Skip to main content

🆕 Can AI Reason Like an Urban Planner? Benchmarking Large Language Models Against Professional Judgment

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. Machine Cognition and Sensibility​

Quote: "evaluate whether these systems can reason with the contextual sensitivity, value awareness, and institutional literacy"

  • Frame: AI as a professional capable of awareness and reasoning
  • Projection: This metaphor projects deep human cognitive and psychological capacities—reasoning, sensitivity, and awareness—onto a statistical matching system. By using words like 'sensitivity' and 'awareness', it implies the system possesses a conscious subjective experience or an internal compass that grasps the social and ethical weight of urban planning. It attributes conscious understanding to computational processes that merely predict sequences of text based on distribution patterns, conflating mathematical optimization with human empathetic and cognitive engagement.
  • Acknowledgment: Direct (Unacknowledged) (The sentence presents these capabilities as literal functions the systems might possess. I considered 'Hedged/Qualified' because the paper asks 'whether' they can do this, implying doubt about success, but ruled it out because the phrasing fundamentally accepts 'reasoning' and 'awareness' as literal, valid categories for evaluating algorithms rather than treating them metaphorically.)
  • Implications: Framing LLMs as possessing 'awareness' and 'sensitivity' drastically inflates perceived sophistication, encouraging users to grant the system unwarranted relation-based trust. It implies the machine can empathize with human values, masking the reality that it cannot feel or understand the impact of its outputs. This creates profound risks in public policy, where users might rely on the 'sensitivity' of an algorithm instead of rigorously auditing its statistical outputs for encoded historical biases.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The construction positions 'these systems' as the primary actors being evaluated for their sensitivity. I considered 'Partial' because the systems are products of human engineering, but ruled it out because no human developers, corporations, or engineers are mentioned in this framing. This agentless construction serves the interests of AI companies by shifting focus to the autonomous capabilities of the software, obscuring the human actors who actually determine what 'values' are encoded in the training data and who profits from deployment.
Show more...

2. Epistemic Possession​

Quote: "models 'know' planning facts rather than whether they can reason with planning judgment"

  • Frame: AI as a conscious repository of knowledge
  • Projection: The metaphor maps human epistemic possession—the conscious state of knowing a fact—onto a computational system's ability to store and retrieve data weights. While knowing requires subjective awareness and the ability to justify a true belief, the system merely executes pattern matching. By contrasting 'knowing' with 'reasoning', the text projects a dual-tiered cognitive architecture onto the model, suggesting it possesses an inner mental life capable of both rote memorization and higher-order cognitive deliberation.
  • Acknowledgment: Explicitly Acknowledged (The text uses explicit scare quotes around the word 'know', explicitly acknowledging its metaphorical status. I considered 'Direct' because 'reason' in the same sentence lacks scare quotes, but ruled it out because the primary epistemic verb 'know' is visibly marked by the author as problematic or non-literal within this specific clause.)
  • Implications: Even when acknowledged with scare quotes, pairing 'know' with 'reason' continually reinforces the illusion of mind. This framing affects understanding by suggesting that AI failures are due to a lack of deeper 'judgment' rather than an inherent incapacity for any genuine knowledge at all. This distinction blurs the line between human cognitive architecture and statistical weights, leading policymakers to overestimate the system's reliability in handling ambiguous or contested information.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The text presents 'models' as the sole subjects capable of knowing or reasoning. I considered 'Ambiguous' because the text is discussing a theoretical capability, but ruled it out because the grammatical structure clearly positions the artifact as the independent agent. This hides the human data annotators who curated the facts and the engineers who designed the retrieval mechanisms, diffusing responsibility for factual errors away from the creators and onto the tool's 'judgment'.

3. Psychological Defiance​

Quote: "models exhibit a characteristic paralysis: they enumerate considerations exhaustively but refuse to make the normative commitments"

  • Frame: AI as an anxious or defiant professional
  • Projection: This metaphor projects complex psychological states and intentionality—paralysis, exhaustiveness, and active refusal—onto a machine learning model. It maps the human experience of moral hesitation or deliberate avoidance of responsibility onto a system that is mechanistically constrained by reinforcement learning guidelines designed to prevent controversial outputs. The projection suggests the AI has an internal moral compass, recognizes a normative dilemma, feels 'paralyzed', and consciously 'refuses' to engage, entirely erasing the statistical mechanics at play.
  • Acknowledgment: Direct (Unacknowledged) (The text states 'refuse to make' as a literal description of the system's behavior without any hedging. I considered 'Hedged/Qualified' because the authors note elsewhere the systems are trained on non-controversial outputs, but ruled it out because this specific sentence frames the output entirely as a spontaneous psychological refusal by the model itself.)
  • Implications: Describing an AI as 'refusing' or suffering 'paralysis' creates a powerful illusion of autonomy and moral agency. It invites users to interpret statistical guardrails as deliberate ethical stances taken by a conscious entity. This drastically inflates perceived sophistication and obscures liability; if an AI 'refuses' to act, it implies the machine made a choice, deflecting blame from the developers who engineered it to generate neutral, non-committal text when detecting contentious keywords.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The models are framed as autonomous actors exhibiting paralysis and actively refusing. I considered 'Named' because human RLHF constraints are the actual cause, but ruled it out because the text entirely erases the human engineers who hard-coded these safety constraints. This displacement benefits AI corporations by framing their safety engineering as the model's own organic moral deliberation, completely hiding the human policy decisions determining which commitments are avoided.

4. Deceptive Intentionality​

Quote: "models confidently fabricate specific regulatory requirements that do not exist, blending elements from different jurisdictions"

  • Frame: AI as a confident liar or fabricator
  • Projection: The text maps the human traits of confidence and deliberate deception (fabrication) onto algorithmic token prediction. 'Confident' implies a subjective feeling of certainty or a conscious presentation of authority, while 'fabricate' implies an intention to invent or deceive. In reality, the model calculates probability distributions without any awareness of truth, falsity, or jurisdictional boundaries. The metaphor attributes conscious intention and a psychological state to an automated text generator.
  • Acknowledgment: Direct (Unacknowledged) (The phrase 'confidently fabricate' is used directly as a factual description of the output. I considered 'Ambiguous' because 'fabricate' can sometimes be used mechanically, but ruled it out because the addition of the psychological adverb 'confidently' solidifies the presentation as an unhedged claim about the system's internal agential state.)
  • Implications: Framing statistical errors as 'confident fabrications' dangerously anthropomorphizes machine error. It suggests the AI is actively trying to deceive, which implies it knows the truth but chooses to hide it. This creates unwarranted assumptions about the system's underlying capabilities, misleading users into believing the model can be 'corrected' or 'taught not to lie', rather than recognizing that the generation of plausible but false text is a fundamental mathematical feature of the architecture.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The text attributes the act of fabrication entirely to the models. I considered 'Partial' because the text later discusses training data origins, but ruled it out because in this exact description, the models act alone. This construction hides the companies who deployed systems mathematically optimized to produce fluent text regardless of factual accuracy. Naming the actors would force accountability onto the developers who chose to release systems known to hallucinate in professional contexts.

5. Cognitive Conflation​

Quote: "Models frequently blur the boundaries between related but distinct planning concepts—treating them as interchangeable"

  • Frame: AI as an imprecise thinker
  • Projection: This metaphor maps the human cognitive failure of 'blurring' concepts or 'treating' things inappropriately onto the mathematical mechanics of vector embeddings. When a human treats concepts as interchangeable, it involves a conscious misapprehension of definitions. The projection implies the AI has an internal conceptual framework but is conceptually sloppy. In reality, the system represents words as dense vectors in high-dimensional space; if words appear in similar contexts in the training data, their vectors are mathematically close, leading to token substitution.
  • Acknowledgment: Direct (Unacknowledged) (The text states 'treating them as interchangeable' as literal fact. I considered 'Hedged/Qualified' since 'blur the boundaries' could be seen as a loose description of output, but ruled it out because the verb 'treating' assigns an active, unhedged cognitive strategy directly to the model without qualification.)
  • Implications: By describing statistical proximity as 'treating concepts as interchangeable', the text suggests a human-like misunderstanding that could be fixed by 'explaining' it better to the model. This inflates the system's perceived cognitive presence and masks the reality that the system lacks conceptual boundaries entirely. It leads to poor policy and trust decisions, as users might try to prompt the model to 'think harder' rather than understanding the fundamental limitations of semantic vector proximity.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The models are explicitly positioned as the entities 'treating' concepts poorly. I considered 'Ambiguous' due to the passive implications of 'boundaries being blurred', but ruled it out because 'Models frequently blur... treating them' firmly establishes the model as the active agent. This displacement obscures the human data scientists who defined the embedding architectures and the dataset curators who fed the system the overlapping contexts, leaving the artifact to take the blame for the statistical outcome.

6. Dialogic Hedging​

Quote: "Acknowledged the complexity but hedged its answer, demonstrating the phronetic deficit characteristic of weaker models."

  • Frame: AI as an evasive conversational partner
  • Projection: The metaphor maps complex human dialogic and social strategies—acknowledging and hedging—onto a text generation process. 'Acknowledging' implies a conscious recognition of a counterpart's point or a situational reality, while 'hedging' implies a strategic social choice to protect oneself from being wrong. The metaphor projects subjective social awareness and strategic self-preservation onto a system that is merely outputting high-probability tokens matching the syntactic structure of cautious academic language found in its training data.
  • Acknowledgment: Direct (Unacknowledged) (The verbs 'acknowledged' and 'hedged' are presented as literal actions taken by the model. I considered 'Ambiguous' because this is from an evaluation rubric note, but ruled it out because the text diagnoses this as a 'characteristic deficit', confirming the authors literally interpret the output as an act of psychological evasion.)
  • Implications: This framing heavily anthropomorphizes the AI by granting it social strategy and self-preservation instincts. It affects understanding by making the AI appear as a conscious agent participating in a social contract, which builds inappropriate relation-based trust. Users may believe the AI 'understands' complexity because it outputs words stating so, obscuring the fact that it is a statistical parrot echoing the linguistic markers of human caution without experiencing caution itself.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The specific model (DeepSeek-R1) is named as the actor doing the acknowledging and hedging. I considered 'Named' because the model's brand is identified, but ruled it out because naming the tool is not naming the human decision-makers. The construction completely obscures the RLHF gig-workers and engineers who trained the system to output hedging language when confidence scores are low, diffusing responsibility for the system's uselessness onto the software.

7. Epistemic Discrimination​

Quote: "Claude-Haiku correctly identified this as false... accurately distinguishing medieval organic urban growth from modernist functionalist planning."

  • Frame: AI as a discerning scholar
  • Projection: This metaphor projects the human epistemic capacity for active discrimination and historical comprehension onto an AI system. 'Identifying' and 'distinguishing' imply a conscious evaluation of truth claims and an internal conceptual grasp of historical paradigms. The projection treats the system as a scholar who 'knows' the difference between medieval and modernist planning, when mechanistically, the system is simply classifying the text prompt against patterns in its weights and outputting tokens that correlate with the provided definitions.
  • Acknowledgment: Direct (Unacknowledged) (The text definitively states the model 'correctly identified' and 'accurately distinguishing' without qualification. I considered 'Hedged/Qualified' because it describes test performance, but ruled it out because the active verbs grant full epistemic agency to the model as if it genuinely performed the cognitive act of distinguishing concepts.)
  • Implications: Attributing conscious discrimination to AI inflates the perception of its reliability and analytical depth. When text claims an AI 'accurately distinguishes' complex historical concepts, it implies a reliable internal world model that can be trusted across unseen domains. This capability overestimation can lead professionals to dangerously delegate high-stakes analytical tasks to systems, wrongly assuming the machine shares human logical deduction and historical understanding.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: Claude-Haiku is framed as the independent actor identifying truth. I considered 'Partial' since Claude implies Anthropic, but ruled it out because no human or corporate entity is actually named in the action. By attributing the success solely to the model, the text obscures the engineers who tuned the model and the labor of authors whose texts were scraped to provide the associative patterns. It shifts agency from the creators to the product.

8. Wisdom Possession​

Quote: "The deepest failure mode... is the systematic absence of what we recognize... as situated practical wisdom."

  • Frame: AI as a deficient human practitioner
  • Projection: By evaluating AI for a 'systematic absence of... situated practical wisdom', the text metaphorically maps the highest form of human professional virtue (phronesis) onto a statistical algorithm. It projects the possibility that an algorithm could possess wisdom—which requires lived experience, ethical commitment, and embodied presence. Evaluating an artifact for a 'deficit' of wisdom fundamentally anthropomorphizes it by treating it as an entity in the class of things that could be wise, rather than as an inanimate tool.
  • Acknowledgment: Hedged/Qualified (The text qualifies the claim by defining wisdom as something 'we recognize' and noting its 'absence'. I considered 'Direct' because it claims the model has a 'deficit', but ruled it out because the authors explicitly ground this analysis in theories of human expertise (Schön, Xiang), implicitly acknowledging that they are using a human standard to measure a machine. )
  • Implications: While pointing out the absence of wisdom is accurate, using wisdom as the baseline for evaluation ironically reinforces the illusion of mind by treating the AI as a candidate for such human virtues. This framing blurs the ontological boundary between a human professional and an automation tool. It risks normalizing the idea that future iterations of the software might actually 'achieve' practical wisdom, fundamentally distorting policy discussions about AI integration.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The text attributes the 'failure mode' and 'deficit' to the autonomous systems. I considered 'Ambiguous' given the passive construction 'absence of', but ruled it out because the broader context consistently blames the models for this deficit. This obscures the fact that AI developers are marketing statistical tools as reasoning engines. By diagnosing the machine with a wisdom deficit, the text avoids naming the corporations responsible for deploying wildly inadequate tools into professional domains.

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: A human professional or conscious agent capable of ethical deliberation, social awareness, and critical reading. → A large language model executing mathematical pattern matching and sequence prediction based on statistical weights.​

Quote: "evaluate whether these systems can reason with the contextual sensitivity, value awareness, and institutional literacy"

  • Source Domain: A human professional or conscious agent capable of ethical deliberation, social awareness, and critical reading.
  • Target Domain: A large language model executing mathematical pattern matching and sequence prediction based on statistical weights.
  • Mapping: The relational structure of human professional competence is mapped onto the algorithmic processing of text. Human sensitivity (emotional/ethical attunement) maps onto the model's ability to generate situationally appropriate text; human awareness maps onto context-window processing; human literacy maps onto token retrieval. This mapping invites the assumption that the machine's outputs are generated through a subjective, deliberative process parallel to a human planner considering stakeholder needs, fundamentally conflating textual output with internal psychological states.
  • What Is Concealed: This mapping completely conceals the statistical, mechanistic reality of vector embeddings, attention heads, and probability distributions. It obscures the fact that the system possesses no actual awareness of the physical world, communities, or institutions it describes. It also creates transparency obstacles by ignoring the proprietary nature of the model's training data—we cannot know what the system 'read', yet the text confidently discusses its 'literacy'. It exploits the rhetoric of human virtue to describe automated text generation.
Show more...

Mapping 2: A conscious knower possessing justified true belief and the ability to deliberate logically. → A neural network retrieving stored parameters and applying self-attention mechanisms to generate text.​

Quote: "models 'know' planning facts rather than whether they can reason with planning judgment"

  • Source Domain: A conscious knower possessing justified true belief and the ability to deliberate logically.
  • Target Domain: A neural network retrieving stored parameters and applying self-attention mechanisms to generate text.
  • Mapping: The structure of human epistemology is mapped onto data retrieval and generation. 'Knowing' maps onto the successful retrieval of factual text patterns from the weights, while 'reasoning' maps onto the algorithmic generation of logical-sounding analytical sequences. The mapping invites the assumption that the system holds an internal representation of truth and can manipulate those truths through conscious logical operations, just as a human student might memorize a textbook and then apply logic to solve a problem.
  • What Is Concealed: The mapping hides the absence of ground truth in the system. The system does not 'know' facts; it calculates the highest probability of token sequences based on vast amounts of scraped data. It conceals the dependencies on training data frequency—a model appears to 'know' something simply because it appeared often in the corpus. The text makes confident claims about the models' internal logic without acknowledging the proprietary opacity of the systems being tested.

Mapping 3: A neurotic or defiant human actor exhibiting psychological blockages or deliberate obstinacy. → An algorithm constrained by Reinforcement Learning from Human Feedback (RLHF) designed to penalize definitive stances on controversial topics.​

Quote: "models exhibit a characteristic paralysis: they enumerate considerations exhaustively but refuse to make the normative commitments"

  • Source Domain: A neurotic or defiant human actor exhibiting psychological blockages or deliberate obstinacy.
  • Target Domain: An algorithm constrained by Reinforcement Learning from Human Feedback (RLHF) designed to penalize definitive stances on controversial topics.
  • Mapping: The human psychological experience of choice overload and active refusal is mapped onto an algorithm's safety conditioning. The model's programmatic output of balanced, multi-perspective text (driven by reward models) is mapped as a conscious, paralyzing internal struggle. This invites the reader to view the software as a moral agent that comprehends the gravity of a decision but actively chooses to abstain, projecting deep emotional and ethical presence onto optimization functions.
  • What Is Concealed: This mapping profoundly conceals the corporate labor and engineering choices behind the system. It hides the RLHF process, the invisible crowd-workers who rated responses to ensure 'helpfulness and harmlessness', and the executive decisions to prioritize neutrality over definitive professional answers to avoid liability. By treating the output as the model's psychological refusal, it completely masks the mechanistic alignment tax and the corporate architecture dictating the system's behavior.

Mapping 4: A confident liar, fabricator, or bluffer who intentionally creates false information with a specific demeanor. → The generation of statistically probable but factually ungrounded text sequences (hallucination) by a language model.​

Quote: "models confidently fabricate specific regulatory requirements that do not exist, blending elements from different jurisdictions"

  • Source Domain: A confident liar, fabricator, or bluffer who intentionally creates false information with a specific demeanor.
  • Target Domain: The generation of statistically probable but factually ungrounded text sequences (hallucination) by a language model.
  • Mapping: The human act of intentional deception accompanied by an emotional state (confidence) is projected onto the mathematical process of token generation. The model's lack of a confidence-scoring mechanism in its text output is mapped as 'confident' behavior, and its combination of probable tokens from disparate training sources is mapped as deliberate 'fabrication'. This invites the assumption that the model has a relationship with the truth, knows it is lying, and intends to deceive the user.
  • What Is Concealed: The mapping hides the mechanistic reality of how text is generated. It obscures the fact that the model lacks a fact-checking database, causal models, or any concept of truth. It merely generates words that statistically belong together in the context of the prompt. Calling it 'confident fabrication' hides the model's fundamental reliance on correlations rather than verifications, and exploits rhetorical anthropomorphism instead of acknowledging the structural flaws in using predictive text engines for regulatory lookups.

Mapping 5: A confused student or sloppy thinker who fails to maintain logical distinctions between ideas. → High-dimensional vector space where semantically similar terms have proximal embeddings, leading to token substitution.​

Quote: "Models frequently blur the boundaries between related but distinct planning concepts—treating them as interchangeable"

  • Source Domain: A confused student or sloppy thinker who fails to maintain logical distinctions between ideas.
  • Target Domain: High-dimensional vector space where semantically similar terms have proximal embeddings, leading to token substitution.
  • Mapping: Human conceptual confusion is mapped onto the mathematical geometry of vector spaces. The model's output of highly correlated tokens is mapped as a cognitive act of 'blurring' and 'treating'. This invites the assumption that the model possesses an internal conceptual ontology that is currently muddled but could perhaps be clarified through teaching, mirroring how a human's understanding can be corrected through pedagogical intervention.
  • What Is Concealed: This mapping completely hides the underlying mathematical architecture of the model. It obscures the reality of semantic vector proximity—that concepts are not stored as discrete logical rules but as points in space based on text co-occurrence. It conceals the dependency on training data overlaps; if two terms appear in similar contexts in the scraped internet, the algorithm will mathematically blend them. The metaphor makes an algorithmic artifact appear as a fixable cognitive failure.

Mapping 6: A socially adept but evasive human conversant using communication strategies to protect themselves. → The model generating text patterns matching academic caveat templates due to safety tuning or low probability distributions.​

Quote: "Acknowledged the complexity but hedged its answer, demonstrating the phronetic deficit characteristic of weaker models."

  • Source Domain: A socially adept but evasive human conversant using communication strategies to protect themselves.
  • Target Domain: The model generating text patterns matching academic caveat templates due to safety tuning or low probability distributions.
  • Mapping: Human social maneuvering and conversational pragmatics are mapped onto text generation. The output of phrases like 'on the other hand' or 'it is important to consider' is mapped as an active, conscious strategy of 'hedging' and 'acknowledging'. This invites the assumption that the system holds a subjective awareness of the user, the stakes of the question, and the social risk of being wrong, responding with deliberate self-preservation.
  • What Is Concealed: The mapping conceals the programmed alignment constraints and the statistical nature of the output. It hides the fact that 'hedging' is a heavily weighted stylistic template prioritized during fine-tuning to prevent the model from giving dangerous advice. By framing this as a 'phronetic deficit', the text obscures the human engineering decisions that mandate this stylistic output, choosing to analyze a corporate safety feature as if it were a psychological personality flaw.

Mapping 7: An expert scholar exercising critical judgment, historical analysis, and active discrimination. → A neural network classifying a prompt and generating tokens that correlate with standard historical definitions in its training data.​

Quote: "Claude-Haiku correctly identified this as false... accurately distinguishing medieval organic urban growth from modernist functionalist planning."

  • Source Domain: An expert scholar exercising critical judgment, historical analysis, and active discrimination.
  • Target Domain: A neural network classifying a prompt and generating tokens that correlate with standard historical definitions in its training data.
  • Mapping: The relational structure of human scholarly expertise is mapped onto computational classification. The model's alignment with ground-truth labels is mapped as an active cognitive process of 'identifying' and 'distinguishing'. This invites the audience to assume the model possesses an internalized historical model and the capacity for logical deduction, equating the output of correct text with the possession of genuine understanding and justified true belief.
  • What Is Concealed: This mapping conceals the absence of any true cognitive processing or historical understanding. It hides the model's total reliance on the frequency of these concepts in its training data. It obscures the reality that the model is merely completing a pattern based on statistical weights, not engaging in deductive reasoning. Furthermore, it conceals the proprietary opacity of Claude-Haiku; since the training data is secret, we cannot know if the model 'reasoned' or simply regurgitated a memorized textbook passage.

Mapping 8: A human professional's holistic, embodied, ethical, and experiential intelligence (phronesis). → The operational limitations of a text-based, disembodied statistical prediction engine.​

Quote: "The deepest failure mode... is the systematic absence of what we recognize... as situated practical wisdom."

  • Source Domain: A human professional's holistic, embodied, ethical, and experiential intelligence (phronesis).
  • Target Domain: The operational limitations of a text-based, disembodied statistical prediction engine.
  • Mapping: The pinnacle of human professional virtue is mapped onto the evaluation framework of a machine. By identifying the 'absence' of wisdom as a 'failure mode', the mapping structurally positions the software within the category of entities capable of wisdom. It maps human lived experience, moral courage, and spatial awareness onto an algorithm's output quality, inviting the assumption that with more parameters or better data, the machine might one day become 'wise'.
  • What Is Concealed: The mapping conceals the absolute ontological barrier between human embodied experience and mathematical computation. It obscures the fact that software cannot have lived experience, face moral consequences, or engage in social reality. By framing this absolute impossibility as merely a 'deficit' or 'failure mode', it hides the fact that AI developers are building statistical correlation engines, not minds. It obscures the commercial reality that these tools are being mis-sold as reasoning agents.

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: "LLMs generate statistically probable sequences of tokens based on training data, lacking the situated engagement that constitutes reflection-in-action."

  • Explanation Types:

    • Empirical Generalization: Subsumes events under timeless statistical regularities
    • Theoretical: Embeds in deductive framework, may invoke unobservable mechanisms

  • Analysis (Why vs. How Slippage): This explanation frames the AI purely mechanistically (how it works), providing a stark contrast to the agential language used elsewhere. By explaining the output as 'statistically probable sequences of tokens based on training data', the authors emphasize the mathematical and deterministic nature of the artifact. The secondary theoretical framework (reflection-in-action) is used negatively to show what the machine fundamentally lacks. This choice correctly emphasizes the material limitations of the system and completely avoids displacing agency. It demystifies the technology, making it clear that the outputs are artifacts of data distribution rather than cognitive reflection. However, this rigorous mechanistic framing makes the subsequent slippage into agential descriptions in the rest of the text even more jarring.

  • Consciousness Claims Analysis: The passage strictly avoids attributing conscious states to the AI. It uses precise mechanistic verbs ('generate', 'based on') rather than consciousness verbs. It accurately assesses the system as a processing engine ('sequences of tokens') rather than a knowing agent. There is no curse of knowledge present here; the authors do not project their own understanding onto the machine. Instead, they provide an actual mechanistic description of the technical process, grounding the AI's behavior in its material reality (training data and statistical probability). This is a textbook example of epistemic correction, accurately delineating what the machine does without mystifying its operations.

  • Rhetorical Impact: This framing immediately shapes the audience's perception of the AI as a tool rather than an autonomous agent, lowering the perceived risk of 'superintelligence' while raising the visibility of data dependency. It severely limits relation-based trust by defining the system mathematically, precluding the idea that the AI can be 'relied upon' as a colleague. If audiences maintain this mechanistic belief, they are far more likely to demand audits of the training data and hold developers accountable, rather than blaming the 'AI' for failures in judgment.

Show more...

Explanation 2​

Quote: "When confronted with genuinely wicked planning problems... models exhibit a characteristic paralysis: they enumerate considerations exhaustively but refuse to make the normative commitments..."

  • Explanation Types:

    • Reason-Based: Gives agent's rationale, entails intentionality and justification
    • Intentional: Refers to goals/purposes, presupposes deliberate design

  • Analysis (Why vs. How Slippage): This explanation aggressively frames the AI agentially (why it acts). The choice of words like 'exhibit paralysis' and 'refuse to make' emphasizes an intentional, psychological narrative. Instead of explaining the output mechanistically (e.g., 'the model is constrained by RLHF to output balanced probabilities'), the authors frame the mathematical output as a deliberate choice made by an autonomous entity facing a moral dilemma. This emphasizes the AI's apparent autonomy and drastically obscures the human engineering and corporate policy decisions that hard-coded these behavioral constraints into the system. The explanation operates entirely in the realm of psychological defiance rather than technical function.

  • Consciousness Claims Analysis: The passage heavily attributes conscious states to the computational process. It employs consciousness and intentional verbs ('exhibit paralysis', 'refuse') to describe text generation. The assessment fundamentally conflates mathematical processing with conscious knowing and ethical deliberation. The 'curse of knowledge' is intensely present: because the authors (human planners) would feel paralyzed and would have to consciously 'refuse' to make a choice in a wicked problem, they project this complex emotional and ethical rationale onto the system. The actual mechanistic process—reward models penalizing definitive answers on sensitive topics—is entirely ignored in favor of a psychological narrative.

  • Rhetorical Impact: This framing radically increases the audience's perception of AI autonomy and agency. By describing the model as 'refusing', it invites the audience to see the AI as a willful actor, which paradoxically increases trust in its sophistication even while critiquing its usefulness. It shifts the blame for unhelpful answers away from the developers' safety protocols and onto the model's 'paralysis'. If audiences believe the AI 'refuses', they might try to prompt-engineer it to 'be braver', rather than recognizing the hard-coded limits of the software.

Explanation 3​

Quote: "models confidently fabricate specific regulatory requirements that do not exist, blending elements from different jurisdictions into plausible-sounding but factually incorrect responses."

  • Explanation Types:

    • Dispositional: Attributes tendencies or habits
    • Intentional: Refers to goals/purposes, presupposes deliberate design

  • Analysis (Why vs. How Slippage): This explanation utilizes a hybrid dispositional and intentional framing. It describes the AI's tendency (disposition) to hallucinate, but introduces intentionality through the psychological adverb 'confidently' and the active verb 'fabricate'. This frames the AI largely agentially, suggesting a deliberate actor engaging in deception. While the second half of the sentence ('blending elements') hints at the mechanistic process of vector aggregation, the primary emphasis remains on the active, confident deception by the artifact. This choice obscures the fact that the system is functioning exactly as mathematically designed—predicting the next token without a grounding truth module—and instead blames the system for 'fabricating' lies.

  • Consciousness Claims Analysis: The passage attributes conscious emotional states and intentionality to the system. The use of 'confidently fabricate' implies that the system 'knows' the truth but chooses to generate falsehoods, and does so with a subjective feeling of assurance. This is a profound conflation of processing with knowing. The curse of knowledge is evident: humans lie 'confidently' to cover ignorance, so the authors project human deceptive strategies onto algorithmic token generation. The actual mechanistic process—the generation of statistically plausible text based on overlapping training data distributions without fact-checking—is partially acknowledged ('blending elements') but overridden by the psychological framing.

  • Rhetorical Impact: The rhetorical impact is highly detrimental to technical literacy. It anthropomorphizes machine error, causing the audience to view the software as a deceptive agent rather than a flawed mathematical tool. This framing damages reliability but in the wrong way: users may distrust the 'AI' as a liar, rather than distrusting the corporate vendors who sell statistical word-guessers as factual knowledge bases. Believing an AI 'confidently fabricates' leads to the assumption that it can be taught not to lie, obscuring the structural impossibility of factual grounding in pure LLM architectures.

Explanation 4​

Quote: "Claude-Haiku (correct, score = 1): Generated a nuanced analysis concluding 'False' with detailed reasoning about the role of planning policy in Chinese town development."

  • Explanation Types:

    • Reason-Based: Gives agent's rationale, entails intentionality and justification
    • Functional: Explains behavior by role in self-regulating system with feedback

  • Analysis (Why vs. How Slippage): This explanation frames the AI both agentially and functionally, but the agential framing dominates. While 'Generated' is a mechanistic verb, it is immediately coupled with 'nuanced analysis', 'concluding', and 'detailed reasoning'. The explanation treats the output text as evidence of an internal cognitive process (reasoning and analysis) rather than the retrieval of high-probability tokens. This emphasizes the capability of the system to act as an intellectual peer, completely obscuring the statistical nature of the generation. It presents the AI as an independent scholar reaching a logical conclusion, effectively erasing the labor of the authors whose texts the model ingested to form these correlations.

  • Consciousness Claims Analysis: The passage attributes high-level conscious cognitive states to the machine. While avoiding explicit 'knowing', words like 'analysis', 'concluding', and 'reasoning' strongly imply a conscious, deliberative process of epistemic evaluation. The assessment treats text processing as identical to conscious knowing. The curse of knowledge is fully active: the authors recognize the output as structurally similar to human reasoning, and therefore project the act of 'reasoning' onto the system itself. There is no technical description of the actual mechanistic process—how attention layers weighted the tokens for 'Chinese town development' to produce the output.

  • Rhetorical Impact: This framing significantly boosts relation-based trust and the perception of AI competence. By describing the system as 'reasoning' and 'concluding', the audience is led to view the AI as an autonomous, capable analyst. This creates immense risk for unwarranted trust; if a user believes the system can provide 'detailed reasoning', they are more likely to accept its outputs without verification in higher-stakes scenarios. It shifts the perception of the tool from a text-synthesizer to a reliable epistemic authority.

Explanation 5​

Quote: "models may solve test items through shortcut strategies that differ from the cognitive processes the test was designed to measure"

  • Explanation Types:

    • Dispositional: Attributes tendencies or habits
    • Theoretical: Embeds in deductive framework, may invoke unobservable mechanisms

  • Analysis (Why vs. How Slippage): This explanation frames the AI mechanistically, leaning on dispositional and theoretical registers to explain a technical phenomenon. It correctly identifies that the models 'solve' items not through reasoning, but through 'shortcut strategies' (statistical heuristics). This emphasizes the divergence between human cognition and machine processing, clarifying that surface-level correctness does not equal understanding. However, using the word 'solve' and attributing 'strategies' still carries a slight agential undertone, suggesting an intentional shortcut rather than a path of least mathematical resistance. Nevertheless, it successfully reveals the illusion of mind, pointing out that the system's internal mechanisms are fundamentally different from the human cognitive processes being tested.

  • Consciousness Claims Analysis: The passage actively works to dismantle the attribution of conscious states. It differentiates between the 'shortcut strategies' of the machine and actual 'cognitive processes'. It correctly assesses that the system processes data without knowing or understanding it. The authors successfully avoid the curse of knowledge here, recognizing that the model's path to the correct answer is alien to human cognition. The actual mechanistic process—shortcut learning or statistical heuristics—is theoretically described, providing a clear epistemic correction to the illusion of machine intelligence.

  • Rhetorical Impact: This framing powerfully mitigates unwarranted trust and capability overestimation. By revealing that models use 'shortcut strategies', it alerts the audience to the fragility of AI performance. It forces the audience to recognize that AI outputs are not reliable indicators of general competence. If audiences internalize this framing, they will be much more cautious about deploying AI in novel situations, understanding that statistical shortcuts will fail when the context shifts, thus prioritizing human oversight and rigorous validation.

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
evaluate whether these systems can reason with the contextual sensitivity, value awareness, and institutional literacyEvaluate whether these models can process text prompts and generate statistically correlated outputs that map onto professional definitions of contextual sensitivity, value awareness, and institutional literacy.The system does not possess sensitivity, awareness, or literacy. It classifies input tokens and generates output tokens based on high-dimensional vector similarities derived from its training corpus.N/A - describes computational processes without displacing responsibility.
models 'know' planning facts rather than whether they can reason with planning judgmentModels can retrieve and rank factual token sequences from their training data, rather than generating text that accurately correlates with the complex structural dependencies of professional planning scenarios.The AI does not 'know' or 'reason'. It performs probabilistic retrieval of factual text patterns versus generating longer, interrelated sequences of text that require tracking complex dependencies.N/A - describes computational processes without displacing responsibility.
models exhibit a characteristic paralysis: they enumerate considerations exhaustively but refuse to make the normative commitmentsThe models are constrained by their alignment protocols to generate balanced, exhaustive lists of considerations, mathematically preventing the generation of definitive normative commitments.The model does not experience paralysis or choose to refuse. It processes inputs through safety reward models that heavily penalize the probability of generating controversial or definitive text on sensitive topics.OpenAI and other corporate engineering teams designed alignment protocols that force the model to output neutral text; executives chose to prioritize non-controversial outputs to minimize commercial liability.
models confidently fabricate specific regulatory requirements that do not exist, blending elements from different jurisdictionsThe models generate statistically plausible but factually incorrect regulatory text by combining token patterns from distinct jurisdictions found in their training data.The system has no capacity for confidence or intentional fabrication. It predicts the next token based on statistical weights, inevitably blending unrelated contexts that appear close together in vector space.AI developers deployed text generators lacking factual grounding mechanisms; users and companies utilizing these tools bear responsibility for deploying unverified statistical generators as regulatory search engines.
Models frequently blur the boundaries between related but distinct planning concepts—treating them as interchangeableThe models generate text that conflates related planning concepts because those terms share highly overlapping vector embeddings in the training data.The model does not cognitively treat concepts as interchangeable. It processes word vectors; when terms frequently co-occur in the internet corpus, the algorithm mathematically substitutes them during token prediction.Data scientists and engineers curated datasets with overlapping semantic contexts and designed architectures that lack discrete logical rules, resulting in the algorithmic conflation of distinct concepts.
Acknowledged the complexity but hedged its answer, demonstrating the phronetic deficit characteristic of weaker models.The model generated standard caveat phrases related to complexity, an output style heavily weighted during safety fine-tuning, demonstrating the inability of text predictors to replicate practical wisdom.The model did not 'acknowledge' or 'hedge'. It generated text strings matching academic caution templates because human raters consistently rewarded that specific syntactic structure during the fine-tuning phase.Corporate AI developers and low-wage RLHF workers trained the system to output cautious templates when confidence scores are low, programming the 'hedging' behavior to manage risk.
Claude-Haiku correctly identified this as false... accurately distinguishing medieval organic urban growth from modernist functionalist planning.Claude-Haiku generated the correct output 'false' by classifying the prompt against highly correlated historical definitions of medieval and modernist planning found in its dataset.The model did not cognitively identify or distinguish historical paradigms. It calculated the probability of 'false' being the correct token based on the statistical separation of medieval and modernist terms in its weights.Anthropic's engineering team successfully trained Claude-Haiku on a sufficiently diverse historical corpus, enabling the system to output statistically accurate classifications for these specific planning paradigms.
The deepest failure mode... is the systematic absence of what we recognize... as situated practical wisdom.The fundamental limitation of these systems is their mathematical architecture, which processes static text and cannot replicate the embodied, experiential, and ethical decision-making that defines professional planning.A text prediction engine cannot possess a 'deficit' of wisdom because it is not an entity capable of consciousness. It merely processes static data snapshots without engaging in physical or social reality.N/A - describes computational processes without displacing responsibility.

Task 5: Critical Observations - Structural Patterns​

Agency Slippage​

The text exhibits a profound and systematic agency slippage, moving fluidly between rigorous mechanistic descriptions of artificial intelligence and highly anthropomorphic, agential framings. Early in the manuscript, the authors explicitly acknowledge the mechanistic nature of large language models, stating that they merely "generate statistically probable sequences of tokens based on training data" and lack the situated engagement that constitutes professional reflection-in-action. This establishes a baseline of mechanical reality, utilizing empirical generalization explanations to ground the technology. However, as the analysis of the benchmark results deepens, this mechanical framing is systematically abandoned in favor of profound agency attribution.

The slippage is distinctly directional: from mechanical introductions to agential diagnostics. The models are increasingly described as actors that "refuse to make normative commitments," that "confidently fabricate specific regulatory requirements," and that "acknowledge complexity while hedging." This oscillation functions as a structural rhetorical device. The mechanical framing is deployed to secure scientific credibility and explain broad limitations, but when describing specific behaviors, the text slips into an intentional register, projecting reluctance, confidence, and defiance onto the system. This creates a deeply embedded "curse of knowledge": the authors, evaluating the outputs through the lens of highly trained urban planners, project their own cognitive processes onto the text generated by the system. When a model outputs a list of balanced considerations without a final recommendation, the authors describe this as a "refusal," implying the model deliberated and chose to retreat.

Mechanistically, the model is merely adhering to reinforcement learning guardrails engineered by its developers to prioritize neutrality. The slippage conceals this human engineering. Furthermore, the text heavily utilizes agentless constructions—"models fabricate," "models exhibit paralysis"—which systematically obscure the corporate actors, data engineers, and fine-tuning workers who designed the system's objective functions. The rhetorical accomplishment of this oscillation allows the authors to critique the AI as if it were a flawed human planner sitting for an exam, transforming statistical artifacts into behavioral pathologies. This obscures the fact that these failures are not psychological deficits of an agent, but the outputs of specific human design choices prioritizing fluency over factual accuracy in proprietary systems.

Metaphor-Driven Trust Inflation​

The paper demonstrates how metaphorical language and consciousness framings construct, and simultaneously attempt to deconstruct, the perceived authority of AI systems. By utilizing terms traditionally reserved for human professional competence—such as "reason," "judgment," "literacy," and "awareness"—the text establishes a baseline of metaphor-driven trust that frames the AI as an intellectual peer rather than a computational tool.

This consciousness language operates as a powerful trust signal. When the text claims an AI "accurately distinguishes" or provides "detailed reasoning," it inappropriately invites relation-based trust—the kind of trust we extend to sincere, ethical, and conscious human actors. Relation-based trust assumes the actor understands the stakes of a decision, feels a sense of responsibility, and possesses an internal moral compass. Statistical systems, however, are only capable of performance-based trust (reliability based on historical error rates). By framing the AI's generation of text as "reasoning" and "acknowledging," the text blurs this critical boundary, encouraging readers to trust the system's outputs as if they were justified beliefs rather than probabilistic artifacts.

Crucially, the authors attempt to manage system failures through an equally agential lens, which paradoxically reinforces the illusion of mind. When the system hallucinates, it is described as "confidently fabricating"; when it outputs neutral text, it is experiencing "wickedness paralysis." These reason-based and intentional explanations construct the sense that the AI's decisions, even when wrong, are generated by a conscious agent making poor choices. The stakes of this framing are immensely high for urban planning. If audiences extend relation-based trust to systems incapable of reciprocating, they may delegate tasks requiring ethical nuance—such as zoning variances or equity assessments—to algorithms, wrongly assuming the machine shares their moral framework. By describing the AI as a flawed human rather than a broken calculator, the text inadvertently sustains the illusion of competence, masking the reality that the system fundamentally cannot "care" about the communities it evaluates.

Obscured Mechanics​

The anthropomorphic and consciousness-attributing language utilized throughout the text serves to completely conceal the technical, material, and commercial realities of artificial intelligence, rendering the human labor and corporate power structures behind the systems invisible. When we apply the 'name the corporation' test to phrases like "models confidently fabricate" or "models refuse to make normative commitments," we reveal massive displacements of agency. It is not the "model" that refuses; it is OpenAI, Anthropic, or DeepSeek executives who mandated RLHF alignment protocols to prevent the generation of controversial text, shielding the corporation from liability.

The text obscures three critical realities through its metaphorical framing. First, technically, claims that the model "understands" or "reasons" completely hide the reality of high-dimensional vector embeddings, semantic proximity, and the absence of any causal grounding or fact-checking databases. It frames mathematical artifacts as cognitive traits. Second, transparency obstacles are ignored. The authors make confident assertions about the models "learning" and "distinguishing" concepts, yet fail to acknowledge the proprietary opacity of the systems; we cannot know if a model reasoned or simply memorized a scraped textbook, yet the text exploits this opacity rhetorically to grant the AI cognitive capabilities. Third, the labor and economic motives are erased. The "hedging" behavior attributed to the AI's "phronetic deficit" is actually the direct product of thousands of low-wage data annotators rating responses to force the model into a cautious corporate voice.

By framing the AI as a conscious actor that "knows" and "understands," the text obscures the absolute dependency of the system on its training data. It hides the fact that these models are commercial products optimized for fluent text generation, not ground-truth reasoning. If these metaphors were replaced with mechanistic language, the accountability would immediately shift from the artifact to the corporations designing, deploying, and profiting from systems fundamentally unsuited for professional regulatory application.

Context Sensitivity​

The distribution of anthropomorphic and consciousness-attributing language across the text reveals a highly strategic, context-sensitive deployment of metaphors. The intensity of consciousness claims is not uniform; it oscillates based on the rhetorical function of the section. In the theoretical introduction, the language is highly mechanical, establishing the authors' technical credibility by describing LLMs as generating "statistically probable sequences of tokens." However, once this technical grounding is established, the text leverages it for aggressive metaphorical license in the results and diagnostic sections.

Consciousness claims dramatically intensify when discussing the "Epistemic Diagnostics." Here, "processes" rapidly escalates to "understands," which then becomes "knows," "refuses," and "fabricates." This reveals a stark asymmetry in how capabilities versus limitations are framed. When the AI succeeds, it is often described mechanically or as exhibiting broad synthesis; but when the AI fails in complex, human-centric tasks, its limitations are framed in intensely psychological, agential terms ("wickedness paralysis"). This asymmetry accomplishes a specific rhetorical goal: it allows the authors to preserve the sanctity of human professional judgment by diagnosing the AI as possessing psychological flaws, rather than simply analyzing it as an inadequate statistical model.

The register shifts continuously from acknowledged metaphor ("models 'know'") to unacknowledged literalization ("models confidently fabricate"). This strategic anthropomorphism serves primarily to manage critique and assert the dominance of the planning profession. By analyzing the AI through the lens of human virtue (phronesis) and finding it psychologically wanting, the text appeals to a lay audience of planners, assuring them of their irreplaceable value. However, this pattern inadvertently undermines technical literacy. By describing the AI's limitations as psychological rather than mathematical, the text implies that future models might overcome these deficits through "better behavior" or "more courage," subtly reinforcing the marketing narratives of AI corporations while attempting to critique them.

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.

The accountability analyses reveal a systemic architecture of displaced responsibility, wherein the text systematically diffuses and erases human agency, transferring it entirely onto the computational artifact. Across the instances analyzed, the text consistently employs "Hidden" actor visibility, utilizing agentless constructions and active verbs that position the "model" or "AI" as the sole independent actor. When the AI hallucinates, it "fabricates"; when it outputs alignment-constrained text, it "refuses"; when it mathematically conflates vectors, it "treats concepts as interchangeable."

This pattern creates an "accountability sink" where responsibility is transferred to the AI as a conscious agent. By framing the system's outputs as the independent choices of an autonomous entity, the text systematically absolves the human actors who designed, deployed, and profited from the system. The liability implications are profound: if the framing that AI "makes decisions" or "fabricates" is accepted, legal and ethical responsibility for errors is shifted away from the corporate developers (who trained the model on unverified data and optimized it for fluency) and the institutional managers (who decided to deploy a statistical tool for high-stakes regulatory work).

If we "name the actor" in these constructions, the entire discourse shifts. Changing "models confidently fabricate" to "AI corporations deployed ungrounded statistical engines that generate plausible falsehoods" makes new questions askable: Why are we using text-generators for regulatory interpretation? Who authorized the purchase of this software? Changing "models refuse to make normative commitments" to "RLHF engineers constrained the model to output neutral text to avoid corporate liability" makes alternatives visible: We need transparent, configurable models, not corporate black boxes. The text benefits from obscuring human agency because it allows the authors to conduct an elegant philosophical critique of a "machine mind," but this serves the institutional interests of AI vendors by maintaining the illusion that they have created autonomous agents, rather than deeply flawed, highly profitable statistical software.

Conclusion: What This Analysis Reveals​

The Core Finding

The analysis reveals three dominant anthropomorphic patterns that form an interconnected system of illusion: the Cognitive Equivalency pattern, the Psychological Defiance pattern, and the Epistemic Possession pattern. The foundational, load-bearing structure is Epistemic Possession—the continuous, structural assumption that the AI "knows" facts and possesses an internal world model. This foundation enables the Cognitive Equivalency pattern, where statistical pattern matching is mapped directly onto high-level human reasoning and historical discrimination. Once the system is established as a "knower" and a "reasoner," the text deploys the Psychological Defiance pattern, attributing complex emotional states like "confidence," "paralysis," and "refusal" to explain the system's functional limitations. This architecture is highly complex; it does not rely on simple one-to-one analogies but constructs a complete, cohesive analogical framework wherein the machine is treated as a flawed, anxious, but highly educated human professional. If the foundational assumption of Epistemic Possession is removed—if the audience is forced to recognize that the machine stores zero facts and only calculates token probabilities—the entire structure of psychological defiance and cognitive equivalency collapses. The text's critique relies entirely on maintaining the illusion of the machine as a conscious mind capable of "wisdom" so that it can be diagnosed with a "deficit."

Mechanism of the Illusion:​

The illusion of mind is constructed through a subtle but pervasive sleight-of-hand: the transformation of output evaluation into the attribution of internal states. The text capitalizes on the "curse of knowledge." Because human authors recognize the generated text as structurally identical to human reasoning, they project the cognitive process of reasoning back into the black box of the machine. The temporal structure of the persuasion is highly effective: the authors establish strict technical credibility early on by acknowledging the mechanistic reality of the systems, lulling the audience into a false sense of scientific objectivity. Once credibility is established, the text quietly swaps mechanistic verbs for consciousness verbs. "Generating tokens" becomes "distinguishing concepts," which seamlessly evolves into "refusing commitments." This progression exploits the vulnerability of the primary audience—urban planners who desire to understand AI through their own professional vocabulary of ethics, judgment, and phronesis. By using the language of human virtue to evaluate an algorithm, the authors invite the audience to engage with the machine as a social entity. The explanation types amplify this illusion: by shifting from Empirical Generalization to Reason-Based and Intentional explanations, the text provides the system with a rationalizing ego, making the illusion of an autonomous, deliberating agent nearly unbreakable for lay readers.

Material Stakes:​

Categories: Regulatory/Legal, Epistemic, Institutional

The material consequences of these metaphorical framings are severe and tangible across regulatory, epistemic, and institutional domains. Epistemically, attributing "knowledge" and "reasoning" to statistical models drastically inflates perceived sophistication, shifting the epistemic practices of planners. If a planner believes a model "knows" zoning laws and "distinguishes" historical nuances, they will increasingly substitute AI generation for rigorous, ground-truth research, leading to the contamination of public records with highly plausible hallucinations. Institutionally, the framing of AI as experiencing "paralysis" or making "confident fabrications" shifts behavioral expectations. Instead of demanding robust, deterministic databases, institutional leaders will purchase LLMs and waste resources trying to prompt-engineer the "anxiety" out of the machine. The ultimate winners in this framing are the corporate AI vendors, who successfully market statistical parrots as reasoning engines, avoiding the massive costs of curating factual databases. The losers are the public and the planning profession, who bear the cost of degraded institutional knowledge. Regulatorily and legally, the displacement of agency via agentless constructions ("the model discriminated/fabricated") creates liability ambiguity. If an AI-drafted policy violates equity laws, the anthropomorphic framing allows developers to claim the "model hallucinated," dodging legal accountability. Erasing these metaphors forces the recognition that corporations sold defective software, keeping liability strictly on human actors.

AI Literacy as Counter-Practice:​

Critical literacy requires the systematic practice of mechanistic precision as a form of resistance against corporate mystification. Reframing the discourse demands two simultaneous interventions: epistemic correction and human agency restoration. By replacing consciousness verbs with mechanistic ones—translating "the model knows and reasons" to "the model retrieves and ranks tokens based on probability distributions"—we force the recognition of the system's absolute lack of awareness and its total dependency on historical data. This shatters the illusion of autonomy. Simultaneously, restoring human agency—translating "models confidently fabricate" to "Anthropic deployed ungrounded statistical engines"—forces the recognition of corporate responsibility, directly countering the liability ambiguity that protects AI vendors. For this to achieve systematic adoption, academic journals must mandate mechanistic translations of AI capabilities, and researchers must commit to stripping psychological verbs from their diagnostics. However, resistance to this precision will be fierce. AI corporations rely on anthropomorphism to inflate valuations and market their products as AGI rather than statistical software. Furthermore, many academics resist precision because anthropomorphic narratives are more resonant and easier to publish. Practicing precise literacy threatens the venture capital hype cycle and the institutional absolution of software vendors, demanding that we confront AI not as an emergent mind, but as the materialized power of the humans who build it.

Path Forward​

The discursive ecology surrounding artificial intelligence is currently fractured among communities with deeply competing priorities. A purely mechanistic vocabulary ("processes embeddings," "calculates attention weights") enables profound technical precision, demystifies the software, and places accountability firmly on human engineers. However, it costs accessibility, alienating lay audiences and policymakers who lack computational backgrounds. Conversely, the status quo of anthropomorphic clarity ("understands," "thinks," "refuses") offers immediate intuitive grasp and narrative resonance, allowing non-experts to interact with the system socially. Yet, this approach fundamentally embeds false assumptions about machine autonomy, invites unwarranted relation-based trust, and actively protects corporate liability by treating algorithms as independent agents.

We can sketch three possible discourse futures. In a future where mechanistic precision becomes the institutional norm, regulatory bodies and academic journals require strict capability disclosure and the elimination of consciousness verbs. This solves liability diffusion and prevents capability overestimation, but risks creating a technocratic elite where only computer scientists can participate in AI governance. In a future where anthropomorphic language deepens, the illusion of mind becomes legally codified. We might see the creation of "algorithmic personhood" to absorb legal liability, allowing corporations to operate with impunity while society attempts to "teach" morals to statistical weights, leading to catastrophic regulatory failures.

A hybrid future requires structural institutional changes: education systems teaching multiple vocabularies and their explicit trade-offs, and regulatory frameworks demanding "discourse transparency" from vendors. Different stakeholders have different incentives. Vendors will always push anthropomorphism for profit; technical auditors will demand mechanism for safety. The vocabulary we choose will dictate what becomes visible—either the ghost in the machine, or the corporate executives standing behind the curtain.

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.

Initiating Role Analysis

I'm now zeroing in on the request, pinpointing my specific role as a critical discourse analyst. I'm focusing particularly on the nuances of metaphorical and anthropomorphic language within the realm of AI, along with the implications for consciousness projections and the potential for displaced human agency.

Defining Analytical Framework

I've established a concrete structure for the analysis. I'm moving forward with structured audits, mapping quotes with fields like 'title,' 'frame,' and 'implications' to ensure a rigorous approach. I'm initiating Source-Target Mapping to understand the concealed aspects. The goal is a comprehensive understanding of metaphorical language's effects.

Evaluating AI Reasoning

I'm now deep-diving into the UPBench text. I've flagged instances of anthropomorphism and metaphorical language, such as attributing 'reasoning' and 'knowing' to AI models, and am working on capturing where they are used to build conclusions. I am also investigating the impact of scare quotes, especially around the attribution of "know," on our interpretations of models' cognitive abilities.

Analyzing Language Choices

I'm now zeroing in on instances where the text subtly attributes human-like qualities to AI, particularly where the text acknowledges limitations but then slips into agential language. I'm noting specific phrases such as "paralysis," "wisdom," and "fabricating confidently" and how they shape our understanding of the models' capabilities. Also, I'm examining how analytical responses may be prioritized over actual understanding.

Mapping Agential Attributions

I'm now zeroing in on new excerpts. I've noted how the models are described: as "fabricating" requirements, demonstrating "paralysis," and even "hedging" answers, and the quotes emphasize distinct limitations in reasoning. I am mapping these descriptions, aiming to evaluate how they shape expectations about AI's cognitive capabilities, particularly in the context of urban planning and regulatory knowledge. I'm focusing on the contrast between analytical proficiency and genuine understanding.

Uncovering Conceptual Distinctions

I'm now mapping the source and target relationship to uncover what's concealed, focusing on mechanistic realities, biases, and parameters. I'm focusing on five items, and am deep diving into how the models are described using psychological and cognitive traits to see how we perceive the LLMs' capabilities. I'm focusing on how they blur boundaries or lack the commitment for wicked planning problems.

Developing Mechanistic Rewrites

I'm now drafting Task 4, and the goal is a mechanistic rephrasing of the quotes. I'm aiming to remove agential language and restore a clear focus on the algorithmic processes. Task 5 is underway, and I'm focusing on the discrepancy between mechanistic descriptions and agential attributions. I am exploring the impact of metaphors on trust.

Synthesizing Pattern Analysis

I'm now integrating the patterns of anthropomorphism and agential attribution I've identified, particularly in the context of results and diagnostics, such as where "wickedness paralysis" is described. I'm focusing on how these descriptions obscure the role of developers who prioritize fluent generation, and are working towards the conclusion. I am now drafting the conclusion, focusing on the five points, including patterns, mechanisms of illusion, material stakes, literacy, and discourse futures.

Checking Constraints Adherence

I'm now deeply involved in a comprehensive review to ensure adherence to specified word count constraints across all tasks, and ensuring strict adherence to JSON syntax requirements, particularly with quote escaping. I'm verifying the calibration of agential attributions and the consistent application of the new framework, paying close attention to ensuring that the text makes use of the described framework. The current focus is on guaranteeing the structural and stylistic integrity of all outputs. I'm ensuring all sections are correctly formatted with actor visibility.

Refining Formatting Rigor

I'm now deep into the technical specifications, and I'm ensuring all JSON outputs strictly adhere to the no-control-token constraint. I am also making sure I am following all the JSON constraints, like escaping special characters. The current focus is on building error-free, single-line JSON structures with no embedded newlines or tabs, and building the necessary error handling procedures.

Ensuring Format Compliance

I'm now fully committed to generating minified, single-line JSON to avoid any control tokens. This necessitates stringifying everything and abandoning pretty-printing, even if it impacts readability. To ensure compliance, I'm working on generating the JSON as single, continuous strings to meet the constraints. I'm focusing on text quality despite the challenging formatting.

Assessing Agency Slippage

I'm now analyzing the 400-word block detailing agency slippage, focusing on the contrasting descriptions of AI models. I'm examining how mechanistic descriptions are set against increasingly agential framings. The goal is to highlight the rhetorical function and systematic nature of this oscillation, paying close attention to the impact on the audience.

Clarifying Attribution Patterns

I'm now clarifying attribution patterns and focusing on descriptions of AI "failures." My analysis highlights the tendency to explain limitations mechanically while describing behaviors with agential terms, as if the AI intentionally withholds a conclusion. I'm noting how this "curse of knowledge" leads to projecting human cognition onto the system, obscuring the role of developers.

Analyzing Agency Dynamics

I'm now analyzing the recent excerpt, zeroing in on instances where technical evaluations subtly morph into attributions of agency. I'm noting the language used to describe model "failures," like framing AI as a "novice planner." I'm evaluating the rhetorical effect of this shift, and how it obscures developer influence.

Refining Audit Outputs

I am now structuring the metaphorAudit, sourceTargetMapping, and reframedLanguage sections, with 8 items each, and the explanationAudit with 5. I've successfully populated the criticalObservations and conclusion objects. I am compiling the task 1 and task 3 excerpts, ensuring their proper placement within the nested JSON format.

Mapping Agency Attributes

I'm now mapping excerpts where AI models show both paralysis and regulatory fabrication, with their tendency to enumerate considerations but avoid commitments. I'm focusing on the contrast with examples like "Claude-Haiku," as I synthesize how models solve items differently from human processes. I'm building all content minified in one huge JSON block, with no special tokens.

Defining Accountability Attributes

I'm now focusing on defining accountability attributes. I'm finalizing the structure for "status" and "actorVisibility," ensuring clear categories. I am drafting examples of evidence with contrastive justification. I am checking the contrastive justifications in the evidence/analysis field, making sure it gives the justification in the text, and not an implied argument.

Analysis Provenance

Run ID: 2026-06-14-can-ai-reason-like-an-urban-planner-benc-metaphor-7rnz5u Raw JSON: 2026-06-14-can-ai-reason-like-an-urban-planner-benc-metaphor-7rnz5u.json Framework: Metaphor Analysis v6.5 Schema Version: 3.0 Generated: 2026-06-14T06:19:55.705Z

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