Work

Can a soft body of identical contractile cells produce edge-sensitive behaviour without assigned sensor or motor roles? Selected synthetic specimens learn to approach an edge, distinguish a cliff from a goal, and reverse when needed. Performance is strong on the terrain family used in evolution but transfers only partially to new distances and body shapes; one specimen, S4, suggests a distinct morphology-dependent strategy.

Artificial Life 2026, Waterloo, Canada

What makes a molecular distinction informational in a living system? The talk treats biological information as organised molecular difference within evolved sender–receiver systems. Rate–distortion identifies which distinctions must persist and with what fidelity, while physical chemistry constrains how long they can survive and how they can be reset. Bacterial chemotaxis supplies a concrete case through CheY-P signalling and MCP methylation.

Artificial Life 2026, Waterloo, Canada

How do living systems differ from other dissipative structures, and what would a thermodynamic account of agency require? Bauer’s free structural energy describes life’s indirect use of energy: first to maintain internal structure away from equilibrium, then to fund work. Organisational closure does not by itself provide regulatory control; the argument therefore turns to Pattee’s epistemic cut and symbols whose effects on dynamics depend on their selection history. Maxwell’s demon supplies a minimal model, while Landauer’s principle and KL divergence motivate a proposed lower bound on the thermodynamic cost of divergence from a passive distribution.

ECogS, OIST (2025)

How does activity relate mind to world? For Ilyenkov, the ideal emerges through material activity and mind is structured by internalised materiality. Nishida’s acting-intuition makes self and thing determine one another, while the historical body makes embodiment historical and history embodied. The accounts place consciousness differently: as the structure of material activity for Ilyenkov, and as the self-differentiation of the primary basho of pure nothingness for Nishida. Cognitive developmental robotics offers concrete instantiations through which to examine these principles.

Asada Laboratory, Osaka University (2025)

DevNCA
presentation

Can structured developmental activity improve plastic self-organisation in artificial spiking controllers? On a cart-pole task, evolved STDP rules are compared under three developmental conditions: no injected activity, random activity, and patterns generated by a neural cellular automaton. In the reported setup, removing development reduced mean fitness by 66%; NCA-generated patterns increased it by 7.5% over random developmental activity. These results do not identify which properties of the patterns matter, and broader tasks, hyperparameters, and embodied activity remain open.

Artificial Life, Kyoto (2025)

How might intrinsic variability contribute to agency and intelligence rather than count merely as noise? Brembs motivates treating intrinsic activity as primary and sensory input as its modulation. Pattee supplies an account of symbols as replicable, selected physical constraints whose meaning lies in the specificity acquired through their selection history. A proposed vector model treats agency as the costly magnitude of divergence from a passive path distribution and intelligence as the direction, or qualitative specificity, of that divergence. The model is conceptual and outcome-agnostic.

TONAL workshop, Artificial Life, Kyoto (2025)

What continuity would a substrate require if phenomenal experience itself is temporally continuous? Perceptual continuity is projected by an observer; behavioural continuity lies in coherent state-transition rules; intrinsic continuity belongs to a substrate that unfolds without update steps. Thought experiments involving separated projectors and language-model inference suggest that behavioural continuity is inadequate if consciousness requires spatiotemporally localised continuity. This conditional argument favours analogue systems with a continuous dynamical core, but treats continuity as necessary at most, never sufficient.

Models of Consciousness, Sapporo (2025)

Can single cells learn under a broad functional definition, and should their adaptive changes count as learning proper? Cases include E. coli chemotactic adaptation, Stentor habituation, and disputed conditioning reports in Paramecium and amoebae; their evidence and controls differ substantially. A proposed substrate-neutral distinction classifies behaviour modified by scalar molecular concentrations as plasticity and behaviour modified by complex spatiotemporal patterns as learning. This criterion does not rule out unicellular learning in principle, but the reviewed cases do not meet it.

ANZPhilBio (June 2025)

What can Nishida’s acting-intuition and historical body contribute to cognitive developmental robotics? Acting-intuition joins action and perception; the historical body describes embodiment as historically conditioned and history as enacted through bodies. Comparison with Vygotsky, Leontiev, and Meshcheryakov finds a shared emphasis on embodied and social development, but a difference over consciousness: Nishida treats it as a precondition of human culture, whereas the Marxist account treats it as a product of material culture. Motor babbling, social robotics, and lifespan development show where these commitments bear on artificial systems.

International Society of East Asian Philosophy, Fukuoka University (December 2024)

Are nervous systems primarily input–output systems, or do they primarily generate behaviour through intrinsic activity that sensory input modulates? Brembs argues for the second view through variable spinal reflexes, phototaxis, escape responses, Aplysia feeding, C. elegans reversal, and bacterial chemotaxis. He uses these examples to support two related claims: organisms use reafferent behaviour to probe their surroundings, and variability can be evolutionarily useful. The remaining questions concern how to define intrinsic activity and whether artificial or cultured neural systems should constrain, harness, or otherwise adapt to it.

Cortical Labs journal club (December 2024)

What would it mean to ground symbols from world to model rather than merely align them from model to world? On this account, embodied activity and perceptual experience act as morphological computation, supplying a minimal ontology that reduces the linguistic data needed for learning. The hypothesis is expressed as K(L|E) < K(L), where E is integrative access to embodied perceptual experience. Two visual-language studies offer initial cross-modal support, but neither tests genuinely multimodal developmental experience. Whether artificial intelligence needs such grounding depends on whether the aim is practical alignment or an account of how meaning originates.

Robophilosophy, University of Aarhus (August 2024)

The symbol as such
presentation

What might perceptual consciousness contribute to the origin of symbols when cross-modal data can improve language learning without it? Primary sensory integration extinguishes inputs in outputs; secondary integration represents sensory objects as a distanced presence. Combined with Pattee’s account of symbols as constraints, conscious access to the symbol as such supplies a proposed structure around which language can first crystallise. Later models can exploit structures already established through language use. This is a conceptual claim, not an empirical demonstration that consciousness is required.

The Epistemic Significance of Perceptual Consciousness, Jagiellonian University (May 2024)

Can Vygotsky’s account of egocentric speech provide a useful analogy for chain-of-thought in language models? In children, self-directed speech organises attention, divides operations into subordinate problems, and is later internalised. In a language model, intermediate tokens extend serial computation, provide a recurrent scratchpad, and alter the attention available at the next step. The analogy proposes that chain-of-thought combines autoregression with self-attention to exploit patterns learned from language; it does not claim that children and models share the same mechanism.

Agency and Intentions in AI, University of Göttingen (May 2024)

How does Ellul’s modern Technique continue the theological pattern of the City? Enoch, Resen, Babel, and Nineveh unfold themes of human security, mastery, self-naming, and war against rival powers. Judge Holden’s acts of recording and destruction in Blood Meridian make inscription the bridge from City to Technique. Technique resumes Babel’s impossible project as Nineveh’s horizontal war, using inscriptions to construct a human order that remains vulnerable to what those inscriptions exclude as the remainder.

Philosophical Journal of Conflict and Violence (2023)

How can Ellul’s concept of Technique be both unified and internally diverse? The proposed metaphysics of inscription concerns the material practice and mindset of working through texts, images, codes, maps, and drawings. Inscriptions decontextualise, immobilise, flatten, and organise aspects of actuality into a manipulable virtual reality, while encouraging an allocentric “view from nowhere.” This ordering mode helps explain Technique’s automatism, self-augmentation, monism, and universalism, although the dissertation explicitly stops short of claiming that Technique and inscription are identical.

Master’s dissertation, University of Auckland (2022)

How do the image schemas of containment and many-sidedness organise different forms of inference? Aristotle’s logic treats predication as containment; Jain anekāntavāda qualifies claims by standpoint through nayavāda and syādvāda. The comparison does not rank the systems: container logic favours objective precision, while many-sidedness accommodates vagueness by accepting perspectival limits. Its application to wave–particle duality remains suggestive rather than settled.

Unpublished