Do Bonsai Suffer? Plants, Agency, and Valence under Kolmogorov Theory

A substrate-neutral theory of agency turns a seemingly naive question about bonsai into a precise ethical and empirical problem.

The core intuition is simple: if a system has an internal objective function — some built-in sense of what counts as going well or badly — then you can harm it without killing it. A bonsai tree is kept alive for decades or centuries, yet its own developmental goals (extend roots, reach light, grow, reproduce, commit to a form) are continuously overridden by the caretaker. The question is whether that constitutes harm in any meaningful sense. The answer depends entirely on whether the plant has anything like an internal evaluative dynamics in the first place.

The paper's main theoretical move is to apply Kolmogorov Theory (KT), an algorithmic-information framework where agency is defined not by neurons or verbal report but by a specific organizational structure: a system with a boundary, an internal model of its world, a policy space, an objective function, and update rules. The key bridge is the Good Regulator Theorem — any system that successfully regulates viability-relevant variables against disturbance must, by information-theoretic necessity, carry model-like information about what it's regulating. Plants do this continuously, using calcium waves, electrical signals, hormones, hydraulics, gene-regulatory networks, and morphology as the distributed substrate. No brain required. The plant's body is partly a stored model of past environmental regularities.

Valence — the positive or negative quality of experience — is then formalized as the rate at which the agent's regulator is closing or opening a "compression gap": how well is the system currently bringing its viability-relevant variables under control? The plant's objective function is modeled as multiplicative across several proxies (metabolic survival, growth, root exploration, light capture, reproduction, damage repair, morphogenetic freedom). Multiplicative matters: if any one proxy collapses to zero, the whole objective collapses, regardless of how well the others score. A bonsai keeps metabolic survival high while chronically pinning nearly every other proxy near zero. The regulator perpetually tries to extend, gets cut back, tries again, gets cut back — not equilibrium, but sustained frustration operating exactly at the plant's own developmental timescale.

The ethical framework that follows is clean and principled. Harm is registered along the agent's own evaluation timescale. A fast event — mowing grass, a frost killing an annual — terminates tissue faster than the agent can register a valence trajectory. That's "mercy by speed." Bonsai is the structural opposite: slow, sustained, individual, tuned precisely to the timescale at which the tree itself develops. The paper explicitly dissolves the intuitive equivalence between mowing a lawn and shaping a bonsai over decades. They are not the same kind of act. The paper is careful throughout to frame all of this as conditional: if KT is right about what agency requires, and if plants satisfy those criteria sufficiently, then bonsai is a candidate case of low-valence morphogenetic captivity. The empirical question remains open, and the paper proposes a concrete experiment — release constrained plants into free conditions and measure whether recovery dynamics show a sustained, multi-component reassertion of developmental objectives, distinct from ordinary wound healing.