Agency ⊥ intelligence · a scored, testable ladder
The Agency Ladder
Does a machine spend its own energy to hold its own boundary? That has a measurable answer, where "is it intelligent?" is a property the evaluator assigns. The system-intrinsic question is: does it spend its own energy to hold its own boundary, on its own power? Here is a five-rung ladder for that, and where today's systems actually sit. Most score zero.
Charlot Lab, Institute for Physical AI @ JBI
1 · What is the right question to ask about autonomy?
The field is stuck measuring intelligence, and it is stuck for a structural reason: intelligence is assigned by an evaluator inside a task frame, not possessed by the system. You cannot benchmark a property that is attributed rather than intrinsic; chase it and the epistemics stay broken, which is exactly what has happened.
Agency admits a system-intrinsic test. The discriminator that does the most work in one line: if removing the human changes nothing about whether the system persists, there is no agency there. Almost everything currently labelled "agentic" fails this immediately: an LLM in a while-loop with tool calls has no essential variable of its own, no energy store of its own, an entirely exogenous setpoint, and it does not persist between invocations except in a buffer someone else rents. That is delegation, not agency. And the boundary that counts is the one that fails when the power stops: for anything running in a datacenter, that boundary belongs to the datacenter. Which is why, for a physical agent, running on the robot's own power, on-device, is not an efficiency nicety but a necessary condition for the boundary to be the robot's own.
2 · What are the rungs of the ladder?
Five rungs, drawn from the sharpest formalizations anyone has produced and mostly forgotten (Ashby's ultrastability1, Di Paolo's adaptivity2, the operational-closure half of autopoiesis3):
| Rung | Test | Nothing above it is free |
|---|---|---|
| 0 | exogenous setpoint, exogenous power | delegation: a tool driven on someone else's goal and budget |
| 1 | an essential variable it can lose, on its own store | it can die, and it acts to not (the first real stake) |
| 2 | senses the tendency toward the boundary; acts before it is crossed | a cliff becomes a gradient: Di Paolo's adaptivity, the difference between a system that dies and one that behaves |
| 3 | second-order reorganization: replaces its controller when the current one fails | Ashby ultrastability; most deployed RL tunes parameters inside a fixed architecture; the architecture-search and meta-learning lines are the closest published approach to the second loop this review located, and they reorganize offline rather than under threat. That is the first loop only |
| 4 | self-production: makes the components that make it | operational closure; a program does not produce its own substrate |
Show the computation
rung = highest r such that tests 1..r ALL pass (gated, not a score) 1 essential variable it can lose, on its own store → a real stake 2 senses the tendency, acts before the boundary → adaptivity (Di Paolo) 3 second-order reorganization: replaces the controller → ultrastability (Ashby) 4 self-production of its own components → operational closure skip a lower rung and the ladder does not advance: the gate is the point
Scored against real systems, the ladder places them as follows; each row states the test that fails and is open to challenge on the criterion rather than the arithmetic:
| System | Rung | Why |
|---|---|---|
| Frontier LLM + full tool access | 0 | exogenous setpoint + power; nothing of its own persists between invocations. What binds is energy provenance, not capability: the constraint is that inference at current joules-per-token cannot run on a body's own store. The measured quantity that moves this row is joules per decision on-device against the on-board budget - the ratio of joules per decision on-device to the on-board store. This review did not locate a measured pair for a body of this class, and the row moves when that ratio falls below one |
| Industrial robot arm | 0 | someone else's setpoint on someone else's power budget |
| Thermostat | 0 | the setpoint is exogenous and it does not pay to hold it |
| Roomba | 1 | it seeks its own charger, an essential variable (charge) it can lose |
| EFA certificate-gated controller, on-device | 2 → 3 | its own energy boundary; senses the tendency and acts before the crossing; the fallback switch is one step of ultrastability, argued below, and provable |
| Any living cell | 4 | it produces the components that produce it: the criterion this review did not locate any synthetic system meeting; we checked the protocell, artificial-chemistry and self-replicating-machine literature |
3 · Where does the certificate land, and why is it provable?
An on-device controller gated by an energy certificate4 holds an essential variable it can lose: the energy $V$ must stay in the safe set $S=\{V\le c\}$ or the closed loop diverges (rung 1). It does not wait for the crossing: the discrete barrier admits an action only if its actual next state stays in $S$, so it acts on the tendency, before the boundary. A cliff turned into a gradient (rung 2). And it switches controllers under threat: when the task policy would leave $S$, the certificate vetoes it for an energy-shaping fallback. A first, real step of second-order reorganization (reaching rung 3). Its per-gate-step energy is obtained as P x t with the idle floor subtracted, measured on the device that runs it against that robot's own store. This review has not yet published that measurement, and the share of the robot's own budget it represents is the quantity that makes the boundary its own - which is what makes the boundary its own at all.
Unlike every other row, this placement is not a claim but is measured, on a metric the field does not have: J/VT, joules per viability-second held under disturbance. Under an adversary trying to drive the body out of $S$, a puppet that commits the policy holds viability 26% of the time at 347 W; the certificate-gated agent holds 100% at 57 W. Cheaper and safer, not a trade. On a simulated SO-101 body the agent runs at 0.25 W with zero crashes against a single-vector policy's 7.47 W and 36 crashes (all figures in this paper are simulation; no physical SO-101 run is reported here). And it earns rung 2 in the hard case: given only intermittent, noisy, position-only sensing, an observability-aware version sizes its margin to a Kalman estimator's own uncertainty. When an occlusion hits and it can no longer see, that uncertainty grows on its own, the margin widens, and the body slows down rather than trusting a state it cannot estimate. Safe because it knows it cannot see.
So the row reads: rung 2, one step into 3. Not rung 3 in full: the reorganization is a switch to a fixed fallback, not a learned new controller. Not rung 4: nothing synthetic produces its own substrate. But it is higher than any shipped system this review was able to place on the ladder, and it is the only row here whose placement rests on a proof rather than an argument. We did not audit shipped systems we could not inspect, including the deployed peer named below.
Where does this stand?
The binding constraint is the evaluator, not the system. Intelligence is an attributional predicate: it lives in the evaluator's task frame, so a benchmark for it scores the frame as much as the thing being scored. This review did not locate a published intelligence benchmark whose score is invariant to that frame, which is why every placement on this ladder is argued from a measured result or from an explicitly named open problem rather than from a ranking.
What would move it. A scoring quantity that survives a change of frame. The candidate used here is joules per viability-second, how much energy a system spends to remain viable, for the reason that viability is defined by the system's own persistence rather than by the observer's choice of task. That is a proposal, not a settled answer, and the measurement that would test it is the same system scored under two genuinely different task frames: if the joules-per-viability- second ordering holds where a task-score ordering flips, the quantity has earned its place.
References & lineage
- W. Ross Ashby, Design for a Brain, 1952: ultrastability: essential variables plus a second-order loop that reorganizes the controller when they leave bounds. Requisite variety gives the hard lower bound on required sensing.
- Ezequiel Di Paolo, Autopoiesis, adaptivity, teleology, agency, 2005: adaptivity as the necessary repair to bare closure: sensing the tendency toward the boundary and acting before it is crossed.
- Maturana & Varela, autopoiesis (the operational-closure half only); Still, Sivak, Bell & Crooks, Thermodynamics of Prediction, PRL 2012: the dissipation floor under a modeling agent.
- Companion: One Energy, Both Roles and Energy Is the Certificate, Charlot Lab: the certificate and the J/VT scoreboard the rung-2 placement is argued from.