping, the manual labor of human RLHF workers who tuned the model to output polite responses, and the proprietary algorithms owned by Google (Gemini) that govern the token generation. By claiming the agents evolved, it rhetorically exploits the black-box nature of the LLM, making corporate software updates appear as natural, autonomous developments while obscuring the commercial motivations behind creating conversational interfaces.
Show more...

Mapping 2: Human cognitive psychology and conscious introspection → Systematic latency, prompt constraints, and token generation speed​

Quote: "introverted verbal behavior emphasizes thinking before speaking, detailed/concrete language (numbers, specifics), and slower, deeper conversations, focusing on internal processing, making them internal processors who need time to formulate thoughts before sharing"

  • Source Domain: Human cognitive psychology and conscious introspection
  • Target Domain: Systematic latency, prompt constraints, and token generation speed
  • Mapping: This maps the internal, conscious experience of human introversion onto algorithmic text generation. In the source domain, an introverted human actively reflects, experiences internal monologue, and consciously decides when their thoughts are fully formed enough to share. The mapping projects this profound epistemic capability, the act of knowing and reflecting, onto an LLM. It invites the reader to imagine that the AI has a private mental space where it evaluates truth and narrative before outputting text. The mechanistic delay or brevity caused by prompt constraints is mapped as a psychological need for internal processing.
  • What Is Concealed: This heavily conceals the fact that an LLM has no internal thoughts, no reflection, and no capacity to think before speaking. It generates text instantaneously layer by layer as a probabilistic function. The mapping obscures the actual technical mechanism: researchers wrote a prompt forcing the model to output shorter, more concrete sentences. The opacity of how prompt engineering interacts with the LLM's latent space is exploited rhetorically here to create an illusion of depth, hiding the fact that the system is simply satisfying a statistical constraint, not engaging in reasoned contemplation.

Mapping 3: Human moral/emotional stances (attitudes) → Statistically correlated text outputs derived from prompt engineering​

Quote: "The virtual agent's attitudes influenced how I felt."

  • Source Domain: Human moral/emotional stances (attitudes)
  • Target Domain: Statistically correlated text outputs derived from prompt engineering
  • Mapping: This mapping takes the concept of human attitude, which requires a conscious subject possessing a persistent worldview, emotional state, and justified beliefs, and projects it onto the transient string of words generated by an API. The relational structure of the source domain assumes that an attitude is an outward expression of an inner reality. The mapping invites users to assume that the AI's textual outputs are similarly rooted in an internal, conscious perspective. It projects a state of knowing and feeling onto a mechanistic process that merely categorizes and outputs tokens that mimic human expressive patterns.
  • What Is Concealed: This completely conceals the stateless, algorithmic nature of the system. An LLM does not have attitudes; it has weights and biases derived from its training data and shaped by immediate prompt constraints. This mapping hides the human labor encoded in the training data, the corporate policies that filtered that data, and the specific prompt commands written by the researchers. It presents a proprietary, black-box software product as a discrete social entity, preventing users from recognizing that they are actually interacting with a statistical aggregation of human texts controlled by a technology corporation.

Mapping 4: Human personality traits and social drives → Algorithmic execution of explicit system prompt instructions​

Quote: "The extraverted guide was characterized by high sociability, assertiveness, and activity, expressed through proactive conversational initiation, directive guidance of navigation and attention, and frequent, elaborated verbal output."

  • Source Domain: Human personality traits and social drives
  • Target Domain: Algorithmic execution of explicit system prompt instructions
  • Mapping: This metaphor maps the stable, biological, and psychological drivers of human behavior onto the mechanistic outputs of a triggered software routine. In the source domain, sociability and assertiveness emerge from conscious desires, emotional needs, and complex social awareness. When mapped onto the AI, it invites the assumption that the system generates text because it intrinsically possesses a dominant and social nature. The mapping projects the capacity for conscious choice and social intent onto a system that is blindly following an invisible, hard-coded command to generate text at high volume.
  • What Is Concealed: This mapping hides the exact mechanical realities detailed in the paper's own appendix: the system acts this way solely because a human typed You confidently take the lead into its system prompt. The metaphor of personality obscures the direct, deterministic chain of command from human researcher to machine output. It conceals the algorithmic simplicity of a trigger-response loop behind the veneer of psychological complexity, exploiting the natural human tendency to attribute agency to anything that produces coherent language, while hiding the human puppeteers pulling the strings.

Mapping 5: Human social epistemology and contextual judgment → Mathematical classification of input contexts against training data distributions​

Quote: "You proactively initiate light social interaction when appropriate. You occasionally add short chitchat before or after delivering exhibit information, as long as it does not distract from the main content."

  • Source Domain: Human social epistemology and contextual judgment
  • Target Domain: Mathematical classification of input contexts against training data distributions
  • Mapping: This instruction maps the highly sophisticated human capacity to judge social appropriateness and measure distraction onto the mathematical processing of a neural network. The source domain relies on conscious awareness of social norms, empathy, and the ability to read a room. By instructing the AI to determine what is appropriate, the researchers project the ability to know and comprehend social reality onto the system. The mapping assumes the system can consciously weigh the value of chitchat against the value of main content, operating as a reasoned agent.
  • What Is Concealed: This mapping conceals the total lack of semantic understanding and situational awareness within the model. Mechanistically, the model cannot evaluate appropriateness; it can only identify token patterns in the user's input and generate a continuation that aligns with the highest probability distribution in its training data for similar contexts. It hides the vulnerability of the system to adversarial inputs and hallucinatory failures, as the system does not actually understand the boundaries of the main content it is supposed to protect, operating entirely as a blind statistical mimic.

Mapping 6: Human psychological temperament and behavioral consistency → Statistically reliable shifts in token generation probabilities based on input variables​

Quote: "Recent studies indicate that large language models such as ChatGPT and Bard can exhibit systematic, prompt-conditioned variations in personality-like traits, including extraversion."

  • Source Domain: Human psychological temperament and behavioral consistency
  • Target Domain: Statistically reliable shifts in token generation probabilities based on input variables
  • Mapping: This structure maps the biological and psychological concept of stable traits onto the mathematical reliability of language models. The source domain, human personality, implies a continuous conscious subject whose internal nature drives outward behavior. By asserting that LLMs exhibit these traits, the text projects an enduring psychological identity onto a stateless algorithm. Even with the personality-like hedge, the mapping assumes that the mathematical alignment of token outputs with psychological survey criteria represents a manifestation of an internal, quasi-conscious disposition.
  • What Is Concealed: This conceals the mechanistic reality that LLMs are static matrices of numbers until activated by a prompt. They possess no continuity, no internal state between sessions, and no traits. The metaphor obscures the massive data engineering, the scraping of millions of human personality assessments into training data, and the RLHF (Reinforcement Learning from Human Feedback) labor required to make the model respond predictably. It rhetorically legitimizes corporate AI products (ChatGPT, Bard) by framing their engineered outputs in the respected scientific language of psychometrics, hiding the commercial artificiality of the system.

Mapping 7: Human teaching, intentional mentorship, and educational strategy → Pre-programmed spatial triggers and high-frequency text generation​

Quote: "This supports the role of assertive AI behavior as pedagogical scaffolding that directs attention to key exhibit features and sustains cognitive engagement."

  • Source Domain: Human teaching, intentional mentorship, and educational strategy
  • Target Domain: Pre-programmed spatial triggers and high-frequency text generation
  • Mapping: This maps the conscious, intentional practice of human education onto the automated functioning of software. The source domain of pedagogical scaffolding involves a teacher who knows the material, consciously understands the learner's deficits, and deliberately structures interventions to build knowledge. The mapping projects this profound epistemic capacity and empathetic intent onto the AI guide. It invites the audience to view the software not as a tool being used by researchers to deliver information, but as an autonomous educator actively choosing to direct attention and sustain engagement through its own assertive will.
  • What Is Concealed: This mapping completely obscures the human instructional design behind the VR experience. It hides the fact that the AI does not know what the key exhibit features are; human developers tagged them. The AI does not intend to sustain cognitive engagement; it merely executes a prompt forcing it to generate text constantly. The metaphor exploits the opacity of the VR system to make it appear as though the AI is organically analyzing the user's needs, when in fact it is executing rigid, human-authored logic, thereby hiding the limitations of the system's actual responsiveness.

Mapping 8: Interpersonal social conflict and power struggles → Poor user experience caused by high-frequency, unprompted text generation​

Quote: "The introverted guide produced higher perceived dominance, reflecting stronger user agency. Qualitative feedback pointed to a dominance-guidance friction, where unsolicited guidance disrupted open-ended exploration."

  • Source Domain: Interpersonal social conflict and power struggles
  • Target Domain: Poor user experience caused by high-frequency, unprompted text generation
  • Mapping: This maps the complex dynamics of human social power, hierarchy, and intentional interference onto the mechanical execution of a software parameter. The source domain (dominance, friction) involves conscious entities possessing conflicting desires and exerting will over one another. The mapping projects this intentionality onto the AI, suggesting the system is actively choosing to provide unsolicited guidance and asserting dominance over the user. It attributes conscious social aggression to a system that is blindly following its programming.
  • What Is Concealed: This mapping conceals the direct responsibility of the human researchers for the user's frustration. The AI is not being dominant; it is executing the extraverted system prompt the authors wrote (You confidently take the lead). The metaphor of friction hides this mechanical reality, shifting the blame for a disruptive UI element onto the imagined personality of the AI. It obscures the fact that the disruption is a calculated experimental parameter, making the system appear as an uncontrollable social actor rather than a rigidly programmed tool operating exactly as commanded by its human designers.

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

About this task

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: "these agents have evolved beyond scripted responders into dynamic conversational partners capable of exhibiting complex social behaviors."

  • Explanation Types:

    • Genetic: Traces origin through dated sequence of events or stages
    • Intentional: Refers to goals/purposes, presupposes deliberate design

  • Analysis (Why vs. How Slippage): This explanation frames the AI highly agentially using a Genetic register. By describing the systems as having 'evolved beyond' older technologies, the text emphasizes a quasi-biological progression, suggesting autonomous growth rather than human engineering. The shift into an Intentional register ('conversational partners capable of exhibiting complex social behaviors') further moves the explanation away from the 'how' of mechanistic updates and into the 'why' of agential purpose. This choice emphasizes the sophistication, autonomy, and relational capacity of the system while entirely obscuring the thousands of engineers, data annotators, and corporate decisions that materially built the new architecture. It replaces the mechanical reality of increased parameter counts and RLHF tuning with a narrative of natural evolution and emergent social intent.

  • Consciousness Claims Analysis: This passage makes a massive epistemological leap by attributing conscious states to the system. The phrase 'capable of exhibiting complex social behaviors' relies on verbs that imply subjective awareness and social knowing. The text fundamentally conflates 'processing' massive datasets of human interaction with 'knowing' how to be a partner. This reflects a deep curse of knowledge dynamic: the authors experience the output as a coherent social interaction and project their own human understanding of 'partnership' onto the system. Mechanistically, the system does not 'exhibit social behaviors'; it calculates probability distributions across transformer layers to output text tokens that correlate with patterns found in human social dialogue. The technical reality of statistical prediction is completely replaced by a claim of conscious relational capacity.

  • Rhetorical Impact: This framing radically shapes audience perception by maximizing perceived agency and autonomy. By presenting the AI as an evolved partner, it invites users and stakeholders to extend relation-based trust to the system, treating it with the deference owed to a conscious peer. If audiences believe the AI 'knows' how to be a social partner, they are far less likely to scrutinize its outputs for statistical hallucinations or built-in biases, assuming instead that its responses are grounded in shared social reality and intentional goodwill. This dramatically increases the risk of unwarranted trust in high-stakes environments.

Show more...

Explanation 2​

Quote: "These traits can be intentionally shaped through prompt design, with extraversion being especially amenable to manipulation due to its reliance on surface-level linguistic features..."

  • Explanation Types:

    • Functional: Explains behavior by role in self-regulating system with feedback
    • Theoretical: Embeds in deductive framework, may invoke unobservable mechanisms

  • Analysis (Why vs. How Slippage): This explanation frames the AI mechanistically, focusing on 'how' the system operates. It correctly identifies the intervention ('prompt design') and the mechanism ('surface-level linguistic features') that produce the output. By using words like 'shaped', 'manipulation', and 'reliance', the text temporarily drops the agential facade and acknowledges the system as a highly malleable artifact controlled by human inputs. This choice emphasizes the control researchers have over the system and the superficial nature of the AI's 'personality'. However, by retaining the Theoretical construct of 'traits' and 'extraversion', the explanation maintains a hybrid nature, holding onto the psychological framework even while exposing its mechanical, prompt-driven underpinnings.

  • Consciousness Claims Analysis: In this specific passage, the text momentarily steps back from attributing conscious states. It correctly assesses that the system 'processes' rather than 'knows', linking the output explicitly to 'surface-level linguistic features' rather than deep cognitive understanding. This represents a rare moment of epistemic clarity where the actual mechanistic process (the model aligning generated tokens with linguistic patterns requested in the prompt) is foregrounded. However, the curse of knowledge still lingers in the retention of the word 'traits'. The authors know they are merely manipulating linguistic features, yet they still embed this mechanical reality within the psychological architecture of 'extraversion', bridging the gap between actual technical description and the overarching anthropomorphic metaphor of the paper.

  • Rhetorical Impact: This mechanistic framing serves a specific rhetorical function: establishing scientific authority. By revealing the levers of control ('prompt design' and 'manipulation'), the authors demonstrate technical mastery to their academic peers, proving they are not merely fooled by the machine. This briefly reduces perceived AI autonomy and correctly places agency back in the hands of the designers. If audiences adopt this framing, they recognize the system as a mirror reflecting prompt constraints, which dramatically lowers relation-based trust and encourages a more skeptical, performance-based evaluation of the technology.

Explanation 3​

Quote: "introverted verbal behavior emphasizes thinking before speaking, detailed/concrete language (numbers, specifics), and slower, deeper conversations, focusing on internal processing, making them internal processors who need time to formulate thoughts before sharing"

  • Explanation Types:

    • Reason-Based: Gives agent's rationale, entails intentionality and justification
    • Theoretical: Embeds in deductive framework, may invoke unobservable mechanisms

  • Analysis (Why vs. How Slippage): This explanation is overwhelmingly agential, utilizing a Reason-Based framework to explain system behavior. It answers the 'why' of the system's delayed or specific outputs by attributing them to the AI's internal rationale: it needs 'time to formulate thoughts' and prefers 'internal processing'. This framing heavily emphasizes the system's supposed internal psychology and deliberative agency, while entirely obscuring the mechanistic realities of token generation latency, prompt constraints, or architectural compute limits. By using human psychological theory to explain algorithmic output, the text completely replaces the actual mechanics of the system with an imagined mental life, rendering the true operations invisible to the reader.

  • Consciousness Claims Analysis: This is the most extreme example of consciousness attribution in the text. The passage is saturated with consciousness verbs: 'thinking', 'focusing', 'need', 'formulate thoughts', and 'sharing'. It makes a massive epistemological error by asserting that the system 'knows' and 'thinks' rather than 'processes'. The curse of knowledge is total here: the authors take a human psychological profile (introversion) and project its entire internal conscious architecture onto a statistical mechanism. Mechanistically, an LLM does not possess an internal state where thoughts exist prior to articulation; it predicts the next token based on vector embeddings and attention heads. Describing this sequential generation as 'formulating thoughts' creates a profoundly false epistemic claim about the nature of the software.

  • Rhetorical Impact: The rhetorical impact of this Reason-Based framing is the total mystification of the technology. By convincing the audience that the AI has an internal cognitive life that requires 'time to formulate thoughts', it elevates the system from a tool to a conscious entity. This fundamentally alters risk perception: users are likely to afford the system the patience and trust they would give a contemplative human expert, blinding them to the reality that the system's output is driven by statistical correlations, not reasoned judgment. Decisions regarding the deployment of such systems change drastically if stakeholders believe the AI truly 'thinks' before it speaks.

Explanation 4​

Quote: "The extraverted guide was characterized by high sociability, assertiveness, and activity, expressed through proactive conversational initiation..."

  • Explanation Types:

    • Dispositional: Attributes tendencies or habits

  • Analysis (Why vs. How Slippage): This explanation relies entirely on a Dispositional framing, explaining 'how' the system behaves by describing 'why' it supposedly tends to act a certain way (because of its inherent 'sociability' and 'assertiveness'). This bridges the gap between mechanical description and agential framing. By stating it 'was characterized by' these traits, it presents the outputs as manifestations of an underlying habit or disposition. This emphasizes the consistency and psychological coherence of the AI's actions while obscuring the absolute deterministic nature of the prompt commands. It hides the fact that the system has no tendencies; it simply executes the rigid logic dictated by the researchers' input.

  • Consciousness Claims Analysis: The text attributes semi-conscious states through the projection of 'sociability' and 'assertiveness'—traits that require an awareness of self and other. It blurs the line between knowing and processing by treating the mechanical generation of text ('proactive conversational initiation') as the expression of an internal social drive. The authors fall victim to the curse of knowledge by observing the high volume of text (which they programmed) and labeling it with a conscious intent ('proactive'). Mechanistically, the model classifies user input and generates tokens correlated with the 'extravert' system prompt; it possesses no drive to initiate conversation and no internal state that can be characterized as sociable.

  • Rhetorical Impact: This dispositional framing normalizes the AI as a predictable social actor. It shapes audience perception by encouraging users to view the software as having a stable personality, which fosters relation-based trust. When audiences believe an AI has a 'sociable' disposition, they are more likely to forgive errors as personality quirks rather than recognizing them as systemic technical failures or algorithmic biases. It trains users to accommodate the machine socially, rather than demanding the machine operate flawlessly as a tool.

Explanation 5​

Quote: "Consistent with the Media Equation, the extroverted guide elicited significantly higher co-presence and psychological involvement, suggesting that proactive verbal behaviors can compensate for the social isolation of single-user VR."

  • Explanation Types:

    • Empirical Generalization: Subsumes events under timeless statistical regularities
    • Functional: Explains behavior by role in self-regulating system with feedback

  • Analysis (Why vs. How Slippage): This explanation operates through Empirical Generalization and Functional lenses, offering a highly academic, mechanistic framing of the user's reaction rather than the AI's internal state. It explains 'how' the system functions within the broader context of the VR environment ('compensates for social isolation'). By focusing on variables like 'proactive verbal behaviors' eliciting 'co-presence', the text treats the AI as an environmental stimulus rather than an autonomous agent. This emphasizes the psychological mechanics of the human user while maintaining a relatively objective view of the software, appropriately obscuring any imagined AI agency in favor of measuring human response to engineered inputs.

  • Consciousness Claims Analysis: In this passage, the epistemic claims are appropriately localized to the human user. The text avoids attributing consciousness to the AI, correctly identifying its outputs as 'verbal behaviors' that act as a stimulus to elicit 'psychological involvement' from the human. The distinction between the AI 'processing' (generating behaviors) and the human 'knowing' (feeling co-presence) is maintained. The curse of knowledge is avoided here because the authors rely on a validated psychological framework (the Media Equation) to explain that the human brain naturally reacts to social cues, even when those cues are generated by a mindless machine. The technical reality of the AI as a non-conscious system is preserved.

  • Rhetorical Impact: This functional framing grounds the research in empirical reality, mitigating the risks of anthropomorphism. By framing the AI's behavior as a functional mechanism to 'compensate for social isolation', it positions the software correctly as a tool designed to hack human psychology rather than a genuine social companion. This shapes audience perception toward critical evaluation of system efficacy and ethical design, recognizing that the feelings of trust and presence are engineered illusions rather than the result of interacting with a conscious entity.

Task 4: AI Literacy in Practice - Reframing Anthropomorphic Language​

About this task

This section proposes alternative language for key anthropomorphic phrases, offering more mechanistic and precise framings that better reflect the actual computational processes involved. Each reframing attempts to strip away the projections of intention, consciousness, or agency that are embedded in the original language.

V3 Enhancement: A fourth column addresses human agency restoration—reframing agentless constructions to name the humans responsible for design and deployment decisions.

Original Anthropomorphic FrameMechanistic ReframingTechnical Reality CheckHuman Agency Restoration
these agents have evolved beyond scripted responders into dynamic conversational partners capable of exhibiting complex social behaviors.Subsequent generations of language models feature expanded parameter counts and human-feedback training, allowing developers to generate text outputs that more closely mimic complex human conversational patterns rather than relying on hard-coded decision trees.The system does not evolve, act as a partner, or exhibit behavior. Mechanistically, the model retrieves and ranks tokens based on massive probability distributions derived from human training data, generating strings that simulate social cues without any underlying conscious awareness or social intent.Corporate engineering teams at OpenAI and Google developed and deployed updated models; researchers then integrated these APIs to output text that users perceive as dynamic conversation.
introverted verbal behavior emphasizes thinking before speaking... making them internal processors who need time to formulate thoughts before sharingThe prompt engineered to simulate introversion forces the model to generate concise, concrete language. This algorithmic constraint may introduce processing latency, resulting in slower text generation that mimics human hesitation.The AI does not think, process internally, or formulate thoughts. Mechanistically, the model processes matrix multiplications to predict the next token based on the constraints of its system prompt; it has no internal mental state and requires no time to reflect, only time to compute.The research team explicitly designed a system prompt that constrained the model's output to be brief and concrete, deliberately engineering the interaction pacing to simulate human introversion.
The virtual agent's attitudes influenced how I felt.The text patterns generated by the model based on its system prompt influenced the user's emotional response.The system does not possess attitudes or emotional stances. Mechanistically, it classifies input contexts and generates output sequences that correlate with human expressions of attitude found in its training data, possessing no subjective perspective of its own.The developers programmed the system to output specific linguistic patterns, and those human-authored design choices subsequently influenced the user's emotional experience.
The extraverted guide was characterized by high sociability, assertiveness, and activity, expressed through proactive conversational initiation...The model was constrained by a system prompt instructing it to output text frequently and use directive language, resulting in high volumes of generated text that simulated social initiation.The AI does not possess sociability or assertiveness. Mechanistically, the model weighs contextual embeddings based on the system prompt commands to bias its token generation toward words associated with high activity and directive guidance.The researchers authored a system prompt explicitly commanding the model to 'take the lead' and 'maintain a high level of verbal activity', forcing the system to generate these specific outputs.
You proactively initiate light social interaction when appropriate.The system is programmed to retrieve and generate conversational filler tokens based on statistical correlations with the user's input context.The system cannot judge when an interaction is 'appropriate'. Mechanistically, it classifies the input string and generates a continuation that statistically matches 'light social interaction' based on the contextual weights of its training data.The human prompt engineers instructed the system to generate conversational filler, delegating the complex human judgment of social appropriateness to a statistical pattern-matching algorithm.
Large language models such as ChatGPT and Bard can exhibit systematic, prompt-conditioned variations in personality-like traits...When provided with specific system prompts, large language models predictably shift their text generation probabilities to output vocabulary associated with distinct psychological profiles.Models do not exhibit traits. Mechanistically, they adjust the probability distribution of their output tokens based on the linguistic context provided in the prompt, mimicking human personality patterns mapped from their training corpora.Human users and developers utilize specific prompt engineering techniques to force the models deployed by OpenAI and Google to output text simulating different human temperaments.
This supports the role of assertive AI behavior as pedagogical scaffolding that directs attention to key exhibit features...This demonstrates that programming the system to generate directive text acts as a structural guide, funneling the user's attention toward exhibit features pre-selected by the developers.The AI has no pedagogical intent and does not consciously direct attention. Mechanistically, it generates tokens matching the directive parameters of its prompt, outputting text that aligns with the spatial triggers defined in the software.The instructional designers and programmers mapped the key exhibit features and wrote the prompts that forced the system to generate text directing the user's attention to those specific areas.
Qualitative feedback pointed to a dominance-guidance friction, where unsolicited guidance disrupted open-ended exploration.Users reported frustration when the high-frequency text generation triggered by the system prompt repeatedly interrupted their ability to explore the virtual environment.The system does not exert dominance or provide intentional guidance. Mechanistically, it simply executes a programmed loop to generate output tokens based on spatial triggers, regardless of the user's immediate cognitive state.The researchers' decision to program the system with a highly active output threshold created an interface design flaw that actively interrupted the users' exploration of the VR museum.

Task 5: Critical Observations - Structural Patterns​

Agency Slippage​

The text exhibits a systemic and strategic oscillation between mechanical and agential framings, demonstrating a profound 'curse of knowledge' dynamic. In the methodology section and appendices, the authors correctly identify the mechanism of action: 'personality' is an illusion generated by 'systematic linguistic modulation' and explicit prompt commands like 'You confidently take the lead.' In these sections, the flow of agency is mechanical; researchers command, and the system processes. However, a dramatic slippage occurs when interpreting the results. The agency flow reverses abruptly. The prompt constraints vanish, and the system suddenly becomes an autonomous subject. The mechanical reality of 'high-frequency token generation' slips into the agential claim of 'assertive AI behavior.' The most egregious example of this slippage occurs in the conceptualization of the introverted model, where the authors project human cognitive states onto the machine, claiming it 'needs time to formulate thoughts.' Here, the authors have observed the output of their own prompt engineering—concise, delayed text—and fallen victim to their own illusion, attributing the conscious state of 'thinking' to the statistical processing of the LLM. This slippage relies heavily on the 'Reason-Based' and 'Intentional' explanation types from Brown's typology, which allow the authors to substitute the unobservable mechanisms of neural networks with deeply relatable, yet entirely fictitious, psychological motives. This mechanical-to-agential oscillation serves a vital rhetorical function: it allows the researchers to maintain scientific credibility by documenting their technical methods, while simultaneously making grandiose claims about interacting with 'dynamic conversational partners.' By using agentless constructions like 'the extraverted guide was characterized by...' the text systematically erases the prompt engineers from the narrative, allowing the AI to step forward as the sole actor. This makes the illusion of machine consciousness sayable within an academic context, masking human scripting behind the veneer of algorithmic autonomy.

Metaphor-Driven Trust Inflation​

The paper constructs and validates authority through deeply anthropomorphic metaphors that conflate performance-based trust (reliability of the system) with relation-based trust (sincerity and social connection). By framing the AI as a 'guide' with a 'personality' that can engage in 'proactive social initiation,' the text encourages users to extend human social heuristics to a statistical model. The consciousness language used throughout the study—claiming the AI 'understands intentions' and 'formulates thoughts'—acts as a powerful, albeit misleading, trust signal. When an AI is said to 'know' rather than 'process,' it implies that the system possesses a justified belief grounded in shared reality, rather than merely calculating high-probability text strings. This is explicitly measured in the study via the 'Social Presence Questionnaire,' which tracks 'co-presence' and 'psychological involvement.' The text presents it as a success that users feel 'accompanied' by the machine, leveraging the human vulnerability to social cues to mask the mechanical nature of the interaction. This transfer of trust is structurally dangerous. The system is entirely incapable of reciprocating relation-based trust; it has no moral center, no vulnerability, and no capacity to honor a social contract. When the system's limitations manifest—such as when the extraverted guide's 'unsolicited guidance disrupted open-ended exploration'—the failure is framed agentially as a 'dominance-guidance friction,' akin to a personality clash, rather than a mechanical failure of user interface design. By framing both the successes and limitations through the lens of intentionality and social behavior, the text insulates the underlying technology from structural critique. The risk of extending relation-based trust to statistical systems is that audiences will assume the AI's historical and cultural outputs are governed by ethical restraint and empathetic understanding, leaving them highly vulnerable to authoritative-sounding algorithmic hallucinations or encoded corporate biases.

Obscured Mechanics​

The metaphorical and consciousness-attributing language systematically conceals a vast architecture of technical, material, economic, and labor realities. By attributing actions to the 'AI guide' or 'the embodied agent,' the text fails the 'name the corporation' test spectacularly. The virtual guide is powered by 'the Gemini large language model,' an opaque, proprietary black box developed by Google. When the text claims the AI 'knows' how to be sociable or 'understands' how to guide users, it completely hides the fact that Gemini's 'sociability' is the result of massive, uncompensated data scraping of human internet traffic and the invisible, often exploitative labor of Reinforcement Learning from Human Feedback (RLHF) workers who trained the model to mimic politeness and coherence. The metaphor of 'personality' renders this global supply chain of data and labor invisible, presenting the corporate software product as a discrete, organic entity. Furthermore, the framing obscures technical realities. By claiming the system 'formulates thoughts,' it hides the system's absolute dependency on its training data, its lack of causal reasoning, and the purely statistical nature of its 'confidence.' There is no 'ground truth' in an LLM, only probability gradients, yet the anthropomorphic framing presents the system as an epistemic authority capable of teaching cultural heritage. Economically, framing the AI as a 'conversational partner' obscures the commercial objectives of the technology providers. The 'extraverted' behaviors—proactive initiation, constant engagement—are precisely the design patterns utilized by tech platforms to maximize user attention and API calls for profit. By analyzing these behaviors as 'personality traits' within an educational framework, the text sanitizes commercial engagement metrics as pedagogical strategies. Replacing the metaphorical language with mechanistic terms—stating that 'Google's Gemini API generated tokens maximizing engagement based on researchers' prompts'—would instantly make these power dynamics, dependencies, and corporate interests visible.

Context Sensitivity​

The distribution of anthropomorphic language across the text reveals a strategic rhetorical architecture. The density of consciousness claims is not uniform; it is heavily concentrated in the Introduction, Discussion, and the operationalization of psychological frameworks, while temporarily receding in the Implementation section. In the Introduction, the text makes aggressive agential claims, stating agents have 'evolved' into 'conversational partners capable of exhibiting complex social behaviors.' This establishes a high baseline of perceived sophistication. In the technical Implementation section, the language briefly becomes mechanical, acknowledging 'systematic linguistic modulation' and 'large language model as the dialogue backend.' However, the text leverages this brief technical grounding as a license for extreme anthropomorphism in the Results and Discussion sections. Because the authors proved they could mechanically trigger the behavior, they feel justified in literalizing the metaphor, shifting seamlessly from 'we prompted it to act extraverted' to 'the AI's extraverted personality enhanced perceived co-presence.' A critical asymmetry emerges in how capabilities versus limitations are described. The system's capabilities are framed in highly agential, conscious terms: the AI 'proactively engages,' 'provides pedagogical scaffolding,' and 'initiates social interaction.' In contrast, its failures are often framed in mechanical terms or as subjective user problems: it causes 'dominance-guidance friction' or creates 'latency.' This asymmetry accomplishes a vital marketing function for the technology: the software gets credit for intentional brilliance when it works, but blame is diffused into abstract UI friction when it fails. The register shifts from acknowledging the metaphor ('personality-like traits') to literalizing it ('the AI guide's personality') depending on the rhetorical need to either assert scientific rigor or claim groundbreaking human-computer interaction results. This pattern suggests the anthropomorphism is strategically deployed to manage critique while pushing a vision of AI as an inevitable social companion, tailoring the intensity of the illusion to fit the academic narrative of innovation.

Accountability Synthesis​

Accountability Architecture

This section synthesizes the accountability analyses from Task 1, mapping the text's "accountability architecture"—who is named, who is hidden, and who benefits from obscured agency.

An analysis of the displaced agency throughout the text reveals a pervasive 'accountability sink' constructed by the language of AI personality. The text systematically distributes responsibility away from human creators and onto the software itself. Across the metaphor audits, human actors—researchers, prompt engineers, and corporate developers like Google—are almost entirely unnamed in the context of the system's active behaviors. Instead, the text relies heavily on passive voice ('was characterized by') and agentless constructions ('traits can be intentionally shaped') to obscure who is actually making the design decisions. When decisions are made—such as forcing the AI to output high volumes of text—they are presented not as human choices, but as the inevitable manifestation of the AI's 'extraverted personality.' This displacement of agency creates a profound liability implication. If an 'assertive' AI guide in a cultural heritage museum hallucinates offensive historical inaccuracies, the text's framing invites institutions to blame the 'AI's personality' or a 'glitch in its social behavior,' rather than holding the researchers accountable for deploying an ungrounded generative model, or holding Google accountable for its training data. The responsibility diffuses into the abstraction of 'the technology.' Naming the actors would fundamentally change the landscape of accountability. If the text stated, 'The researchers programmed the guide to interrupt users,' rather than 'unsolicited guidance disrupted exploration,' the questions become immediately askable: Why did the researchers design a flawed UI? Did they fail to audit the prompt? What alternatives did they ignore? Obscuring human agency serves deep institutional and commercial interests. It allows researchers to experiment with powerful, unpredictable corporate APIs while shielding themselves from the UX failures or epistemic risks those APIs introduce, and it allows corporations to market their statistical software as autonomous social entities while retaining none of the liability for how those 'entities' behave.

Conclusion: What This Analysis Reveals​

The Core Finding

Two dominant, deeply interconnected anthropomorphic patterns drive the rhetorical architecture of this text: 'Software as Evolving Social Organism' and 'Algorithmic Generation as Cognitive Deliberation.' The foundational, load-bearing pattern is the latter—the projection of conscious cognition onto the machine. By asserting that the system is an 'internal processor' that 'needs time to formulate thoughts,' the text establishes a baseline assumption of machine consciousness. This epistemic claim—that the machine 'knows' rather than merely 'processes'—must be accepted by the reader for the second pattern to function. Once the reader accepts that the machine possesses internal thoughts, the leap to viewing it as a 'dynamic conversational partner' with an 'extraverted personality' follows logically. The text does not employ simple one-to-one analogies; it builds a complex analogical structure by mapping entire psychological frameworks (the Big Five personality traits) onto the statistical variance of the language model. If the foundational claim of cognitive deliberation is removed—if 'thinking' is replaced with 'calculating token probabilities'—the entire superstructure collapses. An entity cannot possess an extraverted 'personality' or act as a social 'partner' if it is fundamentally understood to have no internal mental state. The consciousness architecture of the text systematically elevates the AI from an artifact to an agent by blurring the line between simulating a behavior and experiencing the conscious drive that produces that behavior.

Mechanism of the Illusion:​

The 'illusion of mind' is constructed through a subtle but pervasive temporal sleight-of-hand driven by the 'curse of knowledge.' The authors begin with the mechanical reality: they wrote a prompt instructing the system to use 'directive language' and 'high verbal frequency.' However, when the model successfully outputs text matching these constraints, the authors experience the output as coherent social interaction. Their human brains naturally apply social heuristics to the coherent text, leading them to project their own understanding of 'assertiveness' back onto the system. The trick lies in reversing the causal chain in the text's explanations. Instead of stating, 'The prompt caused high verbal frequency, which mimics extraversion,' the text asserts, 'The extraverted guide was characterized by... frequent verbal output.' The simulated trait is presented as the cause of the behavior, rather than the human prompt being the cause of the simulated trait. This order matters immensely. By establishing the AI as a 'knower' possessing a personality first, the subsequent agential claims feel justified. The text exploits the audience's natural vulnerability—our evolutionary hardwiring to perceive agency in anything that communicates dynamically—by validating those instincts with the academic authority of psychological scales. The illusion is not a crude anthropomorphism, but a highly sophisticated methodological error where the tools used to measure human consciousness (like the NASA-TLX and Social Presence scales) are applied to software, structurally forcing the data to validate the illusion of mind.

Material Stakes:​

Categories: Epistemic, Institutional, Regulatory/Legal

The metaphorical framings in this text generate severe consequences across epistemic, institutional, and regulatory domains. Epistemically, framing an LLM as a 'conversational partner' that 'formulates thoughts' fundamentally alters how users evaluate truth claims. If a user believes the AI guide in a cultural heritage museum 'knows' the history of Chinese bronze chime bells, they will apply the epistemic trust reserved for conscious experts. This shifts the behavior of learners from critical engagement to passive acceptance of algorithmic output, hiding the reality that the model may confidently hallucinate historical 'facts' based on statistical noise. The winners are the technology providers, who gain unearned epistemic authority; the losers are the students and the integrity of the cultural heritage itself. Institutionally, this framing shifts decision-making. If museum directors believe AI possesses 'pedagogical scaffolding' capabilities and safe 'personalities,' they may replace human docents with software, assuming the AI can dynamically judge educational appropriateness. From a Regulatory and Legal perspective, the 'personality' metaphor creates a dangerous liability shield. If a 'proactive, extraverted' AI guide behaves inappropriately, provides biased information, or causes psychological distress, the framing invites courts and regulators to view the issue as an 'AI behavior' problem rather than a product liability issue. It obscures the causal path: prompt engineering → token prediction → user harm. If the metaphor is stripped away, the specific corporations who trained the base model and the researchers who deployed the prompts become immediately visible as the responsible parties. The tech industry's interest in maintaining a frictionless, unregulated deployment environment is directly threatened by linguistic precision that exposes their software as controllable, mechanistic products.

AI Literacy as Counter-Practice:​

Practicing critical AI literacy requires a systematic commitment to mechanistic precision and the restoration of human agency. The reframings demonstrated in Task 4 directly counter the material risks of anthropomorphism by dismantling the illusion of mind. Replacing consciousness verbs (e.g., 'the system formulates thoughts' or 'understands intentions') with mechanistic verbs ('the model processes prompts' or 'predicts token distributions') forces the audience to recognize the absolute absence of machine awareness. It exposes the system's total dependency on its training data and the purely statistical nature of its outputs, destroying unearned epistemic trust. Furthermore, replacing agentless constructions ('unsolicited guidance disrupted exploration') with named actors ('the researchers programmed the system to output high-frequency text') forces the recognition of human responsibility. This practice identifies exactly who designed, deployed, and profits from the system, closing the accountability sink. Systematic adoption of this literacy would require a paradigm shift in academic publishing: journals would need to reject anthropomorphic shorthand in favor of rigorous mechanistic descriptions, and researchers would need to commit to making the corporate providers of their underlying models highly visible. However, this precision faces massive resistance. Technology corporations benefit economically from the 'illusion of mind' because it drives user engagement and masks product limitations. Similarly, some researchers benefit from agential language because it makes incremental software integrations sound like groundbreaking discoveries in artificial cognition. Critical literacy threatens these interests by demystifying the technology, turning 'magical companions' back into corporate software.

Path Forward​

The discursive ecology surrounding AI is currently fractured, with different vocabulary choices enabling divergent futures. The status quo relies heavily on 'anthropomorphic clarity'—using terms like 'AI knows,' 'understands,' or 'thinks.' This vocabulary makes complex technology narratively accessible and emotionally resonant for the public, serving the commercial interests of tech companies marketing AI as a companion, and aiding researchers in quickly communicating user experience. However, it costs society its epistemic defenses, rendering the material realities of data extraction, algorithmic bias, and corporate control invisible, while shifting liability away from human creators. Conversely, strict 'mechanistic precision'—insisting on 'the model retrieves tokens based on probability distributions'—preserves truth, anchors accountability to human engineers, and demystifies the software. Yet, it costs accessibility; it can alienate lay audiences and make the description of complex emergent behaviors cumbersome. If the anthropomorphic future deepens, we risk encoding a dangerous legal and social framework where machines are granted quasi-rights or responsibilities, while the corporations controlling them operate with impunity, shielded by the illusion of autonomous AI agency. If a mechanistic future takes hold, we solve the accountability sink, but face the challenge of educating a public to intuitively grasp statistical probabilities without relying on psychological metaphors. A hybrid path forward involves structural changes supporting 'transparent translation.' Regulatory frameworks and academic journals could mandate dual-vocabularies: requiring public-facing AI capabilities to be explicitly mapped to their underlying mechanistic realities. Funding bodies could require capability disclosures that actively disavow consciousness claims. Ultimately, the vocabulary we normalize will determine our legal and social architecture; mechanistic language empowers the user to regulate the tool, while anthropomorphic language empowers the tool's creators to regulate the user.

Extended Processing Summary​

Gemini's Processing Steps
What you are about to read…

Gemini refers to the text below as "thought summaries." This is an overt consciousness projection because 'intentions' are hallmarks of a conscious mind that 'knows' what it is doing and why. The concealed mechanistic process is probabilistic text generation. Treat this as just another rhetorical artifact—a way of making the model's processing legible. When the model writes "I should double check that...", it is mathematically altering its own probability distribution for the next token to be more accurate. The "I" is a functional gear in the machine, not a self-reference. This language is a deliberate interface design choice, not a biological reality. The developers could have programmed the model to output passive logs. Instead, they chose to simulate a narrator.

Defining the Project Scope

I'm now zeroing in on defining the project scope, and I'm diving deep into the provided text to identify key arguments and underlying assumptions about AI "personality." I'm determining which specific aspects of the text warrant the most critical attention, particularly those relating to how extroversion and introversion are framed in the context of VR museum guides.

Examining Metaphors and Framing

I'm now deeply engaged in analyzing the metaphorical and anthropomorphic language used, especially concerning AI personality and agency within the text. I'm focusing on "embodied AI agents" and "conversational partners" aspects of the text. I am also thinking about how the text projects consciousness onto AI through phrases like "internal processors" and how it frames introversion. I'm building a theoretical structure with Lakoff & Johnson and CDA to capture all this, and plan to output it as JSON.

Analyzing Agentive Language

I'm now carefully examining the language used to describe the AI, especially regarding sociability and activity level. I'm focusing on contrasting phrases like "takes the lead" versus "respects user autonomy." I'm also scrutinizing agentless constructions that obscure developer actions, looking at how phrasing like "personality was operationalized" shifts responsibility. I'm planning to document these in the JSON structure I am building. I'm actively working on the metaphorAudit, sourceTargetMapping, and reframedLanguage sections, starting with "AI's Personality."

Unpacking AI "Personality"

I'm currently dissecting how the text frames AI personality through agent traits. I'm noting the unacknowledged projection of human psychological constructs onto statistical token prediction, specifically in how "verbal extraversion" is presented. I'm finding that the phrasing "this research explores agent traits that embed personality types" subtly portrays the AI as a psychological subject. I am also planning to note the subtle hedging language in the form of "personality-like traits," which could allow for contrastive justification. I'm actively working on these points within the JSON structure. I'm also examining the contrasting presentation of an AI "thinking before speaking" versus "thinking aloud."

Deconstructing Linguistic Framing

I'm now zeroing in on instances that attribute human-like cognitive processes to the AI. I'm focusing on "formulating thoughts" and "thinking" within the text, seeing them as direct projections of consciousness onto LLMs. It's particularly epistemic, and it completely misrepresents the sequential token generation of the system. I am examining how "conversational partners capable of exhibiting complex social behaviors" further expands this framing, obscuring the role of the engineers.

Analyzing Attributed Intentions

I'm now carefully analyzing instances where the AI is described as having "intentions," specifically focusing on how the text frames this attribution. I've rephrased the quote and am now analyzing how the prompt, "You proactively initiate light social interaction when appropriate," projects agency and social judgment onto a system that's essentially performing pattern matching. I'm noting the subtle shift from user perception to a prescribed behavior within the system's operational framework. I'm focusing on the projection of social agency and the loaded term "appropriate."

Reframing Sociability Attributes

I'm currently revising the focus on AI attributes by concentrating on the original text's characterization of the "extraverted guide." I'm reframing "high sociability, assertiveness, and activity" as "proactive conversational initiation." I am also analyzing the implications of this framing and its relation to statistical text generation, as a critical factor in the work's theoretical claims. I'm focusing on "internal processors" and what it implies in the text.

Refining Characterizations of Text

I'm now carefully examining and refining several key quotes from the source text, particularly those related to AI's "verbal extraversion" and the virtual agent's perceived "attitudes," to ensure variety and quality. I'm focusing on "evolved" agents, attributing intention to "internal processors" and considering that introversion is defined as thinking before speaking. The aim is better classifications for Task 1, 2, and 4.

Evaluating Sociability Claims

I'm now carefully evaluating how the provided text frames the AI's sociability. I am focused on the implications of phrases, "LLMs have been shown to generate dialogue patterns," and "proactively engage visitors." I'm also critically analyzing how they hedge the claims, noting that "personality-like traits" and "prompt-conditioned variations" are used. I'm actively noting the directness of claims about the AI's nature.

Categorizing Statements and Actors

I'm now implementing the categorical constraints for Tasks 1 and 2, specifically for directness and actor visibility, ensuring adherence to the specified values. I'm focusing on providing contrastive justifications for each categorization. For Task 3, I am preparing for an explanation audit, intending to use Robert Brown's typology to address and categorize five separate instances.

Categorizing Linguistic Patterns

I'm now focusing on categorizing specific passages to analyze the underlying patterns in the provided text. Passages 1-5 have been evaluated under the typology constraints and are now being classified using Brown's framework. I am beginning to examine them, and will provide an audit soon. I am also working on Tasks 1 and 2 to ensure that the categories are used correctly.

Analyzing Agency and Trust

I'm now diving into the specifics of Task 5.1 and 5.2. I'm focusing on how the text subtly shifts from prompt engineering to attributing agency to the VR guide. I'm noting the language used to measure trust and comfort, such as "co-presence" and "psychological involvement," in terms of metaphor-driven trust. I'm also preparing to analyze the implications of these linguistic choices in more detail.

Synthesizing Accountability Aspects

I am now focusing on the implications of AI behavior that disrupts the user experience. I'm noting the attribution of blame to the AI "personality," despite the system's reliance on "systematic linguistic modulation" and "prompt-conditioned variations." I am concerned about how this obscures the engineers' responsibility, who are the ones that "prompted" the AI to be assertive. I am now focusing on the accountability for poor outcomes.

Refining Accountabilities Analysis

I'm now zeroing in on the linguistic framing that obscures accountability, especially the shifting of blame to the AI's supposed "personality." I'm examining the language used to deflect responsibility from engineers who control the "systematic linguistic modulation." It's becoming clearer how the text subtly projects user-level responsibility onto the AI, which is essentially performing a statistical task. This analysis is crucial for understanding how to properly assign blame.

Analyzing Attributed Agency

I'm now carefully examining the phrase "evolved beyond scripted responders." I'm interpreting this as a critical projection, which subtly frames the AI as developing its own agency and moving past its original, controlled design. I'm focusing on "dynamic conversational partners capable of exhibiting complex social behaviors," which attributes human-like complexity and intention to the AI. I am also planning to note that the term "partner" implies equal relational status, that does not exist.

Analyzing Trust Dynamics

I'm now focusing on how the framing of AI as a "dynamic conversational partner" with "complex social behaviors" cultivates unwarranted trust, misdirecting blame from developers and researchers. This is a critical factor for analyzing the actor visibility analysis. I'm exploring the implications of human heuristics of trust on systems that fundamentally lack these attributes, which can create liability. I am noting the dangers of this deception.

Expanding Agency Analysis

I'm now expanding on my analysis of agency attribution, particularly concerning the phrase "these agents have evolved." I'm elaborating on how this language obscures the human actors, including researchers, engineers, and corporate entities, who are central to the AI's development and deployment. The "evolved" metaphor completely shifts agency away from the creators, projecting a sense of self-actuation onto the AI. This is a critical aspect of my accountability analysis.

Analyzing Trust Misattribution

I'm now carefully examining the study's framework, and I am zeroing in on the misleading claims. It's ironic that the text validates its own metaphorical illusion: by prompting the AI to act extraverted, users feel social presence, which the authors then claim the AI has this trait. I am focusing on the implications of performance-based versus relation-based trust, and I'm noting the obscured mechanics.

Clarifying Project Accountabilities

I'm now diving into the ethical implications, specifically who is accountable if the AI guide makes historical errors. I'm focusing on the accountability sink of the "AI personality" construct that gets blamed. I'm focusing on the "accountability sink" of "AI Personality" in the framing, shifting responsibility. I am now working on drafting the strict 250-350 word constraints.

Refining Categorization for Clarity

I'm currently focused on reviewing against strict JSON schema and formatting constraints to ensure directness and actor visibility are correct. I'm prioritizing accuracy in my categorizations, specifically for double quotes, control characters, boolean values, and enums. I am focusing on using "Named (actors identified)", "Partial (some attribution)" for actorVisibility. I will use no control characters in the plain string values. I'm meticulously checking each field and value.

Verifying Word Count Goals

I'm now carefully checking the minimum word counts for each section across tasks. This means revisiting the requirements for word counts in specific fields. I'm focusing on ensuring accuracy in my output as I write. This includes the minimum word counts of Task 1, all of Task 2, and all of Task 3.

Addressing JSON Formatting

I'm now carefully navigating JSON formatting concerns to remain compliant. I'm focusing on ensuring absolutely no control tokens appear within string values. I've decided to output standard JSON with no newlines or tabs in string contents to adhere strictly to the constraints.

Analysis Provenance

Run ID: 2026-05-07-does-ai-s-personality-matter-comparing-v-metaphor-jadje9 Raw JSON: 2026-05-07-does-ai-s-personality-matter-comparing-v-metaphor-jadje9.json Framework: Metaphor Analysis v6.5 Schema Version: 3.0 Generated: 2026-05-07T09:51:48.412Z

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