Work
| Internally-funded improbability presentation | What distinguishes a living system from a merely dissipative structure like a hurricane? Both maintain themselves far from equilibrium, but only living systems invest energy indirectly—first into internal structures (Bauer’s “free structural energy”), which then fund work. Argues that organisational closure alone is insufficient for agency: Pattee’s epistemic cut, where symbols constrain dynamics according to their selection history, is necessary. Develops Maxwell’s demon as a minimal model of the agent, with Landauer’s principle setting its cost and KL divergence quantifying its effect. ECogS, OIST (2025) |
| Activity and knowledge presentation | How should we understand the relationship between mind and world—the central problem of philosophy—in a way productive for cognitive robotics? Traces two parallel lines from the modern tradition: Ilyenkov’s dialectical materialism, where the ideal emerges through material activity, and Nishida’s later concepts of acting-intuition and the historical body, where knowing and acting are inseparable. Both ground knowledge in embodied, culturally-embedded activity, but diverge on a key point: for Ilyenkov, consciousness is a product of this activity; for Nishida, it is a precondition. Draws out implications for developmental robotics as experimental philosophy. Asada Laboratory, Osaka University (2025) |
| DevNCA presentation | Spontaneous activity during early development (e.g., retinal waves) structures neural organisation, but can this principle be harnessed in artificial networks, and does the structure of the activity matter? Evolves STDP plasticity rules for spiking neural networks on a control task, injecting activity during a developmental phase. Developmental activity proves essential (66% performance reduction without it), and NCA-generated structured patterns outperform random activity by 7.5%. Artificial Life, Kyoto (2025) |
| Another kind of clay presentation | Biological systems are unreliable at every level—molecular, genetic, cellular, neural—yet evolution has not eliminated this variability. Why? Drawing on Brembs’ active-dynamic view of nervous systems and Pattee’s biosemiotics, argues that intelligence emerges from the sculpting of spontaneous dynamics through symbolic constraints. Proposes a vector model: agency as the costly magnitude of divergence from passive distributions, intelligence as the direction this divergence takes. TONAL workshop, Artificial Life, Kyoto (2025) |
| Temporal continuity and machine consciousness presentation | If consciousness involves temporally continuous experience, what kinds of computational architectures could support it? Distinguishes three types of continuity—perceptual (observer-projected), behavioural (coherent state-transitions), and intrinsic (continuous substrate unfolding)—and applies thought experiments involving spatially separated processing to LLM inference and beyond. Argues that if spatial contiguity matters, behavioural continuity is insufficient, favouring analog architectures with a continuous dynamical core. Models of Consciousness, Sapporo (2025) |
| Can a single cell learn? presentation | Reports of learning in unicellular organisms suggest that neural networks may not be necessary. But do these cases—habituation in Stentor, conditioning in Paramecium and amoebae—really constitute learning? Reviews the evidence with attention to molecular mechanisms and control procedures, then proposes a substrate-neutral distinction: adaptive behavioural plasticity operates via scalar molecular concentrations, while learning proper requires modification by complex spatiotemporal patterns. ANZPhilBio (June 2025) |
| Logic and artificial life presentation | What can Nishida’s later philosophy—acting-intuition and the historical body—offer the project of cognitive developmental robotics? Finds deep parallels with Soviet developmental psychology (Vygotsky, Leontiev, Meshcheryakov) in their shared emphasis on embodied, culturally-transmitted activity as the ground of knowledge, but also a key tension: where the dialectical materialist tradition takes consciousness as a product of activity, Nishida takes it as a precondition. Argues that robotics offers an experimental method for these philosophical questions, while both traditions offer it a richer theoretical framework. International Society of East Asian Philosophy, Fukuoka University (December 2024) |
| “The brain as a dynamically active organ” presentation | Is the dominant view of brains as input-output processors the right framework? Reviews Brembs’ three-part argument: intrinsic activity is primary, with sensory input modulating rather than driving behaviour; organisms use reafferent behaviour to actively probe their environment; and the variability of intrinsic activity is evolutionarily advantageous. Even the simplest organisms exhibit active dynamics that are sculpted by environmental feedback rather than determined by it. Cortical Labs journal club (December 2024) |
| Symbol grounding in the age of LLMs presentation | LLMs require orders of magnitude more data than children to acquire language—why? Formalises the inefficiency via conditional Kolmogorov complexity: integrative perceptual access reduces the data needed for linguistic competence. Multi-modal training helps, but depends on symbolic structures already laid down by perceptually grounded language use. The fundamental asymmetry is that LLMs build from language to the world, whereas humans crystallise language around experience. Robophilosophy, University of Aarhus (August 2024) |
| The symbol as such presentation | Multi-modal grounding improves LLM efficiency, but it depends on structures established by prior language use. What bootstrapped the process? Argues that perceptual consciousness—secondary sensory integration, where objects are represented as distanced presence rather than mere affordances—provides access to the symbol as such. This conscious grasp of symbolic structure, following Pattee, determines its effect as a constraint, providing the minimal ontology around which language initially crystallises. The Epistemic Significance of Perceptual Consciousness, Jagiellonian University (May 2024) |
| Egocentric speech in children and machines presentation | Chain-of-thought prompting dramatically improves LLM reasoning, but if each token involves only local inference, where does reasoning reside? Draws on Vygotsky’s work on egocentric speech, where children use self-directed language to structure attention and decompose problems. In LLMs, tokens serve as text (for the user), as embeddings (for the model), and as the autoregressive structuring of attention—chain-of-thought emerges from this synergy, leveraging deep patterns in language that parallel the cultural frameworks children internalise for problem-solving. Agency and Intentions in AI, University of Göttingen (May 2024) |
| After Babel, the horizontal war paper | What is the relationship between the sacred and the technological, and how does modernity inherit the structures of the ancient city? Traces a path from the biblical City to the Technique of modernity through Ellul’s theological and sociological works, developing the notion of inscriptions and the relationship between reality and the remainder. Illustrates through a literary analysis of McCarthy’s Blood Meridian, particularly the character of Judge Holden as an avatar of inscriptive mastery. Philosophical Journal of Conflict and Violence (2023) |
| The view from nowhere dissertation | How does Technique—Jacques Ellul’s term for the autonomous logic of modern technology—achieve its totalising unity? Argues that mediating inscriptions (text, images, legal codes, maps, engineering drawings) cultivate a perspective detached from any particular standpoint, and that this “metaphysics of inscription” underlies the specific characteristics Ellul attributes to Technique: automatism, self-augmentation, universality, and autonomy. Master’s dissertation, University of Auckland (2022) |