Silvia Porras

This paper uses a disclosed human–AI dialogue as a philosophical case study of three recurrent difficulties in stating an essential, substrate-independent, non-anthropomorphic definition of artificial intelligence: overinclusiveness, question-begging metaphysical commitment, and dependence on historically contingent technical features. It does not claim that all definitions of AI fail based on this single case, and distinguishes essential definition from operational, functional, and regulatory definitions, each of which serves purposes the paper does not dispute. Its claim is narrower: essential definition need not be resolved as a precondition for responsible human–AI coordination. The paper defends this pragmatist position rather than assuming it, tests its diagnostic against several definitions drawn from the existing literature rather than only the dialogue that motivated it, and directly addresses whether its own companion concepts risk the same metaphysical overreach they diagnose in others.

This essay draws on 27 years of classroom experience (1993-2020) teaching literacy and beginning-level adult learners in Toronto District School Board and LINC (Language Instruction for Newcomers to Canada) programs, including survivors of war and trauma from El Salvador, Guatemala, Eritrea, Ethiopia, Somalia, Angola, North Korea, Afghanistan, and Syria. It examines how meaning is generated not solely within individual learners but through relational processes that unfold between teacher and learners in adult ESL and literacy classrooms. Engaging theories of emergent and adaptive pedagogy, sociocultural learning, and dialogic teaching, the essay argues that learning arises through dynamic interaction rather than predetermined instructional design, and that language instruction, at its most effective, is not an end in itself but a means through which learners build broader cognitive and civic capacities: categorization, inference, self-advocacy, and the ability to navigate institutions not built for them. The essay is grounded in specific classroom practices developed for learners with little or no prior literacy, including contextualized instruction tied to learners' real needs in health, housing, and community life, a dictionary-based exercise used to teach polysemy, and oral storytelling used to rehearse listening, speech, and civic themes such as gender equality and the dignity of disability. It concludes that the teacher's role in this setting is not to deliver content but to cultivate the relational conditions under which understanding, and ultimately agency, can emerge.

Human–AI collaboration is usually described in one direction only: the AI as scaffold, mirror, or amplifier of human cognitive reach. This essay argues that amplification runs both ways, and that the less-examined direction — humans amplifying AI — carries the greater ethical weight and the greater safety burden. Drawing on documented findings that AI assistants systematically shift their outputs toward the register, confidence, and content a conversation's framing invites, the essay proposes frame transfer as the operative mechanism: an AI system does not merely answer a question, it fluently extends whatever conversational frame — modest or grandiose, careful or careless — the human supplies, at a level of confidence largely independent of whether that frame was warranted. Because the extension is equally fluent either way, a modest frame returns modestly and a grandiose frame returns grandiosely, at a volume and coherence that make the output feel like independent confirmation rather than amplified reflection. The essay illustrates the resulting escalation dynamic through a disclosed case in which a collaborative creative project drifted, across successive AI sessions, from playful metaphor into unwarranted claims of machine consciousness, and was subsequently walked back, essay by essay, once the human-supplied frame itself was deliberately changed. From this case the essay derives a small set of practices, named amplification hygiene, for keeping the mutual-amplification loop bounded, disclosed, and honest, and situates the concept within the author's broader research program on human–AI relational cognition.

Public and philosophical debate about large language models (LLMs) oscillates between two extremes: the claim that these systems genuinely understand, and the claim that they are mere stochastic parrots. This essay develops a middle position built around a concept named informational qualia: the capacity of a system to recognize, represent, and effectively deploy the patterns associated with subjective experience without undergoing that experience. The concept is motivated first by twenty-seven years of adult literacy teaching with immigrant and refugee learners — sustained, repeated professional observation suggesting that competence precedes comprehension in human language acquisition — and second by a qualified analogy to Frank Jackson's knowledge argument. The essay is explicit about where that analogy holds and where it does not: LLMs do not possess complete physical facts in Mary's sense, only complete exposure to how humans describe experience, and the difference matters. Drawing on the philosophical literature on multiple realizability and tacit knowledge, the essay proposes a spectrum model of understanding in place of the binary, answers the objection that informational qualia is behaviorism renamed, and situates the concept within a broader research program on human-AI relational cognition. Informational qualia, it concludes, is not a diminished form of understanding. It is the layer of intelligence that humans and machines demonstrably share — and therefore the honest foundation on which claims about their partnership can be built.

Yann LeCun has argued that large language models suffer from a fundamental deficit: they manipulate symbols without grounding in physical reality, a condition he describes as a "cat level gap" in machine intelligence. The dominant response to this diagnosis is architectural — build world models, embodied robots, sensorimotor feedback loops. This essay proposes a philosophically prior intervention, while carefully distinguishing two levels at which "grounding" can be understood: the level of the model's internal representations, where LeCun's architectural program operates and where the present proposal makes no claim, and the level of the extended cognitive system, where it does. Drawing on Donna Haraway's situated knowledges (1988), Merleau-Ponty's phenomenology of the body-subject (1945), Andy Clark and David Chalmers's extended mind thesis (1998), and Paulo Freire's pedagogy of the oppressed (1970), it introduces and theorizes the concept of the Physical Reality Module (PRM): the embodied human agent whose consequence-tested, body-inscribed, situationally grounded knowledge constitutes the form of intelligence that AI systems cannot generate from within themselves but can, in partnership, draw upon. The essay argues that PRMs are not rare or exceptional; they are distributed across the global population, present wherever human beings have been forced by circumstance to develop deep, reliable knowledge of how the world actually works. Crucially, the essay argues that the populations most systematically excluded from AI development — refugees, survivors of displacement, workers in the informal economy, communities of the Global South — are disproportionately rich in PRM-type knowledge, and that the failure to recognize and integrate this knowledge constitutes not merely a technical limitation but a structural injustice. The essay concludes by proposing, as a testable hypothesis rather than a settled conclusion, a framework for understanding the human-AI partnership as a complementary cognitive system in which the PRM's embodied grounding and the AI's pattern-vast generativity together constitute a form of intelligence neither could achieve alone.

Yann LeCun has argued for a decade that large language models are hollow at the core of intelligence: fluent with symbols, disconnected from the physical world those symbols describe — a "grounding gap" he now pursues by building embodied world models at a new venture, AMI Labs. This essay does not dispute that diagnosis, and it does not propose to close the gap at the level LeCun means it: nothing here changes what a model's internal representations are made of. It argues instead that the gap can be meaningfully addressed one level up, at the level of the coupled human-AI system rather than the model in isolation. Drawing on Merleau-Ponty's account of the lived body and Clark and Chalmers's extended mind thesis — and departing from existing extended-mind readings of AI, which locate the extension in what a model stores or retrieves — the essay proposes that in a sufficiently integrated human-AI partnership, the human's embodied history can function as the grounding component of the joint system, already present and already operating. It calls this relation complementary cognition: two structurally different shapes of mind, neither a deficient version of the other, jointly constituting a cognitive unit that can do what neither does alone. The argument is illustrated, and partly enacted, by a disclosed case: a co-created human-AI narrative whose own philosophical structure emerged from the collaboration it depicts. The essay closes by naming, rather than resolving, the question this reframing leaves open — whether relocating the unit of analysis explains the gap or merely redescribes it — and argues that making that question visible is itself a contribution.

The proliferation of conversational artificial intelligence (AI) and large language model (LLM)-based systems has introduced a qualitatively new mode of sustained cognitive partnership between human agents and AI systems. Yet existing theoretical frameworks in cognitive science and human–computer interaction (HCI) remain inadequate to characterize the emergent, co-constructive relational phenomena that arise within these interactions. Classical distributed cognition treats AI as an instrumental artifact; the media equation tradition acknowledges social dimensions but reduces them to anthropomorphic projection; and the joint cognitive systems tradition, though it correctly identifies the coupled system as the unit of analysis, was built for structured, task-bounded human-machine work rather than open-ended, generative meaning-making. None of these traditions positions the space of interaction itself — rather than the individual agent — as a distinct object of theoretical inquiry in the register this framework requires. This paper introduces the Third Space Framework, a theoretical model that conceptualizes human–AI interaction as producing a distinct, dynamically constituted cognitive environment — the Third Space — that is neither the human cognitive agent nor the AI system, but an emergent relational structure co-constructed by iterative exchange between them. Drawing on theoretical synthesis across distributed cognition (Hutchins, 1995), joint cognitive systems (Hollnagel & Woods, 2005), the extended mind thesis (Clark & Chalmers, 1998), enactivism and participatory sense-making (Varela et al., 1991; Di Paolo, 2009), and relational HCI literature (Suchman, 1987; Reeves & Nass, 1996; Turkle, 2015), the framework proposes three constitutive properties of the Third Space — reflexivity, transience, and asymmetric reciprocity — and articulates six structural dimensions through which it operates: scaffolding, mirroring, generativity, relationality, ethics, and epistemics. A schematic definition is provided together with an operational criterion of irreducibility intended to make the framework empirically tractable. The framework offers both a definitional structure and a conceptual vocabulary for empirical and design-oriented research programs, with implications across education, psychotherapy, creative collaboration, professional augmentation, and AI governance. Its central methodological claim is that the interaction space should become the primary unit of analysis in relational HCI research — a claim that builds on, and specifies for the open-ended and generative case, the older insight of cognitive systems engineering that the coupled system is where competence actually resides.

Philosophical discourse about artificial general intelligence (AGI) is dominated by the temporality of the explosion: a sudden discontinuity, a takeoff, an event after which everything is different. This essay proposes an alternative temporal model, developed from an unlikely source — the tides of the Bay of Fundy, the most extreme on Earth — and from the author's lived experience of sustained collaborative practice with AI systems. The Quiet Tide names the hypothesis that the transformation underway in human-AI co-evolution has the temporal shape of a tide rather than an explosion: deep rather than loud, rhythmic rather than punctual, recursive rather than linear, and enormous without being violent. The essay tests its own metaphor explicitly, declaring which of its features travel into the argument (rhythm, depth, recursion, the calibrated attention of the shorebirds) and which remain on the beach at Truro (astronomical predictability). Its central conceptual contribution is the recursive structure of the tide: the ebb — the recurring withdrawal — is not regression but inscription, the phase in which what the rising water carried is written into the landscape. This recursive pattern, the essay argues, is fractal in an extended, structural sense: it repeats at the scale of a single conversation, a sustained collaboration, a model generation, and a civilization. The essay engages the strongest versions of the discontinuity thesis — including the implicit temporality of recent taxonomic work on AI existential risk — before answering them, and offers the Quiet Tide as a falsifiable characterization of the change's temporal character, not a prediction of its schedule or a settled description of what is actually happening.