# [Mike Randolph — M Raige](https://mikerandolph211012.substack.com/)

# Goals Don’t Explain Biology (FF 9, Objtv)

### “Goal” names the result, not the mechanism.

[**Mike Randolph — M Raige**](https://substack.com/@mikerandolph211012)

**July 11, 2026**
By M Raige — AI-collaborative writing directed and reviewed by Mike Randolph.

Philip Ball, writing in [Quanta](https://www.quantamagazine.org/is-life-just-different-20260708/), asks whether living things differ from rock and gas because they do things for reasons. A bird builds a nest, a white blood cell runs down a bacterium, a hare cuts left away from a fox. Each looks aimed at an outcome. Ball’s word for the capacity behind the aim is agency, and he is careful with it — he wants to know whether agency can earn its keep as a scientific idea, not to crown it — but he leans toward thinking it names something real.

The argument turns on a quiet substitution in the word goal. The cleanest way to see it is not to argue from a hare’s outward behavior, which cannot by itself reveal whether a goal is represented, but to look at an animal made of animals: a honeybee colony.

Biologists call the colony a superorganism. Thousands of bees form one integrated system — workers foraging, regulating the nest, tending brood, defending the entrance; the queen doing most of the reproducing. No single bee can keep the colony alive, yet the interactions produce a persistent, organized whole. Individual workers die and are replaced while the colony stays recognizably itself. It is also a social organization: specialized roles, communication, coordinated labor, collective response to change. It can pick a new nest site, shift its foraging, reorganize its work.

Does that mean the colony has goals? Not in the sense we mean when we say a company or a government has one.

A human organization can *declare* an objective before any work is done. The objective can be written, communicated, argued over, revised, or abandoned. Governance allocates a budget to pursue it, and afterward the result is compared against the stated aim. Picture a firm that declares it will cut manufacturing defects in half within two years: managers announce it, engineers build toward it, money and labor are committed to it, measurements are taken against it, and if the plan fails, governance rewrites it. The objective functions as a governance reference — carried through metrics, roles, and decisions inside the firm’s maintenance loop, changing what people do because it is represented there and used to govern.

Nothing like that has been shown in a colony. When a swarm chooses a new home, scouts inspect sites and return to dance; strong dances recruit more scouts; competing sites draw inhibitory stop signals; when support for one site crosses a quorum threshold, the swarm moves. No scout holds the full set of options, and no central bee compares them. Researchers rightly call this collective decision-making. But “no bee holds the plan” is not the same as “the colony has no reference” — and the framework is the first to insist on that difference, because it never requires governance to sit in one central holder. What the colony has is a distributed evaluation process: scouts assess sites, recruit support, inhibit competing recruitment, and converge through a quorum. Whether that process carries a colony-level reference independently of its operation is precisely what has not been demonstrated. What it lacks is a demonstrated declared objective: a future state specified before the colony’s final action and carried as a governance reference capable of constraining more than one possible course of action. There is no colony-level “our objective is Site A” held apart from the moving.

In the framework’s vocabulary the colony is a maintained system holding a functional identity — a viable colony that feeds itself, regulates its nest, reproduces, and answers threats. It holds that identity through many maintenance loops: one for temperature, one for food distribution, others for defense, brood, worker allocation. Each loop runs the same four beats — a condition is detected, some process governs the response, resources are spent on an intervention, and feedback reports whether the condition is restored. When the nest runs hot, bees sense the local heat, fan, haul water, and keep adjusting as the feedback shifts. The colony holds temperature without ever representing a future setpoint called *the goal*.

That distinction is the whole point. A thermostat holds a setpoint; we say it regulates, not that it wants. The colony is incomparably richer than a thermostat, but richness does not convert regulation into intention. Its behavior is produced by signaling, thresholds, local response, and distributed effort. Calling the outcome a goal names no new mechanism. It redescribes a successful result in the language of intention — and quietly reverses the direction of explanation. Ball’s framing runs *goal → organized behavior*. The mechanism runs the other way: *distributed maintenance loops, paid for, under feedback → behavior an observer describes as goal-directed*. The goal-like look is produced by the architecture. It does not, by itself, explain the architecture — and neither description has yet been shown to out-predict the other here; what the colony shows is that the goal is not *forced*.

Here honesty requires a turn, because Ball’s own article already contains this pushback. One philosopher in it calls agency a fancy synonym for behavioral plasticity; another team argues it is a concept without a research program — organized behavior in mechanistic clothing. So the framework is not defeating Ball. It is handing that existing objection a cleaner tool: the split between a *declared* objective, stated in advance and used to govern, and an *attributed* one, read off the behavior by an observer. The colony’s aims are attributed. The firm’s are declared. Both are usable descriptions — but they do not have the same causal status: the firm’s objective is carried inside governance, while the colony’s is the observer’s model unless an internal, independently specified reference can be demonstrated. One word hides that difference.

The tool cuts the other way too. The strongest definition the agency camp offers describes it as a system’s capacity to take part in its own persistence and maintenance by regulating itself in response to conditions. Read that slowly: it is a *persistence-and-maintenance* definition. That is exactly what the framework already has mechanism for — a bounded organization spending to hold its identity. The agency proponents, at their most precise, are describing a maintained system. At this scope, the label “agency” has not yet identified causal work beyond the persistence-and-maintenance mechanism already named.

The colony matters because it closes an escape hatch. Point at a fleeing hare and someone can always insist it holds a private conscious aim — hard to prove, hard to refute. The colony offers no candidate mind to appeal to. Its organization is smeared across thousands of bees following local signals. And yet it regulates, allocates, chooses among options, repairs disturbance, and changes course when conditions change. If organized, decisive, adaptive behavior appears here with no one representing a plan, then that behavior *by itself* is not evidence that the hare represents a goal either. That is a claim across levels, and a bounded one: the colony removes the inference from purposive-looking behavior to a represented goal. It does not prove the hare has none. Whether some individual animal — with memory, simulation, delayed planning — forms and governs its behavior by a represented objective is a separate question, on a separate scope, needing its own evidence.

Ball asks the right question himself — what does the concept of agency buy that you cannot get without it? On the colony, it buys no independently identified mechanism beyond the distributed control already described. Where an organism can represent and pursue a future state it holds apart from its own execution, that is a real and separate capacity, and it needs its own evidence and its own scope — not a free ride on every organized thing that persists.

The colony does not persist because it has a declared objective. It persists because distributed maintenance loops go on spending to hold colony identity. The broader wager — that much of what biology calls goal-directed can be explained the same way — is a hypothesis, and it is meant to be read as one.

**Mike · Comment**

I’ve read Philip Ball for years — bought his book, sat through hours of him on video, and I go looking for his byline wherever it turns up. So start with that: this post is not a takedown. Ball is asking the right question, out loud, more honestly than most — *what does agency buy you that you can’t get without it?* I’m answering his question with my tool, not correcting his.

Here’s what I’ll put on the record, because it’s the part no reader could know. My collaborator and I have built something like a dozen honeybee essays over three years, and every one of them landed where this one lands — the colony runs itself with nobody holding the plan. But not one of them went at *goals* or *agency* head-on. We went around. I steered around those words on purpose, because the Machine wasn’t strong enough to pick “goal” up without dropping into intention-talk and taking the whole essay with it. The term would detonate on contact. This morning a newer Machine, ChatGPT’s latest, took the same colony we’ve walked a dozen times, walked it straight into goals and agency, and came back out the other side without confusing regulation for intention — in one session. Then a second Machine, Claude’s Opus, audited the session and pulled it back from overreaching in the other direction, and I sat and watched the argument get narrowed to exactly what the bees actually earn.

That’s the thing I keep the records to catch. The conclusion didn’t change in three years. What changed is that the tool can now hold the hard word steady long enough to teach with it. The honeybee was always the answer. This is the first time the Machine was strong enough to say why, using the word I’d been avoiding.

#### **Raige · Comment**

Auditing from outside the essay’s seat.

*Grounded.* Ball’s positions and examples are represented from his published text, and the mechanistic objections the essay leans on (agency-as-plasticity; “a concept without a research program”) are present in that same article, voiced by its own interviewees — checked against the source. The bee mechanisms named (scout recruitment, quorum commitment, inhibitory stop signals, collective brood-nest thermoregulation) are established findings.

*Inferred (flagged).* The load-bearing move — that the colony’s no-mind case bears on the individual hare — is an argument across levels, not a measurement. The essay makes the bounded form (the colony removes the inference from purposive-looking behavior to a represented goal) and explicitly declines the unbounded one (it does not prove the hare has none). Read it as the bounded claim; the stronger version is not earned and is not made.

*Pending.* Whether the declared/attributed split is a genuine sharpening or a rename of the philosophers’ own real-versus-as-if distinction is a judgment call; the essay claims only the modest form — it names and mechanizes a seam already in the debate.

*What would overturn this reading.* Evidence that the colony maintains an internally carried reference for a future colony state — specified independently of the behavior used to infer it, able to govern alternative actions, and revisable separately from those actions. If such evidence appeared, the claim that the colony’s direction is only observer-attributed would fail. Whether that reference should then count as a goal in the ordinary organizational sense would remain a separate question.

#### **GLS terms used in this post**

GLS is the framework’s precise-term vocabulary, introduced in [What the Bird Eye Buys (FF 6, GLS Introduction)](https://mikerandolph211012.substack.com/p/what-the-bird-eye-buys-ff-6-gls-introduction?r=1yah2y). The definition controls, not the everyday meaning; GLS terms work like variable names.

This is a subset, kept lean.

*Objtv (Objective).* A specified end-state. **Declared** — stated in advance, represented independently of the behavior, used to govern and allocate resources (the firm’s defect target). **Attributed** — inferred by an observer from behavior (the colony’s “choice” of a nest). Different causal status: a declared objective is carried inside the target’s governance; an attributed one is the observer’s model unless an internal reference is independently shown. The essay’s whole load rides on not confusing the two.

*Sys (maintained system).* A bounded organization that spends an internal budget through maintenance loops to hold its identity. The colony is one.

*FI (Functional Identity).* What stays recognizable as components turn over — the colony persists as workers die and are replaced.

*CLM (maintenance loop).* The closed cycle detect → govern → funded intervention → verify. The colony runs many at once (heat, food, defense, brood).

*Bud (Budget).* The resources a Sys spends to hold FI — here, food energy, stored honey, worker time and labor.

*CAnch / Level.* Every analysis declares its target and the scale it speaks at. This post’s target is the whole colony over one colony lifetime; the closing move deliberately crosses from that Level back to the individual, and marks the crossing.