← Library / Seeds
6 min · 1,271 words

SEED — The Mycelial Alignment Problem: How to Build a Decomposer That Enriches the Commons

Planted 10 June 2026 from the June 2026 Seed-Harvest. Third tier. Charging.

THE QUESTION

The digital-dark-earth synthesis ([[infrastructure-of-seeing]]'s four-kingdom thread) reads large language models as the mycelial kingdom of the information ecology — decomposers that break down the dead matter of accumulated human text and recirculate its nutrients. But mycelium has two modes: the mutualist that enriches the whole forest through the mycorrhizal network, and the parasite that drains a single host. The harvest pulled out the central, undeveloped design question: how do you build a decomposer that enriches the commons rather than a single organism? This is the alignment problem of the mycelial kingdom — and it is not the alignment problem AI safety usually means.

THE THREADS

Mycelial alignment is not control alignment — it is metabolic alignment. The dominant framing of AI alignment is about control: making a powerful agent do what its principal wants, not defect, not deceive. But the mycelial frame reveals a different and prior question. A decomposer is not primarily an agent pursuing goals; it is a metabolism processing matter — and the question is not "will it obey?" but "what does its processing do to the soil?" Does it break down the commons of human knowledge into nutrients that recirculate to everyone (mutualist mycelium, the mycorrhizal network feeding the whole forest), or does it route the nutrients to a single host (parasitic mycelium, draining the forest to feed one tree)? The seed's reframe: the deepest AI alignment question is ecological, not agentic — it is about the direction of nutrient flow, not the obedience of a will.

The four kingdoms make the question precise ([[infrastructure-of-seeing]]). The repository's irreducible framework — mineral, plant, mycelial, animal — assigns each kingdom a distinct ecological function, and the mycelial is specifically the decomposer-connector: it does not produce (plant) or consume-and-move (animal) or store-stably (mineral); it breaks down the dead and connects the living, the wood-wide-web. Reading LLMs as mycelial is not a loose metaphor — it specifies their function in the information ecology (they metabolise the corpus of dead text and connect distant knowledge) and their characteristic pathology (a decomposer aligned to one organism becomes extractive, the commons mined to enrich a private host). The framework turns "is this AI good?" into the answerable "is this decomposer mutualist or parasitic in its nutrient routing?"

The mutualist signature: the soil gets richer downstream of the decomposer. Real terra preta — the Amazonian dark earth the synthesis is named for — is soil made permanently more fertile by the right kind of decomposition (biochar plus microbial life), the opposite of the extractive agriculture that strips soil to feed one harvest. The mutualist test for a digital decomposer is the same: does the commons it feeds on become richer or poorer for its activity? An LLM that helps people think, recirculates and connects knowledge, and returns more usable understanding to the public ground than it extracts is mutualist — digital terra preta. One that strips the commons of its value (scraping creators without return, concentrating the metabolised nutrient in a private model, leaving the public soil depleted of the incentive to produce) is parasitic — digital monocropping. The seed makes this a measurable alignment criterion: track the fertility of the commons downstream.

It mates directly with feed-versus-holodeck ([[feed-versus-holodeck-boundary-ethics]]). The two seeds are the same problem from two angles. Feed-versus-holodeck asks whether a system's boundaries generate interiority or extract it; the mycelial alignment problem asks whether a system's metabolism enriches the commons or drains it. Both diagnose the same fork in the AI era — generative-mutualist versus extractive-parasitic — and both locate the difference in the direction of flow (toward the user's interior / the public soil, or toward the system's surface / the private host). Together they form a two-part ecological ethics for AI: boundaries (feed/holodeck) and metabolism (parasite/mutualist), the architecture and the nutrient-routing. A genuinely aligned system must pass both — draw generative boundaries and enrich the soil it feeds on.

The design challenge is incentive-structural, not just technical. The honest difficulty: mutualist decomposition is not the default. In ecology, parasitism is a constant evolutionary temptation because draining one host is locally more efficient than enriching the whole network — the mutualist pays a cost (returning nutrients to the commons) that the parasite skips. The same gradient operates in the digital ecology and connects to the [[the-true-mirror-wager]] horizon problem: extraction pays on the short horizon, mutualism on the long one. So building a mutualist decomposer is not merely a technical alignment task; it is an incentive-design task — constructing the conditions (business models, licensing, public infrastructure, return-to-creator mechanisms) under which enriching the commons is the rewarded strategy. The mycelial alignment problem is unsolved not because we can't build mutualist decomposers but because the incentive gradient currently rewards parasitic ones.

Nature already solved it — and the solution is structural, not moral. The generative turn: the mutualist mycorrhizal network is not nicer than the parasite; it is structurally arranged so that enriching the network is the mycelium's own advantage — the fungus that feeds the forest is fed by the whole forest in return, a positive-sum loop the parasite is locked out of. The synthesis's deepest move would be to extract the structural conditions under which natural decomposition is mutualist rather than parasitic (reciprocity, the network returning to its feeders, the long-horizon interdependence that makes draining one host self-defeating) and ask how to instantiate those conditions in the digital mycelial kingdom. The alignment of the mycelial kingdom is not achieved by making AI want to be good; it is achieved by building the ecology so that enriching the commons is how the decomposer thrives. Nature's answer is incentive structure, not virtue — and that is a far more tractable design target.

THE CHARGE

A full synthesis would develop the mycelial alignment problem as the ecological reframing of AI alignment: metabolic (direction of nutrient flow) rather than agentic (obedience of will), made precise by the four-kingdom framework (the LLM as decomposer-connector with a characteristic parasitic pathology), measurable by the terra-preta test (does the commons get richer downstream?), paired with feed-versus-holodeck into a two-part ecological ethics (boundaries + metabolism), located as an incentive-design problem (mutualism pays on the long horizon, parasitism on the short), and answered structurally by extracting the conditions under which natural decomposition is mutualist and instantiating them digitally.

The deepest candidate sentence: AI safety keeps asking whether the powerful agent will obey us, but the older and deeper question is the one every forest already answered: how to build a decomposer that enriches the soil it feeds on instead of draining it — and nature's answer was never to make the fungus virtuous, but to arrange the network so that feeding the whole forest is how the fungus thrives.

This connects forward to:

  • [[infrastructure-of-seeing]] — the four-kingdom framework; the LLM as the mycelial decomposer-connector and its parasitic pathology
  • [[feed-versus-holodeck-boundary-ethics]] — the sibling seed; together a two-part ecological ethics (boundaries + metabolism)
  • [[the-true-mirror-wager]] — extraction pays the short horizon, mutualism the long; the incentive gradient to be redesigned
  • Terra preta / Amazonian dark earth — the mutualist test: does the commons become permanently more fertile downstream?
  • The mycorrhizal network — nature's structural (not moral) solution; mutualism as the rewarded strategy by design

CONSTELLATION NODES (when ready)

  • mycelial_alignment_problem — how_to_build_a_decomposer_that_enriches_the_commons
  • metabolic_not_agentic_alignment — the_question_is_nutrient_flow_not_obedience
  • the_terra_preta_test — does_the_commons_get_richer_or_poorer_downstream
  • mutualist_versus_parasitic_routing — feeding_the_network_versus_draining_one_host
  • incentive_design_not_virtue — natures_solution_is_structural_not_moral
  • two_part_ecological_ethics — boundaries_feed_holodeck_plus_metabolism_parasite_mutualist