Skip to main content

🆕+🤔 Tracing the ongoing emergence of human-like reasoning in Large Language Models

What survives?

🤔 "What survives...?" A rewriting experiment that tests whether anthropomorphic AI discourse can be translated into strictly mechanistic language while preserving the phenomena described.

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. Cognition as Biological Evolution​

Quote: "suggesting that pragmatic reasoning is still an emerging ability in the cognitive toolkit of artificial systems."

  • Frame: Model as evolving organism with psychological interiority
  • Projection: This metaphor projects profound biological maturation, evolutionary psychology, and subjective interiority onto statistical weight updates. By explicitly using the phrase 'cognitive toolkit' and framing reasoning as an 'emerging ability,' the text maps the human developmental sequence of acquiring conscious, justified understanding onto the mechanistic optimization of loss functions. It suggests that artificial systems possess a localized, internal mental space where cognitive instruments reside and evolve. This erases the distinction between processing computational patterns and possessing conscious states of knowing. It invites the audience to imagine an autonomous organism independently developing psychological maturity and genuine comprehension over time, rather than a fixed mathematical matrix generating probabilistic token sequences based on human-curated datasets.
  • Acknowledgment: Hedged/Qualified (The text uses the participle 'suggesting' to introduce the claim, which functions as a semantic hedge. I considered 'Direct (Unacknowledged)' because 'cognitive toolkit' itself is stated without qualification, but the surrounding rhetorical structure frames the entire proposition as an inference rather than an established literal fact.)
  • Implications: Framing algorithmic pattern-matching as an evolving 'cognitive toolkit' significantly inflates the perceived sophistication and autonomy of the system, fundamentally altering human-AI trust dynamics. When audiences believe a system possesses a 'cognitive toolkit,' they are primed to extend relation-based trust—expecting the system to 'know' and 'understand' context, intentions, and moral weight. This creates extreme vulnerability to automation bias and over-reliance, as users project human-like reliability onto statistical text generation. Furthermore, the 'emerging ability' framing creates profound regulatory ambiguity, suggesting that capabilities evolve naturally and inevitably, which implicitly shields developers from accountability for the specific mathematical optimizations they choose to implement.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The agentless construction treats the 'artificial systems' as the sole locus of action, obscuring the specific corporate laboratories, engineering teams, and data annotators who design and tune these architectures. I considered 'Partial' because the broader text mentions human researchers, but in this specific formulation, all human actors are completely erased. This rhetorical displacement serves the interests of technology developers by framing system behaviors as organic phenomena of 'emergence' rather than the direct, deliberate outcomes of highly capitalized engineering choices, thereby creating an accountability sink for system failures.
Show more...

2. Language Models as Independent Actors​

Quote: "LLMs, while undeniably impressive linguistic agents, have cognitive toolkits that remain fundamentally different from those of humans"

  • Frame: Model as autonomous communicative agent
  • Projection: By defining these statistical artifacts as 'linguistic agents,' the text maps the human capacity for intentional, conscious communication onto algorithmic text generation. An 'agent' implies a conscious entity capable of self-directed action, holding justified beliefs, and possessing the subjective will to communicate meaning. This metaphor completely collapses the boundary between 'processing' language (statistically predicting the next likely token based on training distributions) and 'knowing' language (understanding meaning, possessing communicative intent, and engaging in reciprocal discourse). It attributes a subjective awareness to the system, suggesting the AI 'understands' its output in a way that goes far beyond the mechanistic execution of mathematical correlations.
  • Acknowledgment: Direct (Unacknowledged) (The phrase 'undeniably impressive linguistic agents' is stated as a literal, factual premise without any modal qualification or scare quotes. I considered 'Hedged' because the sentence goes on to contrast them with humans, but the classification of LLMs as 'agents' is presented as an unvarnished, undeniable truth within the text.)
  • Implications: Categorizing text generators as 'linguistic agents' profoundly alters the epistemic status of their outputs. If audiences accept that an AI is an 'agent,' they instinctively evaluate its outputs using the frameworks designed for human interaction: assuming sincerity, intentionality, and a commitment to truth. This consciousness projection generates severe risks of unwarranted trust, as users will assume the system 'knows' when it is hallucinating or 'believes' its own claims. It encourages users to treat statistical anomalies not as mathematical errors, but as intentional choices or sophisticated reasoning, thereby vastly overestimating the system's reliability in high-stakes domains.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The term 'linguistic agents' grants full autonomy and agency to the software product itself, entirely erasing the companies (like OpenAI or Anthropic) that designed the algorithms, selected the training data, and deployed the software. I considered 'Ambiguous' but the syntactic elevation of the software to 'agent' status represents a clear, definitive displacement of human responsibility. This construction protects corporate interests by positioning the software as an independent actor, shifting the locus of liability from the human creators who profit from the tool to the inanimate tool itself.

3. Algorithmic Behavior as Psychological Struggle​

Quote: "they nonetheless struggle with meaning-related components of language"

  • Frame: Model as striving, conscious learner
  • Projection: The verb 'struggle' projects intense psychological interiority, conscious effort, and subjective frustration onto a mathematical inability to map certain inputs to accurate outputs. Humans 'struggle' when they possess an awareness of a goal, experience the subjective friction of difficulty, and apply conscious exertion to overcome it. AI systems simply compute activations across a neural network; they do not experience friction, they do not possess goals, and they do not exert conscious effort. This metaphor invites the audience to view the software as an earnest, conscious student attempting to 'know' and 'understand' meaning, thereby masking the reality that the system merely processes vectors and has no conceptual grasp of meaning whatsoever.
  • Acknowledgment: Direct (Unacknowledged) (The verb 'struggle' is deployed exactly as it would be for a human subject, with no qualifying language or meta-commentary. I considered 'Hedged' because it appears in a literature review summarizing other work, but the authors adopt the terminology directly as fact without distancing themselves.)
  • Implications: When an AI's mathematical failure to predict correct tokens is framed as a 'struggle,' it paradoxically increases user empathy and trust. A conscious entity that 'struggles' is viewed as sincere and capable of eventual growth; it is granted grace for its errors. This consciousness projection dangerously masks the brittleness of statistical systems. If audiences believe the system is 'struggling to understand,' they may provide more sensitive data to 'help' it, or they may mistakenly assume that the system grasps the stakes of its failure. This obscures the absolute absence of meaning-making in the system.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: By stating that the models themselves 'struggle,' the text obscures the engineers who designed architectures fundamentally incapable of pragmatic grounding. I considered 'Partial' because the text elsewhere mentions architectural limitations, but in this sentence, the model is the sole struggling actor. This displacement normalizes product failure as a sympathetic psychological trait of an emerging technology rather than a deliberate engineering trade-off made by companies prioritizing scale over precise symbolic logic.

4. Machine Learning as Skill Acquisition​

Quote: "LLMs have acquired formal linguistic competence"

  • Frame: Model as educated cognitive subject
  • Projection: The phrase 'acquired competence' maps the human, conscious process of learning and mastering a domain onto the mechanistic process of weight optimization via gradient descent. Human acquisition of competence involves conscious integration of feedback, justified true belief, and the subjective 'knowing' of a subject matter. By using these terms, the text projects a state of achieved cognitive mastery onto a system that merely processes massive statistical correlations. It suggests that the AI 'knows' grammar and syntax in the same conscious, rule-governed way a human linguist does, rather than simply generating sequences that probabilistically mimic competent human outputs.
  • Acknowledgment: Direct (Unacknowledged) (The claim is stated as a definitive empirical fact, without any mitigating adverbs or scare quotes around 'competence.' I considered 'Explicitly Acknowledged' because the authors contrast 'formal' with 'functional' competence, but the acquisition of the 'formal' variant is treated as literal truth.)
  • Implications: Attributing 'linguistic competence' to an AI system fundamentally misleads audiences about the epistemic nature of the machine's capabilities. It invites the false assumption that because the system displays 'competence' in structure, it must also possess consciousness, awareness, and intent. This leads to profound overestimations of capability, where policy-makers or users assume the system is safe for autonomous deployment in legal, medical, or administrative contexts, failing to realize that this 'competence' is entirely devoid of actual comprehension, truth-verification, or ethical grounding.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The phrasing frames the LLMs as active subjects who have successfully 'acquired' a skill on their own, completely ignoring the massive corporate infrastructure, data scraping, and human-in-the-loop reinforcement learning required to optimize the weights. I considered 'Named' because earlier paragraphs cite human literature, but the specific grammatical construction here isolates the LLM as a self-taught entity. This obscures the labor of data annotators and the extraction of copyrighted data that actually constitute this so-called 'acquisition.'

5. Algorithmic Inability as Cognitive Bias​

Quote: "arguing that the reasoning abilities of LLMs are affected by what we term a Decontextualization Bias"

  • Frame: Model as flawed psychological thinker
  • Projection: This metaphor projects human cognitive fallibility onto mechanical architecture. A 'bias' in human psychology involves a systematic deviation from rationality due to subjective heuristics, emotional weighting, or deeply held, unexamined beliefs. By diagnosing the machine with a 'bias' that affects its 'reasoning abilities,' the text maps conscious psychological processing onto statistical inference. It suggests that the AI 'knows' the data but is subjectively failing to contextualize it, when in reality, the mechanistic process cannot integrate context because it does not experience the world; it strictly processes discrete tokens based on training distributions.
  • Acknowledgment: Explicitly Acknowledged (The phrase 'what we term a Decontextualization Bias' explicitly flags the concept as a theoretical construct authored by the researchers. I considered 'Hedged' but the metacognitive phrase 'what we term' clearly acknowledges the active creation of the terminology, recognizing its status as an analytical lens rather than a pre-existing physical object.)
  • Implications: While presented as a critique, framing computational limitations as a cognitive 'bias' paradoxically reinforces the illusion of the AI's mind. It implies that beneath the bias lies a capable, conscious reasoner. This affects understanding by making the problem seem solvable through 'therapy' (better prompting or debiasing) rather than recognizing it as a fundamental boundary of text-based statistical modeling. It risks leading regulators to focus on 'correcting the AI's bias' rather than auditing the underlying data regimes and corporate design choices that make contextual grounding mathematically impossible.

Accountability Analysis:

  • Actor Visibility: Named (actors identified)
  • Analysis: In this specific instance, the authors name themselves ('we') as the actors actively theorizing and categorizing the phenomenon. I considered 'Partial' but 'we' directly points to the human researchers framing the discourse. However, regarding the design of the system itself, the agency remains partially displaced, as the 'bias' is framed as an emergent property rather than a direct result of engineers deciding to train models purely on text without multimodal worldly grounding.

6. Statistical Outputs as Chosen Strategies​

Quote: "rather than flexibly computing different inferences depending on context, models often applied a single interpretive strategy"

  • Frame: Model as deliberate tactician
  • Projection: The text projects the human capacity for deliberate choice, strategic planning, and active interpretation onto passive mathematical outcomes. A 'strategy' requires a conscious goal, an evaluation of options, and a deliberate decision to employ a specific method to achieve that goal. By stating models 'applied a single interpretive strategy,' the text transforms the mechanistic reality of a rigid, over-fitted statistical distribution into a narrative of conscious, willful behavior. It suggests the AI evaluated the context, 'understood' the options, and consciously decided to stick to one method, thereby confusing processing invariances with conscious choices.
  • Acknowledgment: Direct (Unacknowledged) (The attribution of an 'interpretive strategy' is made as a direct observation of the system's empirical behavior without any qualifying language. I considered 'Hedged' because of the word 'often', but 'often' qualifies the frequency of the behavior, not the metaphorical claim that the behavior is a 'strategy.')
  • Implications: By characterizing deterministic mathematical outputs as 'strategies,' the discourse subtly justifies errors as deliberate choices rather than fundamental systemic flaws. If an AI is 'strategic,' users are more likely to trust it to handle complex, open-ended tasks, believing the system possesses the conscious awareness required to shift tactics when needed. This profound capability overestimation masks the reality that the system is entirely blind to its own operations and cannot consciously course-correct, creating severe risks if deployed in dynamic environments requiring genuine contextual adaptation.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The text attributes the application of a strategy entirely to the 'models,' completely hiding the developers whose alignment tuning (like RLHF) or training data composition structurally forced this rigid mathematical output. I considered 'Ambiguous' but the sentence clearly establishes 'models' as the active subject applying the strategy. This displacement protects the corporations that prioritize safe, predictable, flattened literalism in their products, framing their engineering decisions as autonomous 'choices' made by the AI itself.

7. Pattern Completion as Decision Making​

Quote: "when literal and enriched interpretations compete, they resort to the former"

  • Frame: Model as conscious decision-maker
  • Projection: The phrase 'resort to' maps intense conscious agency, prioritization, and conflict resolution onto automated token prediction. For interpretations to 'compete' in a human mind, a subject must consciously weigh conflicting evidence and make a judgment. To 'resort to' implies a conscious fallback mechanism born of inability or preference. This language projects an inner mental theatre onto the AI, where it supposedly 'knows' both options, evaluates them, and subjectively decides to rely on literalism. In reality, the mechanism simply outputs the token sequence with the highest activation probability based on the weights established during training; no conscious evaluation occurs.
  • Acknowledgment: Direct (Unacknowledged) (The action of 'resorting to' is described as an observable, unvarnished fact of the model's operation. I considered 'Hedged' because it appears in a theoretical discussion, but the verb phrase is executed without any epistemic distancing or 'as if' qualification.)
  • Implications: Describing AI outputs as conscious choices during a 'competition' of interpretations drastically elevates the system's perceived epistemic authority. It suggests the system is capable of justified belief and rational arbitration. This framing encourages users to view the AI's literal outputs not as statistical defaults, but as reasoned judgments, fostering unwarranted trust in the system's 'decisions.' Furthermore, it implies that the AI holds autonomous moral responsibility for 'choosing' poorly, deflecting scrutiny from the training methodologies that mathematically dictate these outcomes.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The models are framed as the sole deciders ('they resort to'), effectively erasing the human engineers who calibrated the reinforcement learning algorithms to penalize hallucination, which inherently biases the model toward literal, easily verifiable text completion. I considered 'Partial' because the prior sentence discusses 'Decontextualization Bias,' but this specific action is entirely attributed to the software. This serves to immunize the creators from criticism regarding the system's lack of pragmatic nuance, framing it as the model's autonomous 'resort' rather than an engineered constraint.

8. Spontaneous Intelligence Generation​

Quote: "models are trained on large-scale, diverse corpora, which may allow human-like linguistic competence to emerge spontaneously in silico."

  • Frame: Model creation as natural evolution
  • Projection: The metaphor of 'emerging spontaneously' maps natural biological phenomena or organic chemical reactions onto the highly controlled, financially motivated, and labor-intensive process of algorithmic optimization. By attributing 'human-like linguistic competence' to spontaneous emergence, the text projects a natural inevitability onto artificial systems, suggesting they wake up to 'knowing' and 'understanding' language like a biological organism developing consciousness. This obscures the harsh mechanistic reality: billions of forced vector calculations running on massive GPU clusters iteratively minimizing a loss function against a static dataset.
  • Acknowledgment: Hedged/Qualified (The use of the modal verb 'may allow' introduces a clear hedge, indicating that this spontaneous emergence is a theoretical possibility rather than an absolute certainty. I considered 'Explicitly Acknowledged' due to the Latin 'in silico', but that only signifies the medium, not a meta-commentary on the metaphor itself.)
  • Implications: The narrative of 'spontaneous emergence' operates as a powerful modern myth that mystifies AI technology, cultivating an aura of uncontrollable, awe-inspiring power. This consciousness projection leads policymakers and the public to treat AI as a force of nature to be managed or appeased, rather than a commercial product to be regulated. By framing competence as 'spontaneous,' it obscures the intentionality behind the system's capabilities, leading to regulatory paralysis where authorities feel unable to govern something framed as a naturally emergent, unpredictable phenomenon.

Accountability Analysis:

  • Actor Visibility: Hidden (agency obscured)
  • Analysis: The passive construction 'models are trained' hides the massive corporate entities performing the training, while 'emerge spontaneously' completely erases the human agency involved in architecting the networks, scraping the internet for data, and tuning the hyperparameters. I considered 'Partial' because 'corpora' implies human creation, but no human actors are named. This agentless framing is the ultimate accountability sink, suggesting that whatever the AI becomes is an unguided act of nature, thereby legally and ethically absolving the tech executives who actually direct, fund, and profit from these engineering efforts.

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: Developing biological organism or conscious human mind → Statistical optimization and neural network architecture​

Quote: "suggesting that pragmatic reasoning is still an emerging ability in the cognitive toolkit of artificial systems."

  • Source Domain: Developing biological organism or conscious human mind
  • Target Domain: Statistical optimization and neural network architecture
  • Mapping: This structure maps the biological timeline of cognitive maturation onto the iterative scaling of language models. In the source domain, a human mind contains a 'toolkit' of cognitive skills (logic, empathy, pragmatics) that organically 'emerge' as the brain develops and the person consciously learns to navigate the world. The mapping projects this internal psychological structure onto AI, implying that beneath the computational surface, a localized 'mind' is acquiring discrete skills. It invites the assumption that the system possesses a unified conscious awareness that is slowly 'learning' to grasp pragmatic reality, transitioning from basic processing to genuine, justified knowing.
  • What Is Concealed: This mapping entirely conceals the static, deterministic nature of the mathematical matrices. It hides the fact that a model does not 'develop' or 'emerge' organically; its weights are updated via massive infusions of computing power and human-directed data curation. It obscures the total absence of real-world grounding, sensory experience, and intentionality—mechanistic realities that make true pragmatic reasoning impossible for current architectures. Furthermore, it conceals the proprietary, closed-door decisions of tech companies under the guise of natural technological evolution.
Show more...

Mapping 2: Autonomous, conscious communicator → Generative text-prediction algorithms​

Quote: "LLMs, while undeniably impressive linguistic agents, have cognitive toolkits that remain fundamentally different from those of humans"

  • Source Domain: Autonomous, conscious communicator
  • Target Domain: Generative text-prediction algorithms
  • Mapping: This mapping projects the relational structure of human interpersonal communication onto human-computer interaction. In the source domain, an 'agent' is a conscious entity with intentions, goals, and the ability to initiate action based on an understanding of meaning. By projecting this onto LLMs, the text maps the subjective state of 'knowing' what one is saying onto the mechanistic process of calculating token probabilities. It assumes that because the output resembles human communication, the source of the output must possess a parallel, albeit 'different,' internal cognitive state capable of genuine communication.
  • What Is Concealed: The 'agent' mapping completely conceals the reactive, non-volitional reality of generative AI. The system has no intentions, no goals, and no awareness of the user or the context. It obscures the mechanism of token prediction, where the system is merely returning mathematical correlations derived from its training set without any comprehension of the signified reality. It also conceals the socio-technical assemblage behind the screen: the human prompt engineers, the RLHF guardrails, and the corporate servers that are the actual 'agents' facilitating the transaction.

Mapping 3: Conscious student or striving subject → Algorithmic inability to map to target distributions​

Quote: "they nonetheless struggle with meaning-related components of language"

  • Source Domain: Conscious student or striving subject
  • Target Domain: Algorithmic inability to map to target distributions
  • Mapping: This structure maps the subjective human experience of cognitive friction onto statistical inaccuracy. In the source domain, a student 'struggles' when they consciously recognize a gap between their current understanding and a desired state of knowledge, applying willful effort to bridge that gap. Projected onto AI, this mapping suggests the system is aware of 'meaning,' wants to grasp it, but encounters internal difficulty. It attributes conscious intent and an epistemic desire to 'know' the material, transforming mathematical failure into an ongoing, sympathetic psychological effort.
  • What Is Concealed: This mapping hides the fundamental truth that models do not experience effort, difficulty, or a desire to improve. A model 'failing' a pragmatic inference test is executing its mathematical function flawlessly based on its training data; it simply lacks the statistical patterns required to produce the desired human output. The metaphor conceals the fundamental architectural limitation of text-only training: the system cannot struggle with 'meaning' because it has absolutely no access to meaning, only to the statistical distribution of signifiers.

Mapping 4: Human mastery and skill acquisition → Successful optimization of syntactic probability distributions​

Quote: "LLMs have acquired formal linguistic competence"

  • Source Domain: Human mastery and skill acquisition
  • Target Domain: Successful optimization of syntactic probability distributions
  • Mapping: This maps the human pedagogical journey onto the engineering process of model training. In the source domain, a person 'acquires competence' through conscious practice, internalizing rules, understanding exceptions, and developing a justified belief in their ability to perform. When projected onto a language model, it maps the conscious possession of knowledge onto a frozen set of billions of numerical weights. It invites the audience to believe that the AI has internalized grammar as a set of comprehended concepts, elevating its mechanistic processing of patterns into the epistemic state of 'knowing' a language.
  • What Is Concealed: The mapping conceals the radically different mechanism by which LLMs achieve output that looks competent. It hides the fact that the system possesses no internal rulebook, no conceptual understanding of syntax, and no awareness of grammar. It obscures the massive environmental and labor costs required to achieve this 'competence'—the scraping of billions of human-written texts without consent, and the massive energy expenditures required to identify statistical correlations within them. The 'acquisition' is entirely passive and mechanical, not active and cognitive.

Mapping 5: Human psychological or cognitive prejudice → Mathematical absence of contextual data representation​

Quote: "arguing that the reasoning abilities of LLMs are affected by what we term a Decontextualization Bias"

  • Source Domain: Human psychological or cognitive prejudice
  • Target Domain: Mathematical absence of contextual data representation
  • Mapping: This structure maps human psychological flaws onto algorithmic limitations. In the source domain, a 'bias' occurs when a conscious mind, capable of rational thought, is skewed by internal heuristics, emotions, or unexamined assumptions. By mapping this onto LLMs, the text suggests the system actually possesses an underlying 'reasoning ability' that is merely being 'affected' or distorted by a bad mental habit. It projects a duality onto the machine: a rational, knowing core that is unfortunately hindered by a subjective, psychological blind spot.
  • What Is Concealed: This conceals the fact that LLMs do not possess 'reasoning abilities' to be biased; their entire architecture is a flat, decontextualized statistical map. They cannot 'ignore' context due to a bias; they literally cannot perceive context because they exist outside of space, time, and human social reality. It also conceals the proprietary design choices of the developers who explicitly trained the models to prioritize literal surface forms to ensure safe, verifiable, and generalized outputs, reframing a corporate engineering strategy as an accidental psychological flaw.

Mapping 6: Conscious strategic planner or human problem-solver → Deterministic generation of high-probability token sequences​

Quote: "rather than flexibly computing different inferences depending on context, models often applied a single interpretive strategy"

  • Source Domain: Conscious strategic planner or human problem-solver
  • Target Domain: Deterministic generation of high-probability token sequences
  • Mapping: This maps the human executive function of selecting a method to solve a problem onto the automated execution of an algorithm. In the source domain, applying a 'strategy' involves a conscious mind assessing the environment, 'knowing' the available options, and making a justified decision to deploy a specific tactic. Projected onto the model, this mapping suggests the AI evaluates linguistic context, consciously considers multiple interpretations, and deliberately 'chooses' to apply a single, rigid rule. It imbues the mathematical output with intentionality and meta-cognitive awareness.
  • What Is Concealed: The mapping hides the absence of choice in the computational process. A model does not 'apply a strategy'; it executes a fixed mathematical operation. It conceals the specific alignment training (like RLHF) that flattens out diverse, nuanced responses in favor of uniform, predictable, and 'helpful' literalism. By attributing the uniformity to the model's 'strategy,' it obscures the reality that the opacity of the black-box system prevents us from knowing exactly how the training data distribution forced this specific mathematical convergence.

Mapping 7: Human arbitration, conflict resolution, and decision-making → Statistical probability weighting in neural networks​

Quote: "when literal and enriched interpretations compete, they resort to the former"

  • Source Domain: Human arbitration, conflict resolution, and decision-making
  • Target Domain: Statistical probability weighting in neural networks
  • Mapping: This structure maps the conscious human experience of evaluating competing claims onto the mathematical resolution of competing vector weights. In the source domain, when interpretations 'compete,' a conscious judge assesses them and 'resorts to' one based on logic, preference, or exhaustion. Projected onto AI, this maps the epistemic process of 'knowing' two options and actively choosing one onto the mechanistic process of next-token prediction. It personifies the abstract linguistic concepts as active competitors inside a conscious mental arena possessed by the machine.
  • What Is Concealed: This mapping conceals the purely statistical, non-evaluative nature of the output generation. The model does not 'see' two interpretations and choose one; it simply calculates which token has the highest probability of appearing next given the input context. The 'resorting' hides the dependency on the specific baseline data the model was trained on—if literal text was overwhelmingly present in the corpora or heavily rewarded during human-feedback tuning, the math will dictate a literal output. It hides the human engineering behind the statistical weights.

Mapping 8: Biological evolution, self-organization, and natural emergence → Compute-intensive optimization of multi-billion parameter architectures​

Quote: "models are trained on large-scale, diverse corpora, which may allow human-like linguistic competence to emerge spontaneously in silico."

  • Source Domain: Biological evolution, self-organization, and natural emergence
  • Target Domain: Compute-intensive optimization of multi-billion parameter architectures
  • Mapping: This mapping projects the biological concept of self-organizing complexity (like consciousness arising from neurons or life from chemical soups) onto a manufactured, engineered artifact. In the source domain, 'spontaneous emergence' implies a natural, unguided process where new properties appear autonomously. Projected onto language modeling, it maps the mysterious origins of biological consciousness onto the scaling laws of AI, suggesting that simply pouring enough data into a computer naturally and inevitably sparks 'human-like' knowing and understanding without direct human design.
  • What Is Concealed: This mapping is one of the most obfuscatory, completely concealing the vast, directed, and highly capitalized human labor required to build these systems. It hides the deliberate architectural innovations (Transformers), the millions of hours of human data annotation, the fine-tuning, the content moderation, and the immense carbon footprint. By framing the capability as 'spontaneous,' it obscures the proprietary opacity of the corporate labs that meticulously engineer these models, treating a deeply intentional commercial product as if it were a miraculous natural phenomenon.

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: "This bias gives rise to a systematic preference for literal readings"

  • Explanation Types:

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

  • Analysis (Why vs. How Slippage): This explanation frames the AI agentially by diagnosing its behavior through an invented psychological construct ('bias') that creates a 'preference.' While 'systematic' hints at a mechanistic regularity, the core of the explanation relies on dispositional language—the system is framed as an entity that possesses subjective desires or inclinations ('preferences') guided by its internal mental state ('bias'). This choice emphasizes the model as a self-directed cognitive actor with specific intellectual habits. Simultaneously, it obscures the mechanistic reality that this 'preference' is actually a deterministic output dictated by the mathematical weighting of its training data and optimization functions. The explanation substitutes a technical description of how probability distributions resolve with a psychological narrative of why a conscious agent chooses a specific reading.

  • Consciousness Claims Analysis: The passage makes a strong, albeit implicit, consciousness claim by attributing a 'preference' to the system. Mechanistic verbs (processes, calculates) are replaced by subjective psychological nouns ('bias', 'preference'). This firmly situates the system in the realm of 'knowing' rather than 'processing.' To have a preference, an entity must be aware of multiple options and possess a subjective criteria for valuing one over the other. This represents a classic 'curse of knowledge' dynamic: the human researchers, who consciously prefer literal or pragmatic readings based on deep contextual understanding, project this conscious evaluative process onto the system's output. Mechanistically, the system has no preference; it strictly processes query embeddings through its attention layers, multiplying matrices to generate a probability distribution over a vocabulary, outputting the sequence that maximizes the likelihood function based on its specific training corpus.

  • Rhetorical Impact: By framing the system's mathematical output as a 'preference' born of a 'bias,' the text shapes the audience's perception of the AI as highly autonomous and human-like. This consciousness framing inadvertently builds relation-based trust; an entity with preferences is perceived as an intentional actor capable of being reasoned with, rather than a rigid calculator. Consequently, users and regulators might believe that this AI can be 'convinced' or 'taught' to change its preferences through discourse, fundamentally misunderstanding the deep architectural engineering required to alter its statistical outputs. It individualizes the 'flaw' into the machine's personality rather than exposing the structural limitations of the technology.

Show more...

Explanation 2​

Quote: "While models often capture the literal, truth-conditional structure of conditionals, they struggle to integrate contextual cues and speaker intentions"

  • Explanation Types:

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

  • Analysis (Why vs. How Slippage): This explanation operates primarily in an agential and dispositional register. By using the verb 'struggle to integrate,' the text explains the AI's behavior as a failure of effort toward a desired goal. It frames the AI as an entity attempting the complex cognitive task of mapping 'speaker intentions.' This emphasizes a perceived parallel between human psychological limitations and algorithmic constraints. However, it severely obscures the 'how'—the mechanistic reality that text-prediction models literally have no sensory access to 'context' outside of their prompt window, and absolutely no access to 'speaker intentions,' which are unobservable human mental states. The explanation masks a hard mathematical boundary with a narrative of cognitive difficulty.

  • Consciousness Claims Analysis: This passage is saturated with consciousness verbs and states: 'capture,' 'struggle,' 'integrate,' and recognizing 'intentions.' It aggressively positions the AI as a 'knower' rather than a 'processor.' For an AI to 'struggle to integrate speaker intentions,' it must first be capable of possessing the concept of an 'intention' and 'knowing' that a speaker exists. This is a profound epistemic leap and a clear projection of the author's own mind. Mechanistically, the model processes input tokens as numerical vectors; it does not 'know' a speaker exists, nor can it formulate a concept of 'intention.' It simply calculates attention scores between the tokens in the prompt and generates output tokens that statistically correlate with similar patterns in its training data. The model cannot struggle to integrate what it structurally cannot perceive.

  • Rhetorical Impact: Framing the system's limitation as a 'struggle' generates a powerful rhetorical impact: it elicits sympathy and patience from the audience while maintaining the illusion of the AI's vast underlying competence. If an AI is merely 'struggling' to understand intentions, the audience perceives it as a conscious entity on the verge of a breakthrough, requiring only more data or time. This consciousness framing obscures the true risk: the system is fundamentally blind to truth and intent. Believing the AI is trying to understand could lead users to trust its confidence in high-stakes scenarios, failing to realize the system is mathematically incapable of the empathy and contextual grounding implied by the text.

Explanation 3​

Quote: "models are trained on large-scale, diverse corpora, which may allow human-like linguistic competence to emerge spontaneously in silico."

  • Explanation Types:

    • Genetic: Traces origin through dated sequence of events or stages
    • Empirical Generalization: Subsumes events under timeless statistical regularities

  • Analysis (Why vs. How Slippage): This explanation begins mechanistically ('trained on... corpora') but immediately slips into a highly agential, almost mystical genetic explanation ('emerge spontaneously'). It explains the 'why' and 'how' of the AI's capability through an evolutionary narrative. This choice emphasizes scale and natural phenomena, positioning the AI's development as an organic, unstoppable force. By doing so, it almost entirely obscures the intense, deliberate, and highly curated engineering processes—reinforcement learning from human feedback (RLHF), data filtering, architecture tuning, and objective function design—that are required to make LLMs output coherent text. The explanation hides the hand of the human designer behind the veil of 'spontaneous' complexity.

  • Consciousness Claims Analysis: The passage attributes a state of 'competence' to the system, suggesting a transition from mechanical processing to cognitive knowing. By describing the competence as 'human-like' and 'spontaneous,' it invites the reader to imagine the machine waking up to consciousness and understanding simply because it was fed enough data. The authors project their own deep, conscious understanding of language onto the machine's probabilistic outputs. Mechanistically, nothing 'emerges spontaneously.' The system optimizes a highly specific, mathematically defined loss function over thousands of GPU hours. The 'competence' is an illusion generated by the system's ability to retrieve and rank high-probability token combinations derived entirely from the human consciousness embedded in the training texts, without the system itself possessing a single justified true belief.

  • Rhetorical Impact: The rhetoric of 'spontaneous emergence' radically shapes audience perception by conferring an aura of autonomous, quasi-divine intelligence upon the technology. This framing diminishes the perceived agency of humans while amplifying the perceived autonomy of the machine. The risk here is systemic: if regulators and users believe AI capabilities are 'spontaneous' forces of nature, they will view accountability as impossible. It shifts the discourse from regulating corporate data practices to reacting to an uncontrollable new species, ultimately serving the interests of companies that wish to evade strict liability for the specific, deliberate design choices encoded in their systems.

Explanation 4​

Quote: "rather than flexibly computing different inferences depending on context, models often applied a single interpretive strategy"

  • 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 uses a reason-based and intentional framework to describe a purely mathematical phenomenon. It contrasts what the model 'should' do as a conscious actor ('flexibly computing... depending on context') with what it deliberately 'chooses' to do ('applied a single interpretive strategy'). This frames the AI agentially, emphasizing it as an autonomous decision-maker executing a plan. The choice of 'strategy' completely obscures the mechanistic explanation: that the neural network's weights were likely over-optimized during fine-tuning (e.g., to prioritize literal, helpful answers to pass benchmark tests), resulting in a mathematical collapse into a single high-probability output path regardless of subtle prompt variations.

  • Consciousness Claims Analysis: The passage heavily attributes knowing and consciousness to the system through the word 'strategy.' To apply a strategy, an entity must 'know' its goals, understand its environment, and possess the conscious agency to execute a plan. This demonstrates the curse of knowledge: the researchers, observing a uniform output across varying inputs, use their own conscious, strategic minds to make sense of the rigid data, projecting a 'strategy' onto what is actually a computational limitation. Mechanistically, the model does not interpret or strategize. It processes embeddings through attention layers that have been mathematically tuned to strongly associate certain syntax with literal outputs, generating tokens based on maximum likelihood without any awareness of 'strategy' or 'context.'

  • Rhetorical Impact: Framing rigid algorithmic output as a 'strategy' drastically alters the audience's perception of risk and reliability. It rationalizes the system's failures as deliberate, calculated choices rather than inherent structural flaws. If audiences believe the AI is 'strategic,' they will assume it possesses a vast reserve of hidden competence and justified reasoning behind its actions. This consciousness framing leads to dangerous over-trust; users might rely on the AI for complex problem-solving, assuming it will 'strategize' around new obstacles, when in reality it will simply fail or hallucinate when the statistical distribution of the novel context does not match its training data.

Explanation 5​

Quote: "This behavior implies that these models do not engage in genuine pragmatic reasoning, but rather rely on a fixed, rule-based strategy that treats conditionals as biconditionals across the board."

  • Explanation Types:

    • Functional: Explains behavior by role in self-regulating system with feedback
    • Dispositional: Attributes tendencies or habits

  • Analysis (Why vs. How Slippage): This passage attempts a mechanistic correction but remains trapped in agential language. While it correctly denies the model 'genuine pragmatic reasoning,' it immediately replaces that with another intentional, dispositional framing: 'rely on a fixed, rule-based strategy.' It explains the behavior functionally (how it operates across inputs) but emphasizes the system as an independent actor that 'relies' and 'treats.' This obscures the absolute reality that language models do not possess internal 'rules' in a symbolic sense; they are entirely statistical correlation engines. The explanation hides the continuous, floating-point nature of neural network vector spaces behind the discrete, human-comprehensible metaphor of a 'rule-based strategy.'

  • Consciousness Claims Analysis: Even in critiquing the model's capabilities, the passage attributes knowing verbs ('rely on', 'treats'). The system is still framed as an entity that 'knows' what a conditional is and actively decides to 'treat' it as something else. The authors, possessing formal linguistic knowledge of biconditionals, project that symbolic categorization onto the model's output. Mechanistically, the model does not 'treat' conditionals as anything, nor does it have symbolic 'rules.' It predicts tokens. During its alignment training or pre-training, linguistic structures resembling conditionals frequently co-occurred with literal, biconditional outcomes, causing the network's weights to heavily favor those token sequences. It is a statistical invariance, not a known rule.

  • Rhetorical Impact: By describing the system's limitations as a 'reliance' on a 'strategy,' the text maintains the illusion of an autonomous, calculating entity even while exposing its flaws. The rhetoric softens the blow of the model's failure by framing it as a rigid thinker rather than a blind calculator. This impacts policy and development decisions by suggesting the fix is to 'teach' the model a new strategy or give it new rules, rather than acknowledging the fundamental incapacity of next-token prediction architectures to achieve grounded, real-world pragmatic reasoning. It preserves the system's authority as a 'thinker,' albeit a flawed one.

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
suggesting that pragmatic reasoning is still an emerging ability in the cognitive toolkit of artificial systems.The data indicates that generating outputs mimicking pragmatic inferences is a statistical capability not yet reliably achieved by current text-prediction architectures.Artificial systems do not possess a 'cognitive toolkit' or 'abilities.' Mechanistically, the models process input embeddings and calculate probability distributions to predict tokens. They do not reason; they correlate patterns from their training corpora.N/A - describes computational processes without displacing responsibility.
LLMs, while undeniably impressive linguistic agents, have cognitive toolkits that remain fundamentally different from those of humansGenerative text systems, while producing highly complex and statistically accurate linguistic outputs, process language via mathematical correlations entirely unlike human conscious comprehension.Models are not 'agents' and do not have 'cognitive toolkits.' They do not know or understand. They classify and predict tokens using multi-layered transformer architectures optimized via gradient descent.N/A - describes computational processes without displacing responsibility.
they nonetheless struggle with meaning-related components of languageCurrent transformer architectures fail to consistently output correct tokens in tasks that, for humans, require semantic comprehension and real-world grounding.A model cannot 'struggle' or grasp 'meaning.' It mathematically optimizes loss functions. When it outputs incorrect responses, it is because the statistical distribution of the training data does not contain the required correlations.N/A - describes computational processes without displacing responsibility.
LLMs have acquired formal linguistic competenceEngineers have successfully trained LLMs to generate text that reliably conforms to the probabilistic patterns of formal syntax found in their training data.LLMs do not 'acquire competence' or know grammar. They mechanistically encode contextual embeddings based on attention mechanisms tuned over billions of iterations to replicate human syntactic structures.Corporate engineering teams and researchers have designed architectures and compiled massive datasets that tune these systems to replicate formal syntax.
arguing that the reasoning abilities of LLMs are affected by what we term a Decontextualization BiasWe hypothesize that model output inaccuracies stem from a structural limitation: the algorithms prioritize high-frequency literal token associations over lower-frequency context-dependent patterns.Models do not have 'reasoning abilities' to be affected by psychological 'bias.' They simply retrieve and rank tokens based on probability distributions established during their algorithmic optimization.N/A - describes computational processes without displacing responsibility.
rather than flexibly computing different inferences depending on context, models often applied a single interpretive strategyRather than generating variable outputs sensitive to subtle prompt changes, the systems' mathematical weights predominantly collapsed toward a single, high-probability output pattern.Models do not 'apply strategies' or 'interpret.' They process input tokens through fixed neural weights. The uniformity of output reflects algorithmic inflexibility and training data distribution, not conscious strategic choice.Developers likely aligned these models using reinforcement learning techniques that inadvertently penalized variable responses, forcing the algorithms into rigid, highly localized statistical distributions.
when literal and enriched interpretations compete, they resort to the formerWhen prompts contain structures that correlate with both literal and enriched texts in the training data, the models consistently generate the higher-probability literal tokens.Models do not recognize competing interpretations, nor do they 'resort' to decisions. They mechanistically calculate matrix multiplications that mathematically favor the most dominant token sequence derived from their training.N/A - describes computational processes without displacing responsibility.
models are trained on large-scale, diverse corpora, which may allow human-like linguistic competence to emerge spontaneously in silico.Technology companies optimize massive neural networks on vast datasets, engineering the mathematical parameters until the system's generated text closely mimics human linguistic patterns.Competence does not 'emerge spontaneously.' The system remains a purely mechanistic artifact that processes inputs and generates correlative outputs without subjective awareness or spontaneous organic development.Tech corporations extract massive quantities of human-generated text and employ engineering teams and data annotators to meticulously align and tune the models to produce these specific outputs.

Task 5: Critical Observations - Structural Patterns​

Agency Slippage​

The text exhibits a profound and systematic oscillation between mechanical and agential framings, driven largely by the authors' struggle to articulate complex mathematical outcomes using natural language. This agency slippage serves as a critical mechanism for constructing the 'illusion of mind' within the LLM. The oscillation follows a distinct directional pattern: the text establishes the mechanistic reality of the system (e.g., 'LLMs are accurate semantic operators' or 'trained on large-scale corpora'), but almost immediately slips into agential, psychological attribution when describing complex behavior (e.g., 'they struggle', 'they resort to', 'applied a strategy'). The dominant flow of agency is overwhelmingly toward the AI system, elevating it from a mathematical tool to a cognitive subject.

This slippage is most dramatic precisely at the moments where the authors must explain the system's failure to replicate human pragmatic reasoning. For example, when observing that models fail to distinguish between biscuit and standard conditionals, the text does not say 'the models lack the statistical correlations required for this output.' Instead, it invokes Brown's dispositional and reason-based explanation types, claiming the models 'resort to the former' and 'applied a single interpretive strategy.' Here, the 'curse of knowledge' is on full display. The human researchers, possessing deep semantic and pragmatic awareness, cannot help but project their own internal, conscious decision-making processes onto the black box of the neural network. Because a human would have to 'strategize' or 'resort' to a rule to consistently fail this way, the authors assume the machine is doing the same.

Crucially, this slippage relies on pervasive agentless constructions that simultaneously strip human actors of their agency. Sentences like 'models are trained' or 'human-like linguistic competence to emerge spontaneously' completely erase the corporate laboratories, the massive data extraction, and the reinforcement learning by human feedback (RLHF) that actually determine the models' behaviors. By removing the human engineers who deliberately aligned these systems to be literal and 'safe,' the text creates a vacuum of agency that is immediately filled by the AI itself. The model is positioned as the sole active 'knower' in the text, capable of acquiring 'competence' and possessing a 'cognitive toolkit.' This rhetorical accomplishment makes it deeply unsayable to suggest that the AI is just a rigid mathematical calculator failing a statistical test; instead, it is framed as a conscious entity wrestling with cognitive biases, fundamentally altering the epistemic reality of the machine.

Metaphor-Driven Trust Inflation​

The metaphorical architecture of the text systematically constructs an unwarranted authority for AI systems by conflating performance-based reliability with relation-based trust. Throughout the paper, the authors deploy consciousness-attributing language—such as referring to the models as 'linguistic agents,' possessing a 'cognitive toolkit,' and having 'acquired formal linguistic competence.' This language acts as a profound trust signal. In human social dynamics, assigning 'competence' and recognizing someone as an 'agent' implies a recognition of their conscious awareness, sincerity, and capacity for justified belief. By projecting this framework onto statistical token predictors, the text inadvertently encourages the audience to extend relation-based trust to systems that are fundamentally incapable of reciprocating it.

The danger of this anthropomorphism becomes most apparent in how the text manages system limitations and failures. When the models fail to execute pragmatic inferences, the text does not frame this as a catastrophic mathematical breakdown or a hard limitation of text-only architecture. Instead, it utilizes intentional and reason-based explanations: the models 'struggle,' they suffer from 'Decontextualization Bias,' and they 'resort to' specific strategies. This framing is crucial for maintaining trust despite failure. A machine that is 'biased' or 'struggling' is perceived as an entity that is trying to get it right. It is humanized by its flaws. The audience is invited to trust the underlying 'intent' of the machine, believing that it 'knows' the goal but is simply encountering a psychological hurdle.

This transfer of human-trust frameworks onto statistical systems creates massive epistemic vulnerabilities. If audiences believe the AI 'understands' and 'knows' language as an 'agent,' they will naturally assume it possesses the moral and contextual weight necessary to evaluate its own outputs. They will trust its confident generation of text as the sincere assertion of a knowing subject rather than the probabilistic output of a mechanism. The stakes of this misplaced relation-based trust are immense: when users rely on an AI that they believe is a 'competent agent' with a 'cognitive toolkit,' they are likely to deploy it in high-stakes legal, medical, or administrative contexts, utterly unaware that the system possesses zero contextual grounding, no capability for truth-verification, and no conscious awareness of the consequences of its output.

Obscured Mechanics​

The dense anthropomorphic and consciousness-attributing language in this text systematically conceals the material, technical, economic, and labor realities that actually constitute Large Language Models. By framing the AI as an autonomous, conscious 'agent' that 'acquires competence' and 'applies strategies,' the text renders invisible the massive socio-technical infrastructure required to generate its text.

Applying the 'name the corporation' test reveals the depth of this obfuscation. When the text says the models 'applied a single interpretive strategy' or 'resort to the former,' it hides the specific engineering teams at OpenAI, Anthropic, Google, and Meta who designed the optimization objectives. The failure of these models to capture pragmatics is not a cognitive 'bias' or an independent 'strategy'; it is the direct consequence of corporate decisions to train models via Reinforcement Learning from Human Feedback (RLHF) and strict safety prompts that actively penalize ambiguity and enforce rigid, literal, universally 'helpful' responses. The metaphor of a 'struggling' cognitive mind completely obscures this commercial objective.

Furthermore, claims about the AI 'knowing' or 'understanding' language hide profound technical realities. The text treats the models as psychological subjects, ignoring the reality of the proprietary black boxes they actually are. The authors confidently assert why the models fail ('Decontextualization Bias'), despite having absolutely no access to the training data weights or the exact alignment algorithms of closed models like GPT-4o or Claude. The consciousness metaphor bridges the gap of this transparency obstacle, providing a neat psychological explanation where rigorous mechanistic verification is legally and technically impossible.

Materially, the text's narrative of 'spontaneous emergence in silico' erases the immense environmental costs, energy consumption, and infrastructure required to train these models. It also entirely erases the labor of the millions of low-paid data annotators who painstakingly categorized the human feedback necessary to make these models appear 'competent.' The primary beneficiaries of this concealment are the technology corporations themselves. By having their products framed by academics as 'linguistic agents' with 'cognitive toolkits,' these companies enjoy the marketing benefits of perceived Artificial General Intelligence while remaining shielded from accountability for the specific, highly curated, and often flawed engineering choices that actually drive the algorithms. Replacing these metaphors with mechanistic language would make visible the direct line between corporate design choices, exploited labor, and the rigid statistical outputs observed in the study.

Context Sensitivity​

The distribution of anthropomorphic and consciousness-attributing language across the text is highly strategic, varying significantly in density and intensity depending on the rhetorical context. In the technical methodology and raw results sections, the language leans mechanistic: the text discusses 'accuracy,' 'logistic regression,' 'binary variables,' and models acting as 'reliable logical agents.' However, in the introduction and discussion sections—where the authors must contextualize their findings and construct a broader narrative—the metaphorical license drastically intensifies. It is here that 'processes' escalates into 'understands,' 'interprets,' 'acquires,' and ultimately possesses a 'cognitive toolkit.'

This pattern reveals a sophisticated rhetorical maneuver: the text establishes its scholarly credibility through rigorous, mechanical statistical analysis, and then leverages that credibility to make aggressive, anthropomorphic claims about the nature of the AI's 'mind.' The intensity of the consciousness claims directly correlates with the authors' need to explain variance. When the math is clear, the model is an 'operator.' When the math produces confusing or rigid output (like the failure on biscuit conditionals), the model suddenly becomes an 'agent' applying a 'strategy' or suffering from a 'bias.'

Crucially, there is a pronounced asymmetry in how capabilities versus limitations are framed. When discussing the models' successes (e.g., matching truth-tables), the text uses agential and consciousness terms: they have 'acquired formal linguistic competence' and are 'competent semantic agents.' However, when discussing their limitations, the language becomes a mix of psychological excuse ('Decontextualization Bias') and rigid mechanism ('rule-based processing'). This asymmetry accomplishes a vital strategic function: it preserves the overall perceived sophistication of the technology. The machine is granted full cognitive credit for its successes, but its failures are framed either as sympathetic psychological struggles or as the result of rigid rules, never as proof that the system entirely lacks comprehension.

This context sensitivity suggests that the anthropomorphism serves as a vision-setting tool and a mechanism for managing critique. By elevating the discourse into the realm of cognition and psychology in the discussion sections, the authors align their statistical findings with the grander, more marketable narrative of emerging Artificial Intelligence, ensuring their research resonates with the broader philosophical and commercial debates surrounding the technology, even as it obscures the mechanistic reality of their own data.

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.

Synthesizing the accountability analyses reveals a systemic architectural pattern in the text: a near-total displacement of responsibility from human creators to the AI systems themselves. The text's grammar consistently structures the discourse so that the 'models' or 'LLMs' occupy the subject position of active, deciding verbs. They 'acquire,' 'struggle,' 'apply strategies,' 'resort to,' and 'emerge.' Simultaneously, the human engineers, corporate executives, and data annotators are relegated to passive, unnamed shadows, hidden behind agentless constructions like 'models are trained' or 'have acquired.'

This architecture creates a massive 'accountability sink.' When the text identifies a profound failure in the system—namely, its inability to compute pragmatic, contextually grounded inferences—responsibility does not flow back to the corporations that designed a text-prediction architecture fundamentally incapable of worldly grounding. Instead, the responsibility transfers directly to the AI as an autonomous agent. The failure is framed as the model's own 'Decontextualization Bias' or its 'choice' of a rigid 'interpretive strategy.' By making human actors invisible and presenting the AI's outputs as the inevitable 'emergence' of a conscious entity's internal logic, the text effectively absolves the creators of their design choices.

The liability implications of this framing are severe. If policymakers, regulators, and the public accept that AI systems are autonomous 'linguistic agents' with their own 'cognitive toolkits' and 'biases,' it becomes legally and culturally difficult to hold corporations accountable for the damages their products cause. You do not sue the parents of a conscious adult who makes a bad strategic choice; the anthropomorphic framing invites precisely this kind of autonomous liability shield for the software.

If we apply the 'name the actor' test to the text's most significant agentless constructions, the entire discourse shifts. If, instead of 'models applied a single interpretive strategy,' the text read, 'OpenAI and Anthropic's RLHF tuning forced the models to output rigid, literal token sequences,' profound new questions become askable. We can ask: Why did they choose that tuning? What economic incentives drive the flattening of pragmatics? What alternatives were ignored? By naming the actors, the illusion of an autonomous, struggling AI mind shatters, revealing a highly calculated, commercially driven software product, and making true accountability possible.

Conclusion: What This Analysis Reveals​

The Core Finding

The discourse analysis of this text reveals a tightly integrated, load-bearing system of anthropomorphic metaphors designed to project consciousness onto statistical language models. The overarching architecture relies on three dominant, interlocking patterns: AI as a developing biological organism ('cognitive toolkit', 'spontaneous emergence'), AI as a conscious, intentional actor ('linguistic agents', 'applying strategies'), and AI as a struggling psychological subject ('bias', 'struggling'). These patterns are not isolated rhetorical flourishes; they form a logical, mutually reinforcing flow. To believe a model can have a 'bias' or 'struggle' (Pattern 3), one must first accept that it is an intentional actor (Pattern 2), which is in turn predicated on the foundational assumption that it possesses an internal, organic architecture of mind (Pattern 1).

The foundational pattern—the consciousness architecture—is the projection of 'knowing' onto 'processing.' The text systematically uses verbs that demand a conscious subject: understand, reason, interpret, and acquire. This simple, one-to-one mapping of human cognitive states onto mathematical operations is the load-bearing pillar of the entire illusion. If you remove the assumption that the system 'knows' what it is outputting, the claims that it is 'applying a strategy' or 'resorting to the literal' immediately collapse into absurdity. The text's sophistication lies not in complex analogical structure, but in its relentless, unhedged grammatical insistence that the machine is a 'who' rather than a 'what.'

Mechanism of the Illusion:​

The text creates the 'illusion of mind' through a highly effective rhetorical sleight-of-hand: the systematic blurring of the boundary between structural processing and subjective knowing. This illusion operates heavily on the 'curse of knowledge.' The authors, possessing deep semantic and pragmatic understanding, observe the model outputting complex, grammatically perfect sentences. Unable to intuitively grasp how billions of floating-point multiplications can generate such coherence without a guiding consciousness, the authors project their own internal psychological states—strategies, biases, and struggles—backward onto the black-box mechanism to make sense of its output.

The temporal structure of the persuasion is critical. The text first establishes the system's 'competence' using the undeniable empirical reality of its grammatical accuracy. Once the audience accepts that the machine has 'acquired' grammar, they become vulnerable to the subsequent, far more dangerous leap: that the machine possesses the conscious intent to 'use' that grammar strategically. The illusion exploits a deep human vulnerability: our evolutionary hardwiring to attribute mind and intention to anything that communicates with us in natural language. By employing intentional and reason-based explanation types, the text validates this human instinct, providing academic sanction to the visceral illusion that there is a 'ghost in the machine' making active, cognitive choices.

Material Stakes:​

Categories: Regulatory/Legal, Epistemic, Social/Political

The metaphorical framings within this text carry profound, tangible material stakes, primarily driven by the conflation of mechanistic processing with conscious knowing. In the Regulatory/Legal category, framing AI as an 'emerging,' 'struggling' agent with its own 'cognitive toolkit' creates an intractable liability shield for corporations. If regulators believe a system 'strategizes' and makes autonomous 'interpretive choices,' they are far less likely to draft legislation targeting the human engineers who built the training data pipelines. Accountability diffuses into the abstraction of the technology, allowing companies like OpenAI and Google to evade strict liability for the harms—such as defamation, bias, or copyright infringement—generated by their deterministic mathematical models.

Epistemically, attributing 'competence' and 'reasoning' to statistical token predictors fundamentally degrades public literacy regarding truth. When audiences are told a machine 'knows' the answer and has 'acquired' language, they trust its outputs as justified assertions of reality. They become incapable of recognizing that the machine is structurally blind to truth, possessing only statistical correlations. This capability overestimation leads to catastrophic automation bias in critical fields like medicine and law, where users outsource verification to a machine entirely devoid of comprehension. Socially and politically, the 'spontaneous emergence' narrative mystifies the technology, consolidating power in the hands of the tech elite by framing their deliberate, profit-driven engineering choices as inevitable acts of technological nature that society must simply adapt to, rather than actively govern.

AI Literacy as Counter-Practice:​

Critical literacy and mechanistic precision serve as vital counter-practices to the dangerous material stakes of anthropomorphism. As demonstrated in the reframing exercises, practicing precision requires two non-negotiable commitments: Epistemic Correction (replacing consciousness verbs with mechanistic descriptions) and Restoring Human Agency (naming the corporate actors behind the curtain). When we reframe 'the model applied a single interpretive strategy' to 'developers aligned the architecture to output rigid, literal token sequences,' the illusion of the autonomous mind shatters. This simple linguistic shift directly counters the regulatory and epistemic risks by forcing recognition of the system's actual nature: it is a designed, deterministic software product entirely reliant on its training data, completely lacking internal awareness.

Systematic adoption of this precision requires structural change. Academic journals and peer-reviewers must begin treating unwarranted consciousness verbs (knows, understands, strategizes) as empirical errors equivalent to poor statistical methodology. Researchers must commit to explaining variance through mathematical and architectural constraints rather than psychological metaphors. However, resistance to this practice is fierce. Tech corporations actively deploy and benefit from anthropomorphic language because it drives massive commercial investment, shields them from liability, and markets their statistical tools as revolutionary 'intelligence.' Precision threatens the valuation of these companies by reducing their magical 'agents' back into what they truly are: highly complex, resource-intensive calculators.

Path Forward​

Looking toward the future of AI discourse, we can analytically map three distinct vocabularies and the specific futures they make possible. The 'Anthropomorphic Clarity' approach (the current status quo seen in this text) relies on projecting consciousness and intent ('the AI knows,' 'the model thinks'). This vocabulary enables highly intuitive, narratively resonant communication that allows lay audiences to easily interact with the software. However, it costs society epistemic accuracy, creating a future where over-trust and automation bias are rampant, and corporate accountability remains permanently obscured behind the illusion of the machine's autonomous agency.

Alternatively, a shift toward 'Mechanistic Precision' ('the model retrieves tokens,' 'processes embeddings') forces exactitude and technical reality into the public sphere. If this vocabulary becomes the norm, supported by journal mandates and educational curricula, regulatory frameworks gain immense clarity. Liability can be accurately mapped to corporate design choices, and users develop a healthy skepticism of statistical outputs. The cost, however, is accessibility; the dense technical language may alienate non-experts, making democratic engagement with the technology more difficult for those outside computer science.

A third future involves 'Hybrid Discursive Frameworks,' where transparency mandates require capabilities to be disclosed mechanistically, but user interfaces are permitted to use acknowledged metaphors. This requires institutions to fund and teach multiple vocabularies, ensuring users can switch between seeing the system as a 'helpful agent' (for ease of use) and a 'statistical matrix' (for evaluation and critique). Ultimately, the choice of vocabulary is a choice of values. Emphasizing mechanism serves the communities prioritizing safety, truth, and accountability, while anthropomorphism serves the communities prioritizing frictionless adoption, commercial marketing, and rapid technological scaling. The future we build will be bounded by the words we choose to describe the tools we make.

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 Scope

I'm starting to zero in on the core of this critical discourse analysis. My aim is to define the exact scope, concentrating on AI metaphors, anthropomorphism, and the projection of consciousness in this scientific paper on LLMs. I am looking for ways to integrate this with the analytical process.

Pinpointing the Analysis

I'm now zeroing in on the provided text, particularly the title and key concepts. I'm focusing on "human-like reasoning," the dichotomy between "formal" and "functional linguistic competence," and the author's stance on LLMs' pragmatic limitations. I see the tension between agent-like and mechanical descriptors as central. My goal is to extract the argument's structure and its key terminology to integrate it into my analysis.

Charting the Linguistic Landscape

I am now focusing on identifying the metaphorical language used to describe LLMs. I have begun a metaphor audit, compiling a list of phrases that ascribe human-like qualities to these systems. "Emerging," "toolkit," "linguistic agent," and "tendency" are prime examples. The goal is to see how these metaphors shape our perception of LLMs.

Analyzing Explanations and Metaphors

I'm now conducting a detailed examination, beginning by mapping the source and target elements of the key metaphors I've identified. Following that, I'm auditing the author's explanations, applying Brown's typology to classify explanatory passages. I've noted a dispositional/theoretical bias in the text, alongside empirical generalizations and functional, reason-based explanations. My analysis aims to understand how these explanations and metaphors shape the argument.

Deconstructing Anthropomorphism

I'm now diving into the specifics. I'm focusing on the language used and I'm rewriting the quotes, stripping away the anthropomorphic elements and restoring human agency in my own language. I've begun crafting the sections for the critical observations. The critical observations will cover agency slippage, metaphor-driven trust, the obscured mechanics, and the context-sensitive language use, and how they relate to the argument. My goal is to expose biases and their origins.

Refining the Argument's Structure

I'm now revising my section structure, targeting the conclusion for rigorous adherence to the given guidelines. The conclusion will now be precisely 5 paragraphs, with a word count between 250-350 words each. I'm focusing on "Contrastive justification" in the evidence field for the first section. I've been reviewing quote examples for agency slippage and the application of hedged/direct classifications. I'm also preparing to analyze the linguistic agency, particularly with quotes focusing on the shift from human responsibility to model "bias" or "emergent ability".

Categorizing Direct/Hedged Language

I've completed the detailed linguistic analysis, classifying quotes based on direct/hedged anthropomorphism. The focus now is mapping actor visibility, classifying agency in quotes that obscure human responsibility. I'm noting examples of "Hidden" actors, particularly the implications for our perception of AI systems. My next step will be to classify examples using my analysis.

Classifying Actor Visibility

I'm now refining the classification of actor visibility within the quotes. I have classified several as "Hidden" due to obscured agency and I'm adding a "Partial" category, such as "Partial (some attribution)" for instances that name products but not corporations. I'm focusing on text to provide examples for each category. I am working to define all the categories, and refine the definitions of each.

Evaluating Actor Visibility

I'm now focusing on specific examples to finalize the actor visibility classifications. For "models are trained on large-scale, diverse corpora, which may allow human-like linguistic competence to emerge spontaneously in silico," I've classified it as "Hidden," since it's passive, obscuring those who curate the corpora. I've added a new quote "E architectures..." to the "Partial" category, as it partially attributes agency. I'm finalizing my justification for the visibility categories. I've drafted all eight required items for the metaphor audit.

Analyzing Explanations Further

I'm now deep into the explanation audit. I've finished the source-target mapping for Task 2, ensuring each quote from Task 1 is fully mapped. Now, for Task 3, I am analyzing each passage based on Brown's typology, elaborating on the explanatory and rhetorical impact of each. I'm focusing on passages that reveal bias, and those concerning pragmatic reasoning, and adding detail to each.

Reframing the Language

I am now actively translating the quotes from Task 1 into more mechanistic language, restoring explicit human agency. For example, "cognitive toolkit of artificial systems" is now "architectural mechanisms designed by engineers," and "linguistic agents" becomes "statistical prediction algorithms deployed by tech companies." I'm also carefully drafting Task 5, the Critical Observations, focusing on five substantial sections, each with a focus on agency slippage, metaphor-driven trust, and obscured mechanics. I am writing robust, comprehensive academic prose that is mindful of word count limits.

Finalizing the Conclusion

I'm now fully immersed in drafting Task 6, the conclusion. I'm structuring it precisely into five paragraphs, focusing each on the themes of systemic integration, rhetorical architecture, material stakes, critical literacy, and a path forward. I am paying close attention to the word count, ensuring each paragraph falls within the 250-350 word range. I'm focusing on "Epistemic" and "Regulatory/Legal" in the "Material Stakes" section. I'm writing comprehensive academic prose.

{ "update": { "header": "Calibrating Categorization", "summary": "I am now meticulously calibrating the contrastive justifications for the Acknowledgment and Actor Visibility categories. I am specifically reviewing how the analysis will not be uniform. I am ensuring that the categories are rigorously defined and that examples are used to provide clear, distinctive, and justifiable boundaries between each. My aim is to address the linguistic and rhetorical nuances, focusing on the distinction between explicit attribution and obscured agency." } }

Refining Actor Analysis

I'm now meticulously refining the actor visibility analysis, specifically the contrastive justifications for the Acknowledgement and Actor Visibility categories. I'm focusing on non-uniformity across the analysis, defining each category meticulously with clear, distinctive, and justifiable boundaries. The analysis will now address linguistic and rhetorical nuances, with a focus on distinguishing between explicit attribution and obscured agency. I'm looking for an explicit metaphor, or a partially visible actor.

Refining Actor Analysis

I'm now meticulously refining the actor visibility analysis, specifically the contrastive justifications for the Acknowledgement and Actor Visibility categories. I'm focusing on non-uniformity across the analysis, defining each category meticulously with clear, distinctive, and justifiable boundaries. The analysis will now address linguistic and rhetorical nuances, with a focus on distinguishing between explicit attribution and obscured agency. I'm looking for an explicit metaphor, or a partially visible actor.

Calibrating Categorization

I am now meticulously calibrating the contrastive justifications for the Acknowledgment and Actor Visibility categories. I am specifically reviewing how the analysis will not be uniform. I am ensuring that the categories are rigorously defined and that examples are used to provide clear, distinctive, and justifiable boundaries between each. My aim is to address the linguistic and rhetorical nuances, focusing on the distinction between explicit attribution and obscured agency.

Analysis Provenance

Run ID: 2026-05-25-tracing-the-ongoing-emergence-of-human-l-metaphor-79caj0 Raw JSON: 2026-05-25-tracing-the-ongoing-emergence-of-human-l-metaphor-79caj0.json Framework: Metaphor Analysis v6.5 Schema Version: 3.0 Generated: 2026-05-25T10:17:10.596Z

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