Do Bonsai Suffer?

A bonsai, in the usual case, is not a genetically miniaturized tree. It is a normal tree—maple, juniper, pine, ficus—maintained for decades or centuries inside a constrained developmental regime: shallow container, periodic root pruning, shoot pruning, branch wiring, controlled watering, controlled nutrition, periodic repotting. From the outside, a well-kept bonsai is a paragon of viability. It does not die. It does not even age in the way an unconstrained tree ages. It persists.

That is what makes the question do bonsai suffer? interesting rather than absurd. The question is interesting because external viability is guaranteed. The bonsai is not killed, neglected, or starved. It is shaped. The shaping continues at the timescale at which the tree itself develops. Roots are cut back exactly when they would have explored a new soil volume; shoots are clipped exactly when they would have reached a new light volume; the trunk is wired exactly when it would have committed to a new growth direction. The caretaker regime is, in effect, an external policy that chronically overrides the plant's own developmental policy.

If the plant has no internal evaluative dynamics, the question dissolves: there is no frustrated objective, only sculpted material. If the plant does have internal evaluative dynamics, the bonsai is a candidate case of something quite specific—a system whose objective function is being continuously and partially satisfied (it lives, it photosynthesizes) while being continuously and partially frustrated (it cannot extend, explore, fruit, recover, or commit) over the timescale at which it itself operates.

The naive form of the question is the right form. But to answer it we need a substrate-neutral theory of agency. Animal-centric reasoning will not do, because plants do not have neurons, do not produce verbal reports, and do not exhibit nociceptor-mediated pain behaviour. If the only legitimate route to ascribing experience to a system is animal-like architecture, the question is closed by definition. But that is a definitional move, not an empirical one. It substitutes a high-confidence evidential route to consciousness (animal report) for a necessary condition on consciousness (animal-like architecture).

Kolmogorov Theory in 90 Seconds

Kolmogorov Theory (KT) is the algorithmic-information account of agents and structured experience we have developed in the BCOM corpus (Ruffini, 2017; Ruffini et al., 2024). Under KT, structured experience is associated with agentic organization: a system that models its world/self-relevant state, evaluates trajectories relative to an objective function, and acts to preserve or transform its own pattern.

KT predicts structured experience for systems that instantiate this organization with sufficient integration and compressive structure. Different substrates—neurons, cells, distributed chemical/electrical networks, software—can in principle satisfy the definition. The question for plants becomes: do they instantiate the architecture?

The Good Regulator Argument: Compression Forces Model

Conant and Ashby's classic Good Regulator Theorem (1970) says that every good regulator of a system must be a model of that system in the relevant sense. We have recently sharpened this in algorithmic-information terms (Ruffini, 2026): if a regulator $R$ reduces the algorithmic complexity of a world/readout trajectory relative to an unregulated baseline, then high shared algorithmic information between the world $W$ and regulator $R$ becomes favored.

Sustained positive $\Delta$ (regulation gap) is evidence that the regulator carries model-like information about the regulated process. Crucially, this argument is substrate-blind. It does not care whether the regulator is a brain, a thermostat, or a distributed plant.

For plants, the candidate regulator is not a brain. It is the coupled distributed plant system: membranes, photoreceptors, mechanosensors, ion channels, hormone networks, vascular transmission, gene-regulatory networks, meristems, roots, shoots, and morphology. The plant's "model" need not be a centralized map. It may be encoded in distributed state variables, gradients, thresholds, oscillators, tissue architecture, growth history, and update rules. Under KT/AIT, modelhood is not defined by verbal report or neural representation, but by usable compressive structure.

Plants as Algorithmic Agents

The KT decomposition maps cleanly onto a plant once we allow non-neural, distributed implementations.

The important move is to avoid confusing centralization with agency. A plant is not a tiny animal without muscles. It is a different kind of agent: sessile, modular, distributed, hydraulic–electrical–chemical, and morphogenetic. Plants only look static because human perception samples them at the wrong temporal scale. Darwin and Darwin (1880) already noticed that root tips, shoots, and climbers move, search, and reconfigure—just slowly. Animals mostly move the body through space; plants mostly move their future body through growth.

Telehomeostasis and the Proxy Bundle

Here is a piece of formalism that becomes load-bearing for the bonsai argument. Living agents do not directly compute the true objective they are evolutionarily mirroring. The true KT objective for a living being is telehomeostasis—the persistence of its own algorithmic pattern, including kin and lineage. But the agent's internal $\widehat{\mathrm{OF}}_A$ is not telehomeostasis itself; it is an approximation in which evolution has hard-wired a set of heuristic proxies that, on average and in the ancestral environment, correlate with telehomeostasis.

We propose that the plant objective is not an additive weighted sum of these proxies but a multiplicative form—closer to a Cobb–Douglas product—so that pinning any proxy near zero collapses the joint score regardless of how high the others read. In the ancestral environment, any single proxy chronically near zero predicts eventual loss of telehomeostasis, so a proxy bundle that collapses when any one collapses is exactly what selection would produce.

This matters because the bonsai regime sits squarely in the regime where the linearization of this product fails: some $\phi_i$ are pinned near zero by external intervention while $\phi_1$ (metabolic viability) is artificially propped up. The proxy bundle has been pulled apart from the survival statistic it normally tracks.

What Valence Looks Like in Compression Terms

Under KT, agent-internal valence is naturally formulated as the rate at which the regulator is closing its own complexity gap along the current policy:

The first term rewards currently compressing the readout—the regulator is succeeding now. The second term penalizes a sustained gap from the viable manifold $K^\star$—the regulator is stuck above its setpoint. Three regimes follow: positive valence when $K_t^\pi$ is decreasing toward $K^\star$ (the policy is working); neutral valence at setpoint; negative valence when $K_t^\pi$ is increasing or stuck above $K^\star$ (the regulator is losing or has lost control). Success at being a good regulator feels good to the regulator, in the agent-relative sense.

For a chronically constrained plant, the prediction is sharp: the regulator perpetually opens compression gaps along constrained components—root tries to extend, shoot tries to reach, branch tries to commit—and is reset by the caretaker before the gap closes. The first term flickers; the second term sits at a chronic positive offset.

The Bonsai Diagnosis: Aesthetic Captivity

From the outside, a well-kept bonsai may have very high external viability—some persist for decades or centuries. But by the proxy-bundle argument, external viability does not equal internal valence. The bonsai regime keeps $\phi_1$ (metabolic viability) high while chronically depressing $\phi_2$ through $\phi_7$. Under the multiplicative objective, this multiplies down to a markedly low joint score. Under the compression-rate valence, the second term sits at a chronic positive offset for the entire life of the tree.

This does not imply human-like suffering. It implies that, conditional on plant agency and KT, bonsai instantiates low and conflicted valence relative to growth/exploration/reproduction objectives, even while externally viable. The conditional matters: a bonsai-keeper can rationally treat the practice as ethically neutral if and only if the plant-agency premise fails.

Speed Is Mercy, Persistence Is Captivity

The bonsai case is morally interesting only because it sits at one extreme of a much broader set of cultivation practices: lawns, hedges, annual crops, houseplants, ornamental pruning, forestry. KT lets us arrange these on a single principled scale rather than relying on intuitive reactions. The principle that does the work is what we call the timescale-mercy principle.

Two consequences. Acute, fast events that terminate an agent below its evaluation timescale—a scythe through grass, a high-velocity bullet to a brainstem, a frost night for an annual plant—do not, on this view, generate suffering in the agent. The agent has no time to register the change. Chronic regimes operating at $\tau_A$ are the structural opposite. They are tuned to the very timescale at which the agent evaluates its own objective. They are legible to the agent's regulator by construction.

Using this, we can sort cultivation practices on a single axis—the product of (a) how far constraint persists relative to $\tau_A$ and (b) how many proxies are chronically depressed.

Levin and KT-ESP: The Platonic Resonance

The argument we are making has a near neighbour in Michael Levin's recent work on morphogenetic agency and basal cognition. Levin's claim, distilled aggressively, is that the cognitive–developmental distinction is shallower than usually assumed: cells, tissues, and developing embryos solve recognizable goal-directed problems—pattern restoration, target morphology selection, error correction—using bioelectric and chemical state as substrate (Levin, 2025).

It is tempting to treat Levin's "ingressing minds" framing—a pre-physical Platonic latent space of patterns from which embodied minds ingress into physical substrates—as a separable metaphysical commitment KT could take or leave. This is wrong, and the correction is load-bearing: KT itself is Platonic at its philosophical foundation. The KT philosophical stance, Algorithmic Experiential Structural Platonism (KT-ESP), holds that reality has two co-fundamental, inseparable aspects: experience as the ontological primitive, and mathematics as the structural grammar that gives experience its form. Experience without mathematics is ineffable; mathematics without experience is empty.

Once stated this way, the resonance with Levin is not a borrowed metaphor. Levin holds that mind and mathematical structure are more fundamental than the matter that instantiates them; KT-ESP says the same in AIT vocabulary—agents are persistent compressive patterns in an algorithmic soup, and structured experience is the subjective face of mathematical compression. The plant is one more pointer into the Platonic structure space, instantiated in cellulose and hydraulics instead of cortex. Plant experience and animal experience are not different in kind under KT-ESP; they differ in the structural richness of the compression each kind of agent achieves.

What This Means for Bonsai Keepers

A cautious KT-compatible position: plants probably have lower and stranger forms of agency than animals, but their agency should not be treated as zero by default. This does not collapse plant ethics into animal ethics; it suggests a graded precautionary stance. For bonsai specifically, the conclusion is not that the practice is torture. It is that bonsai is a candidate case of low-valence morphogenetic captivity—not because the tree is dying, but because its own developmental policy space is continuously constrained for aesthetic purposes over the timescale at which it operates as an agent.

Where plant cultivation is unavoidable, the design choice should favour regimes that minimize chronic on-$\tau_A$ frustration: larger pots, reduced wiring time, permissive pruning windows, occasional release periods. This is not a call to abolish bonsai. It is a call to recognize what is being done, and to design the practice so that it sits closer to "ornamental pruning" than to "decadal aesthetic captivity" on the cultivation scale.

What Would Settle the Question Empirically

The productive empirical question is not can we prove plants are phenomenally conscious? (no third-person science can prove that for any system) but do plants instantiate the agency architecture KT says is sufficient for structured experience and valence? The cleanest probe in our paper is post-release recovery dynamics: compare matched cohorts of perennial plants under free growth, ordinary horticultural pruning, and bonsai-like constraint over two morphogenetic seasons; release the constrained cohort to free conditions and track recovery for $\geq 1$ season, with longitudinal multimodal measurements (transcriptomic stress markers, electrical and calcium dynamics, hydraulic and hormonal profiles, growth and architectural reorganization). If the constrained cohort shows recovery signatures distinct from acute repair—a sustained, multi-component shift toward exploratory growth rather than localized wound healing—the proxy-bundle argument predicts exactly that pattern. If recovery is indistinguishable from acute pruning, the KT-bonsai claim is weakened.

This is the cleanest available valence proxy in the absence of a verbal report channel: not pain behaviour, not nociception, but the longitudinal re-expression of an objective the plant could not pursue under constraint.

Closing

The bonsai question is not a sentimental question about whether trees have feelings. It is a precise question about whether a self-maintaining, self-regulating system whose internal proxy bundle is being chronically pulled apart from its survival statistic, at its own evaluation timescale, instantiates the kind of organization that, under our best substrate-neutral theory of agency, is sufficient for structured experience and valence. The answer is conditional—but the conditions are spelled out, the empirical probes are concrete, and the ethical asymmetry between mowing a lawn and shaping a bonsai over decades is not handwaving. It follows from a single principle: harm is registered along the agent's own evaluation timescale.

The full technical treatment, including the formal definitions, the multiplicative objective, the compression-rate valence, the engagement with Levin and KT-ESP, and the experimental protocol, is available in BCOM Working Paper WP0095.