Johannes Kleiner and the Definition Problem in Biological Naturalism
Johannes Kleiner, of the Institute for Psychology at the University of Bamberg, the Munich Center for Mathematical Philosophy and the Association for Mathematical Consciousness Science, posted the preprint Defining Biological Naturalism in 2025 (DOI 10.20378/irb-109961). The paper is a five-page commentary on Anil Seth’s target article Conscious Artificial Intelligence and Biological Naturalism, covered on this site in the Seth biological naturalism analysis. Its argument is short and structural. Biological naturalism is indispensable as a counterweight to computational functionalism, but as Seth presents it, the position has no definition, and without one it degrades into a gap-filling strategy.
| Element | The paper’s content |
|---|---|
| The problem | Seth’s biological properties are only contingently unique to living systems |
| The dilemma | Either the properties will soon be built into AI, or the position silently redefines itself around whatever AI has not yet achieved |
| The method | Generalize Putnam’s 1967 definition of computational functionalism to any Formal Description |
| The definition | Biological naturalism is true only if no Formal Description satisfies the functionalist requirements |
The dilemma Kleiner states
Seth’s case for biological naturalism appeals to properties of living systems, and Kleiner grants the case its force. Metabolism, self-maintenance and substrate-dependent dynamics are real biological properties, and a science of AI consciousness that never hears the biological side is metaphysically lopsided. The difficulty is that every property on the list is under industrial pursuit. Predictive processing and the free energy principle are already used to build next-generation AI systems, as in the active inference ecosystems research of Friston and colleagues in 2024. Continuous-time neuromorphic computation with fine-grained timing is implemented in hardware such as the BrainScaleS wafer-scale system commissioned in 2023. Mortal computation, proposed by Hinton in 2022 for the forward-forward algorithm, has industrial applications as its stated goal. Embodied and quasi-agentic robots are being built now.
Any one of those properties, once instantiated artificially, stops being evidence for a biological boundary. Kleiner’s dilemma follows. A biological naturalist programme either commits to properties that near-future artificial systems may meet, or it survives only by moving its goalposts to whatever artificial systems currently lack.
The definition from Putnam
The paper’s positive move generalizes Hilary Putnam’s 1967 explication of computational functionalism. Putnam defined mental states through a Probabilistic Automaton Description of an organism, with three conditions covering which organisms possess such a description, that the organism is not decomposable into parts with separate descriptions of that kind, and that pain co-occurs with a defined subset of sensory inputs. Putnam chose probabilistic automata because they were the most general description of digital computation available in 1967. Kleiner replaces that choice with any mathematical object. A Formal Description, on his terms, is any description of an organism in terms of a mathematical object, and F*-Functionalism holds that feeling pain is possessing the kind of organization a Formal Description describes.
Biological naturalism then receives an explicit negative definition. It is true only if there is no Formal Description satisfying the functionalist conditions for conscious organisms. Conscious life, on this definition, is a system that admits no formal description of the functionalist kind. The definition has a direct consequence for artificial systems. Building an artificial system requires a formal description of what is being built, so whatever else such a system is, it cannot be conscious in virtue of what it is.
Comparison to The Consciousness AI
This project treats consciousness as an emergent property of physical organization and holds that claims about which systems are conscious must be testable. Kleiner’s paper supplies the methodology for the biological side of that test. A position without a definition cannot be tested, and the paper converts Seth’s intuition into a falsifiable form, no formal description of the required kind exists for conscious systems. The project’s own substrate work, maintained in the tlcdv/the_consciousness_ai repository, operates under computational functionalist assumptions, which makes Kleiner’s negative definition the sharpest statement of what the project would have to be wrong about. The paper’s companion piece with Erik Hoel, the dilemma about falsifiability and non-triviality in consciousness theories, is examined in the Kleiner-Hoel dilemma post.
What the Definition Leaves Open
The definition is a proposal about method, and the paper is explicit that it discusses how such a definition might be obtained rather than settling the metaphysics. What remains open is whether conscious systems truly admit no formal description, which is an empirical thesis about what can be modeled, and whether the definition’s negative form is stable as modeling improves. The commentary literature on Seth’s target article, of which this preprint is one response, is examined in the BBS commentary responses post. The state of the substrate debate is reviewed in the flagship overview of AI consciousness research.