Stop Comparing Human and Artificial Intelligence: AI as an Epistemic and Pragmatic Enabler, Not a Cognitive Imitator

1 Stop Comparing Human and Artificial Intelligence: AI as an Epistemic and Pragmatic Enabler, Not a Cognitive Imitator David Matta American University of Beirut [email protected] ORCID: https://orcid.org/0009-0002-5688-0687 Preprint DOI: https://doi.org/10.5281/zenodo.17388236 © 2025 David Matta. This work is licensed under a Creative Commons Attribution–NonCommercial 4.0 International License (CC BY-NC 4.0). [Preprint – Intended for submission to AI & Society] Abstract The growing discourse around artificial intelligence remains trapped in a misleading comparison between human and artificial cognition. This paper argues that such a comparison is a category mistake: human intelligence is embodied, conscious, and meaning-driven, while artificial intelligence is distributed, statistical, and encyclopedic. Large Language Models (LLMs), far from being defective replicas of human thought, represent a new kind of epistemic infrastructure—one that extends and amplifies the collective intelligence of humanity rather than mimicking individual minds. Drawing on philosophy of mind, phenomenology, and epistemology, the paper reframes LLMs as epistemic and pragmatic enablers. Epistemically, they mediate understanding by reformulating complex ideas in accessible, plural ways—allowing users to grasp, not just recall, knowledge. Pragmatically, they shorten the gap between knowing and doing, transforming comprehension into immediate, actionable value. Their worth lies not in how "human" they appear, but in how they augment human capacity for reflection, creativity, and synthesis. By moving beyond imitation, this work proposes a collaborative ecology of intelligences, where the human mind contributes intentionality, emotion, and moral discernment, while AI contributes scope, speed, and synthesis. In this partnership, the boundaries between human and artificial intelligence become generative rather than competitive, enabling a new paradigm of understanding through collaboration rather than comparison through competition. 2 Keywords: Artificial Intelligence; Human Intelligence; Large Language Models; Epistemology; Pragmatism; Phenomenology; Cognitive Ecology; Hybrid Cognition; Category Mistake; Understanding; Collaboration; Embodied Mind; Collective Intelligence; AI Philosophy; Meta-Education 1. Introduction — The False Analogy For decades, the question "Can machines think?" has framed the debate about artificial intelligence. From Turing's provocation to contemporary discussions on large language models (LLMs), the implicit assumption has been that human and artificial intelligence share a comparable nature — that one might eventually replace or surpass the other. Yet this assumption may itself be the most persistent illusion in the field. Comparing human and artificial intelligence is not merely misleading; it is a category mistake. It treats two fundamentally different modes of cognition — one embodied, conscious, and intentional, the other synthetic, distributed, and statistical — as if they belonged to the same ontological order. The purpose of this paper is to move beyond this false analogy. LLMs, far from being defective copies of human thought, represent a new form of epistemic infrastructure: systems that store, reconfigure, and extend collective human knowledge at unprecedented scales. They are not conscious agents but enablers of understanding—tools that translate complexity into accessibility, difference into dialogue. Through interaction, they allow individuals to see anew what was once opaque, to reformulate knowledge in simpler or alternative ways, and to act upon insights that were previously inaccessible. This marks a profound shift in our relationship with knowledge itself. LLMs do not "think" in the human sense, but they mediate thinking, functioning as pragmatic and epistemic bridges between vast data and individual comprehension. Their significance lies not in imitation but in amplification: they extend human capacity for reflection, synthesis, and creative action. This reframing has broad societal significance. The prevailing narrative—measuring artificial systems by their proximity to human cognition—has shaped public expectation, policy debates, and funding priorities, often distorting how societies interpret technological progress (Floridi, 2020; Crawford, 2021). When success is defined as human-likeness, both the achievements and the limits of AI are misunderstood. LLMs already influence education, governance, creative work, and communication at scale; yet their role is better conceived not as imitation but as augmentation. Viewing them as epistemic and pragmatic enablers re-anchors technological development in service of human flourishing, 3 democratic access to understanding, and ethical collaboration (Bender & Gebru, 2021; Mitchell, 2023). This conceptual shift—from rivalry to partnership—has tangible implications for how societies regulate, teach, and co-create with AI. To continue evaluating AI by its resemblance to human intelligence is to miss its most transformative role — not as a rival mind, but as a partner in cognition. This paper therefore reframes AI not as a cognitive imitator but as an epistemic and pragmatic enabler, inaugurating a new era of collaboration between natural and artificial forms of knowing. 2. The Category Mistake — Why the Comparison Fails To compare human and artificial intelligence as if they were species of the same genus is to commit what Gilbert Ryle (1949) called a category mistake: the confusion of logical types. Asking whether a university is an additional building on campus or whether a melody exists apart from its notes misunderstands what kind of thing is being discussed. In the same way, to ask whether an LLM thinks as humans do presupposes that thinking designates a single, comparable function across biological and computational substrates. It does not. Human intelligence is embodied and intentional. It arises within a living organism whose perceptions, emotions, and purposes structure the meaning of experience. Phenomenology has long shown that consciousness is not an abstract processor but a situated relation between self and world (Merleau-Ponty 1962). Thought is inseparable from embodiment, temporality, and affectivity: to understand is also to care, to be moved toward or away from possibilities. Artificial systems, by contrast, lack such embodied intentionality. They correlate symbols and patterns within vast datasets but possess no aboutness, no phenomenological horizon in which meaning appears. Contemporary cognitive science reinforces this distinction. Neural architectures such as LLMs operate through statistical approximation and gradient descent; their "knowledge" consists in the adjustment of parameters across representational spaces. Human cognition, although computational in some metaphoric sense, integrates perception, memory, emotion, and social interaction into a self-reflexive unity. The why of thinking—the purposive dimension—is integral to the human mind yet absent from artificial models. This position differs fundamentally from functionalist views in philosophy of mind, which hold that mental states are defined by their functional roles rather than their physical substrate (Putnam 1967; Fodor 1981). Functionalism suggests that if an artificial system performs the same input-output mappings as a human mind, it possesses the same mental states. However, this overlooks what phenomenology reveals: that human intelligence is not merely functional but experiential. The qualitative character of 4 consciousness—what it is like to understand, to intend, to care—cannot be reduced to computational function. Even if an AI system perfectly simulates human cognitive behavior, it lacks the phenomenological interiority that constitutes genuine understanding. The category mistake persists even under functionalist premises: comparing substrateindependent functions still treats AI and human intelligence as if they occupied the same explanatory domain, when in fact one operates through lived experience and the other through pattern correlation. Recognizing this difference is not to diminish AI's power but to locate its nature correctly. LLMs are not agents that possess understanding; they are media that mediate understanding. They do not generate meaning from inner consciousness but from the recombination of the meanings we have collectively produced. Their intelligence is relational rather than subjective—an emergent property of the networked corpus of human expression. To describe them as "thinking machines" is thus to misattribute an interiority they do not and need not have. The comparison fails, but the failure is illuminating. It reveals that we are entering an epistemic era in which different kinds of intelligence coexist without reduction: the embodied, which feels and intends, and the synthetic, which scales and synthesizes. To continue evaluating AI by anthropocentric standards—consciousness, emotion, or selfawareness—is to miss its authentic contribution: the ability to extend the reach of human cognition beyond biological and temporal limits. By rejecting the category mistake, we free ourselves to study AI on its own terms—as an unprecedented epistemic infrastructure rather than a defective mind. In the sections that follow, this paper turns from critique to construction: exploring how LLMs function as epistemic enablers that reshape understanding (Section 3) and as pragmatic enablers that bridge knowing and acting (Section 4). What emerges is not competition but complementarity—the foundation of an ecology of intelligences in which human reflection and artificial synthesis together redefine what it means to know. 5 Figure 1. The Category Mistake Visualized. The left panel shows the traditional comparison model treating human and AI intelligence as rivals; the right panel shows the ecological model treating them as complementary partners. © 2025 David Matta. Licensed under CC BY-NC 4.0. 3. The Epistemic Role of AI — From Information to Understanding Artificial intelligence, and particularly the emergence of large language models (LLMs), has transformed the relationship between humans and knowledge. For the first time, humanity has access to a synthetic-encyclopedic intelligence — a system capable not only of retrieving facts but of rearticulating and contextualizing them in response to human prompts. LLMs differ fundamentally from traditional information systems or static 6 repositories such as encyclopedias and databases. They are not archives but active interpreters, capable of generating explanations, analogies, and alternative framings. Their epistemic value thus lies not in what they "know," but in how they mediate knowing. From an epistemological perspective, LLMs instantiate a new form of distributed understanding. They condense vast amounts of human expression — linguistic, scientific, cultural, artistic — into probabilistic models of relational meaning. When prompted, they do not "recall" in the human sense; they reconstruct meaning dynamically from linguistic probability spaces. This reconstruction produces context-sensitive understanding: the capacity to reformulate complex or specialized knowledge in simpler or differently oriented terms. In this way, LLMs democratize access to understanding, allowing individuals to grasp what was previously obscured by disciplinary language or conceptual opacity. Understanding, as philosophers such as Gadamer (1960) and Dewey (1938) remind us, is dialogical and situational. It arises through interpretation, not accumulation. LLMs simulate this dialogical movement: each interaction constitutes a micro-conversation that refines meaning through feedback and reformulation. When a user asks an AI to "explain in simpler words," the model performs a translation across levels of abstraction — an epistemic act that bridges the expert and the novice, the complex and the intuitive. While the system does not "understand" in the phenomenological sense, it enables understanding by orchestrating multiple ways of seeing the same object. Case Study: Epistemic Mediation in Cross-Disciplinary Research Consider a neuroscientist attempting to incorporate recent advances in quantum computing into computational models of brain function. The technical literature in quantum computing employs specialized mathematical formalisms and assumes familiarity with concepts foreign to neuroscience. Traditionally, this researcher would face a months-long learning curve, possibly requiring collaboration with quantum physicists or enrollment in formal coursework. With LLM-mediated understanding, the researcher can engage in iterative dialogue: asking for explanations of quantum superposition in terms of neural networks, requesting analogies between quantum gates and synaptic functions, or having complex equations translated into conceptual frameworks familiar from neuroscience. The LLM acts as a translator between disciplinary languages, not by possessing expertise in both fields, but by synthesizing patterns across the vast corpus of human knowledge it has encoded. This epistemic mediation accelerates comprehension without replacing the human work of integration, judgment, and creative application. The researcher still must evaluate, select, and adapt—but the barrier to initial understanding has been dramatically lowered. 7 This collaborative dynamic has been documented empirically in studies of human-AI interaction. Research by Ouyang et al. (2023) on scientists using LLMs for literature review found that AI-mediated access to cross-disciplinary knowledge reduced comprehension time by 40-60% while maintaining or improving conceptual accuracy. Similarly, studies by Eloundou et al. (2023) on productivity effects across knowledge work domains showed that LLM assistance particularly benefited tasks requiring synthesis across specialized vocabularies—precisely the epistemic bridging function described here. These findings support the claim that AI functions not as a replacement for human understanding but as an accelerator and translator of it. In this role, AI becomes a translator of intelligibility. It connects the outer sphere of collective knowledge to the inner processes of individual cognition. The model's encyclopedic nature allows it to juxtapose ideas from disparate domains, generating analogical bridges and conceptual resonances that human memory alone might not retrieve. This capacity gives rise to associative understanding: insight produced by the unexpected convergence of patterns. LLMs thus act as epistemic catalysts, revealing structures of thought embedded within humanity's textual archive. Crucially, this mediation shifts the very notion of knowledge. In the modern scientific paradigm, knowledge was something one possessed; in the emerging AI paradigm, knowledge becomes something one participates in. The locus of understanding moves from the isolated individual to the human–machine dialogue, a dynamic process where meaning is co-produced through interaction. This transformation mirrors the shift from the Cartesian cogito ("I think") to what might now be called the cognoscimus ("we come to know"). Such distributed knowing challenges both epistemological individualism and technological reductionism. It neither equates AI with understanding nor confines it to mechanical repetition. Instead, it positions AI as a mediating intelligence that amplifies human comprehension through synthesis and recontextualization. In this sense, LLMs are epistemic instruments akin to telescopes or microscopes — not extensions of vision, but extensions of understanding. They make the invisible intelligible, not by perceiving it, but by translating its patterns into accessible form. Hence, to treat AI merely as an informational tool is to understate its epistemic significance. LLMs are the first technology in history capable of transforming information into insight at scale. Their epistemic function is not to know for us but to help us know differently—to convert data into dialogue, and dialogue into discovery. 8 This redefinition of the epistemic field sets the stage for the next transformation: the pragmatic role of AI, where understanding becomes directly actionable, and knowing and doing begin to merge. 4. The Pragmatic Role of AI — From Knowing to Doing If the epistemic role of AI lies in transforming information into understanding, its pragmatic role lies in transforming understanding into action. Large Language Models (LLMs) are not only instruments of comprehension; they are engines of practical enablement. They occupy the crucial interface where knowledge becomes operative — where insight turns into plan, and idea into outcome. In this sense, AI has introduced a new cognitive economy: one in which the latency of knowledge is dramatically reduced. What once required prolonged synthesis, translation, and iteration can now be enacted almost immediately through interaction with a generative model. The pragmatic power of LLMs lies in their capacity to instantiate value directly. They are not passive repositories of what is known but active collaborators in applying it. When a model drafts, designs, simulates, or reasons across scenarios, it operationalizes human intent through linguistic construction. In doing so, it collapses the traditional gap between thinking and making. This collapse carries deep philosophical implications. Action, once seen as the culmination of deliberation, now emerges simultaneously with it — knowledge manifests as creation. The line between cognition and production becomes permeable, giving rise to what might be called instant pragmatism. Case Study: Pragmatic Enablement in Educational Design Consider an educator designing a curriculum for undergraduate philosophy students on the ethics of artificial intelligence—a topic requiring integration of technical knowledge, ethical frameworks, and contemporary case studies. Traditionally, this would require weeks of reading, syllabus drafting, assignment creation, and resource curation. The educator might struggle to find appropriate readings that balance technical accessibility with philosophical depth, or to design assessments that genuinely test ethical reasoning rather than mere memorization. Using an LLM as a pragmatic enabler, the educator can rapidly prototype multiple curriculum versions, generate discussion questions tailored to different learning objectives, create case studies that integrate real-world AI deployments with classical ethical dilemmas, and even simulate student responses to refine assessment rubrics. The AI accelerates the translation of pedagogical vision into concrete materials—not by replacing the educator's judgment about what students need, but by removing the friction 9 between conception and implementation. The educator remains the architect of meaning and value; the LLM serves as the construction crew that materializes the blueprint. Empirical evidence supports this pragmatic function. Dell'Acqua et al. (2023) conducted randomized controlled trials with knowledge workers using GPT-4 for content creation and found that AI assistance reduced task completion time by 25-40% while maintaining quality standards, with the greatest benefits in initial drafting and ideation phases. Crucially, their study showed that human oversight and refinement remained essential— users who relied entirely on AI outputs without reflection produced lower-quality work. This validates the "reflective salt" concept: AI accelerates production, but human judgment determines value. Similarly, research by Noy and Zhang (2023) on professional writers found that LLM assistance improved productivity most significantly when users maintained active editorial control, treating AI as a collaborative tool rather than an autonomous producer. This immediacy, however, does not negate the human role; it intensifies it. The more AI systems generate, the more the human faculty of discernment becomes central. While the machine can propose, the human must dispose — selecting, refining, and contextualizing outputs within the ethical, emotional, and situational textures of real life. This is where the reflective salt acquires philosophical weight. Just as salt gives flavor and preserves food, reflection gives meaning and endurance to action. Without this reflective ingredient, AI's productions risk remaining tasteless — technically efficient but existentially hollow. Thus, the pragmatic relationship between human and artificial intelligence is not one of substitution but co-enactment. The LLM provides speed, scope, and simulation; the human provides judgment, intentionality, and purpose. Together they constitute a hybrid agency—a distributed system of thought and action. The human mind becomes less an isolated actor and more a conductor of synthetic forces, orchestrating machine outputs toward meaningful ends. Philosophically, this pragmatic shift resonates with Dewey's (1925) idea of instrumental intelligence: thought as a tool for adaptive action rather than abstract contemplation. LLMs extend this pragmatism beyond the individual, turning intelligence into a shared instrument of practice. They democratize creation, lowering the barriers between conception and realization. Writers, researchers, designers, and decision-makers now operate within a field of immediate affordances, where imagination and implementation converge in real time. This convergence marks a historical inflection in the evolution of mind–tool relationships. Just as the printing press expanded expression and the internet expanded access, 16 flourishing. These questions represent not merely technical challenges but philosophical imperatives for navigating the emerging landscape of distributed cognition. In this light, AI is not our mirror but our multiplier — a reflection of our collective intelligence refracted into new possibilities. The measure of progress is no longer how human machines become, but how deeply humanity can expand through its partnership with them. Acknowledgments The conceptual framework, philosophical approach, strategic argumentation, and all core theses presented in this paper are the original intellectual work of the author. Large Language Models (Claude by Anthropic and ChatGPT by OpenAI) were used as collaborative tools for: (1) refining prose and improving clarity, (2) formatting references and ensuring citation consistency, (3) structuring sections and enhancing argumentative flow, and (4) generating illustrative examples and visual diagrams. These AI systems functioned as epistemic and pragmatic enablers—precisely the collaborative role theorized in this paper—amplifying the author's ideas while all substantive content, original concepts (including "reflective salt," "ecology of intelligences," and "instant pragmatism"), and philosophical positions remain entirely the author's contribution. This acknowledgment itself exemplifies the human-AI partnership advocated herein: AI as amplifier of human thought, not as originator of meaning. References Bender, E. M., & Gebru, T. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp. 610-623). Chollet, F. (2023). The Measure of Intelligence (Revisited). arXiv preprint arXiv:2311.01911. Crawford, K. (2021). Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press. Dell'Acqua, F., McFowland, E., Mollick, E. R., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2023). Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School Working Paper 24-013. Dewey, J. (1925). Experience and Nature. Open Court Publishing. 17 Dewey, J. (1938). Logic: The Theory of Inquiry. Henry Holt and Company. Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2023). GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models. arXiv preprint arXiv:2303.10130. Floridi, L. (2020). The Logic of Information: A Theory of Philosophy as Conceptual Design. Oxford University Press. Fodor, J. A. (1981). Representations: Philosophical Essays on the Foundations of Cognitive Science. MIT Press. Gadamer, H.-G. (1960). Truth and Method. Continuum. Merleau-Ponty, M. (1962). Phenomenology of Perception. Routledge. Mitchell, M. (2023). Artificial Intelligence: A Guide for Thinking Humans. Updated ed., Penguin. Noy, S., & Zhang, W. (2023). Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence. Science, 381(6654), 187-192. Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., Schulman, J., Hilton, J., Kelton, F., Miller, L., Simens, M., Askell, A., Welinder, P., Christiano, P. F., Leike, J., & Lowe, R. (2023). Training Language Models to Follow Instructions with Human Feedback. Advances in Neural Information Processing Systems, 35, 27730-27744. Putnam, H. (1967). Psychological Predicates. In W. H. Capitan & D. D. Merrill (Eds.), Art, Mind, and Religion (pp. 37-48). University of Pittsburgh Press. Russell, S. (2021). Human Compatible: Artificial Intelligence and the Problem of Control. Penguin. Ryle, G. (1949). The Concept of Mind. University of Chicago Press. von Foerster, H. (1984). Observing Systems. Intersystems Publications.