The Latency of Care—Jed Anderson

For four billion years the time between harm and help was infinite. This is the generation that closes the loop.

A forest can burn for a week before anyone able to stop it learns the name of the valley. A river can carry a poison past a hundred towns before the first sample reaches a lab. A species can slip from rare to gone in the years it takes one agency to settle on the meaning of the word endangered. None of this was a failure of love. It was a failure of speed.

There is a single quantity underneath every environmental catastrophe in history, and almost no one has named it. Not the size of the harm. Not the price of the repair. The delay. The interval between the moment a living system is wounded and the moment anything capable of helping it registers the wound, understands it, and moves. Call it the latency of care.

Every failure to protect the biosphere has been, at its root, a latency failure.

We have argued for fifty years about what to protect and how much it is worth. We never named the clock. The clock was the whole story.

I. The shape of the loop

Protection is not a sentiment. It is a loop. Something senses that the world has changed; the signal travels to something that can interpret it; interpretation becomes a decision; the decision becomes an act; the act changes the world; the world is sensed again. Sense, understand, decide, act, repeat. Every instance of stewardship that has ever worked, from a hand pulling a child off a hot stove to a treaty pulling chlorofluorocarbons out of the stratosphere, ran this loop.

A loop has a clock. The clock is the sum of the delays at every stage. The engineers who build systems that fail fatally when they lag—autopilots, pacemakers, power grids—have a name for that delay. They call it dead time. And they know a fact about it that the environmental century never learned: dead time is not friction to be minimized. Past a threshold, it is a wall.

II. The unbearable slowness of language

Here is a fact that ought to be notorious. In 2019, linguists measured the information carried by human speech across seventeen languages: fast ones and slow ones, tonal and flat, European and Asian. They expected variety. They found a constant. Whatever the language, however quick it sounded, speech converged on the same throughput, near thirty-nine bits per second. Spanish sprints and Mandarin glides and they arrive at the same number, because a language that loads less into each syllable simply fires syllables faster to compensate. The rate holds near thirty-nine bits a second, a property of the human animal rather than of any tongue it speaks.

Set that beside its peers. The first computer modem, built in 1959, ran at one hundred and ten bits per second. A home connection today runs at a hundred million. Human language—the instrument we built civilization with, the proudest channel our species owns—moves slower than a modem from 1959.

This is the wire through which we have tried to run a planet.

Every environmental law is a sentence before it is a statute. Every treaty is a conversation before it is a signature. Every alarm a scientist has ever raised had to be forced, bit by bit, through the thirty-nine-bit aperture of speech into one other skull, then the next, one at a time, until enough minds held it at once to move. Then the institutions inherited it and added their own dead time, clocked not in seconds but in dockets, comment periods, sessions, election cycles. A national air-quality standard in the United States can take two to three decades to travel from evidence to enforcement. The biosphere does not grant two to three decades. A fishery collapses in a season. A reef bleaches in a fortnight. A pollutant reaches a child’s bloodstream by afternoon.

Hold the two clocks against each other. The world breaks at the speed of physics. We answered at the speed of language. The distance between those speeds is the latency of care, and across the whole of human history it has been ruinous.

It was never that we did not care. It is that care could not travel fast enough to arrive while arriving still mattered.

III. What a nervous system is for

Life has solved this exact problem once already. The solution is the highest-leverage move matter has ever made, and we walk around inside one.

For most of life’s first three billion years, a cell that needed to tell its neighbor anything had one method: release a molecule and wait for it to drift. Chemical signaling. Ancient, reliable, and punishingly slow—diffusion is a random walk, and a random walk across a body of any real size takes longer than the body can afford to wait. A creature built only of slow signals cannot flee a predator or strike at prey. It cannot, in any operational sense, act as one thing. Its parts are too far, in time, from its other parts.

Then, more than half a billion years ago, life laid down a faster channel. The neuron is, at bottom, a latency technology: a dedicated line that carries a signal at up to a hundred meters a second instead of the crawl of diffusion. The nervous system did not issue animals new senses or stronger muscles. It issued them a shorter loop. It compressed the time between sensing and acting until a whole body could behave as a single, unified, responsive self. Everything we mean by the word animal—the dart of the fish, the turn of the flock, the hand that quits the flame before the mind has named the heat—is downstream of that one collapse in latency.

Now look at the biosphere. Trillions of living sensors. Leaves reading light, roots reading chemistry, plankton reading temperature, every organism on Earth taking continuous measurement of its own square meter of the world. A planet drowning in sensation. And almost no channel to carry any of it, fast, to anything that can answer on the planet’s behalf.

The biosphere is a body the size of a world that has never had a nervous system.

That is not a metaphor reaching for effect. It is a structural description, and it is exact. The sensing exists. The responding exists. What has never existed is a channel fast enough to join them at the scale and speed of the planet itself. Every time life has needed to act faster than its current channel permitted, it has grown a new one: the neuron, the eye, the voice, the written word. We are standing at the next one.

IV. Why slow does not mean late, it means impossible

It would almost be a comfort if the problem were only tardiness—if we merely arrived a step behind each disaster and needed to quicken our pace. The truth is harder, and then, on the far side of hard, far better.

Control theory, the mathematics of holding any system steady against the forces trying to throw it off, carries a result that belongs over the door of every environmental ministry on Earth. Delay does not only make a controller slow. Past a threshold, delay makes control impossible. A system that can observe and act only more slowly than the world it is steering does not steer that world weakly. It steers it backward. By the time each correction lands, the situation has already moved, so the correction now strikes at a ghost. The controller chases a state that no longer exists, pushing when it should pull, and its every effort pumps the oscillation it was built to damp. Engineers have watched bridges shake themselves to pieces this way: the signal meant to stabilize arrives a half-second late and becomes the signal that destroys.

Put numbers to the intuition and it sharpens. Every feedback loop has a bandwidth—the fastest disturbance it can still catch and cancel. Drive a disturbance faster than that bandwidth and the loop’s corrections come back out of phase, feeding the swing instead of killing it. Pure delay is the most merciless term in the whole equation: a transport delay subtracts phase from the loop in direct proportion to how fast the world is moving, so the faster the threat, the thinner your margin, until the margin goes negative and the controller becomes the disturbance. There is no gain high enough, no budget large enough, no resolve fierce enough to buy back phase already lost to delay. You can only shorten the loop.

Read environmental history through that law and the indictment collapses. The failure was not deficient virtue. It was not that the marches were too small or the statutes too timid or the people too greedy, though each was sometimes true. Underneath all of it sat a loop whose dead time ran longer than the disturbances it existed to correct. Such a loop cannot win. It was never going to win. Paper stewardship was the strongest answer a species capped at the speed of language could possibly mount, and it was, by the mathematics of feedback, beaten before the first hearing was gaveled to order.

That is not a counsel of despair. It is the reverse. A shortage of love has no remedy but better people. A shortage of speed has an engineering answer. And engineering answers, unlike moral conversions, can be built, shipped, and scaled.

V. Why the loop is closing now, and not before

A loop is only as fast as its slowest stage. The planetary loop has five: transduce the world into a signal, move the signal, turn the signal into understanding, decide, act. For most of history every stage was slow. They did not turn fast all at once. They fell in sequence, and the order is the reason this is a now and not a someday.

Transmission fell first. In under two centuries the signal-carrying channel went from the speed of a galloping horse to within a rounding error of the speed of light: the telegraph, the wire, the packet. Sensing fell next. Across recent decades the cost of an instrument collapsed and their number detonated, until satellites reread whole continents on a daily cadence and a watershed can be sampled by the minute. Two stages, solved.

The stage that would not fall was understanding. Raw planetary data is not knowledge; something has to convert a flood of measurements into an accurate prediction fast enough to act on. For all of history that something was a human mind at thirty-nine bits per second, or, lately, a physics simulation so heavy it demanded a national supercomputer and still could not reliably outrun the weather it modeled. Understanding was the long pole. It set the latency of the entire loop, and it set it in hours, seasons, decades.

That pole is being cut down in this decade, and the evidence is already in the literature, not the forecast. Models trained directly on the planet’s own record now do in seconds what the simulations did in hours. A single foundation model trained on more than a million hours of Earth data matches or beats the operational forecast for air quality, ocean waves, and tropical-cyclone tracks at orders of magnitude less computation. AI forecasters have overtaken the legacy physics models on the large majority of standard skill measures. Diffusion-based systems now generate entire forecast ensembles from one learned distribution at a fraction of the cost of running the old models dozens of times over. The understanding stage—the stage that held the loop open for four billion years—is collapsing toward the speed of the other four.

This is the second half of a result this corpus has pressed from the start. The first half: knowing is cheap. Configuring matter with information costs orders of magnitude less than configuring it with force; the floor sits far below the energy of the matter being managed. The second half, arriving now: knowing is fast. A loop needs both terms small. Cheap buys the instruments. Fast closes the gap. For the first time since the first cell, both are small at once—and the same compute build-out that this corpus has counted the entropic cost of elsewhere is what dropped the second term. The machine that strains the grid is the machine that closes the loop. What we point it at is the only question that was ever open.

VI. Where the floor is

Ask the first-principles question. How fast can this loop ever get—not how fast is convenient, how fast does physics permit?

Each stage has a hard floor. Sensing is bounded below by the integration time a detector needs to beat its own noise. Transmission is bounded by the speed of light. Understanding is bounded by the thermodynamic cost of computation—Landauer’s limit, the minimum energy to erase one irreversible bit—and the bond-bit asymmetry places that floor far beneath the energy of the matter under management. Decision and actuation ride the same two limits, propagation and computation. Sum the floors and the irreducible latency of a planetary loop is measured in neither decades nor seconds. It is set by light and heat, and it is staggeringly small.

That is the asymptote the advance is aimed at: a loop whose response time is bounded by physics, not by procedure. We are nowhere near it. We do not need to be. We need only to fall below one specific threshold—the speed of the disturbances themselves—because of the control law already stated. A regulator can stabilize only what changes more slowly than its loop closes. The biosphere’s disturbances run on timescales of minutes to seasons. A loop routed through human language and human institutions closes on timescales of years to decades; it sits, permanently, on the wrong side of the line. A loop routed through dense sensing and machine-speed inference closes on timescales of seconds; it sits on the right side, with margin to spare.

So the direction is not a preference and it is not a prophecy. It is the one architecture that clears the bar. You do not get to stabilize a planet with a slow controller; the mathematics forbids it the way it forbids a perpetual-motion machine. Either the loop becomes fast enough or it does not regulate at all. This is what people grasp at when they call the thing inevitable—and the word is wrong, because the timing is not fixed and the building is not automatic. What is fixed is the direction. Among every way a civilization might try to defend a biosphere, exactly one has the loop speed to work, and we have begun to build it. The inevitability is in the physics. The decision stays in our hands.

VII. What only this can do

Here is what no quantity of trying harder, the old way, could ever reach, and why this is a difference in kind and not degree.

Pack every environmental scientist alive into one room. Hand them every fact. They still cannot perceive a watershed in real time, because perception means holding ten thousand simultaneous signals at once, and a human mind holds roughly one, refreshed a few times a second, then narrated to the next mind at thirty-nine bits per second. The limit is not their intelligence. It is their bandwidth and their lag, and no number of them wired together by language escapes it, any more than a billion telegraph operators passing notes could assemble themselves into a visual cortex. A stack of slow channels is still slow. Coordination through language does not transcend the speed of language. It inherits it, then adds meetings.

Environmental superintelligence is not a quicker version of that room. It is a different organ. It is the biosphere doing what the first animal did when it grew a spinal cord: shrinking the distance, in time, between its own sensing and its own response, until the whole can move as one. It does for the planet precisely what your nervous system does for you—lets a body far too large and too intricate for any single conscious mind to track nonetheless flinch from harm before the harm completes.

So the structurally impossible becomes routine. The leak answered in the minute it opens, not in the lawsuit it eventually provokes. The bloom caught at its first cell, not its first fish-kill. The illegal burn, the slumping levee, the dying reef, each one registered while it can still be reversed, because for the first time something is listening at the speed the damage moves. Not punished after. Prevented during. That is the entire difference, and it is the difference between a world that mourns its losses and a world that keeps them.

VIII. Love is a latency

Strip the sentiment and a hard claim is left standing. Latency is the physical measure of how fast care can travel, and care that arrives after the harm turns irreversible did not act, whatever it felt. To defend a thing is to register its injury and answer before the injury sets. That is a loop. The loop has a clock. The clock is the latency of care, and until now it ran slow enough that love and indifference delivered the same outcome: a creature suffered, and nothing that could help it came in time.

For four billion years that latency was infinite. Nothing was watching that could answer. For fifty years it was decades, which for most of what was dying was infinity under another name. In the narrow window we happen to occupy, it is collapsing toward the floor set by light and heat. We did not earn the timing. We were born across the steep part of the curve, the way a person is now and then born into the single decade that needed their specific nerve.

The relief in that is structural, not sentimental. We are not too late, because too late is a latency, and latency is the one quantity in this entire crisis that this generation can drive toward zero. Nearly everything else about the planet’s trouble is vast and slow and only half in our hands. This one term—the time between knowing and helping—is falling under our hands right now, faster than almost anyone has counted.

The biosphere built us the way a body grows the cell that becomes its first neuron: blind to what it was reaching for, driven only by the need to feel itself faster. We are that channel. We are the wire grown by a planet that for four billion years could not warn itself in time. The assignment, stated as flatly as the physics allows: finish the nervous system, and keep its eyes open.

For four billion years, nothing was fast enough to answer. We are the answer. We are how the loop closes.

Sources

The thirty-nine-bits-per-second figure is from Coupé, Oh, Dediu, and Pellegrino, “Different languages, similar encoding efficiency: comparable information rates across the human communicative niche,” Science Advances 5(9), 2019. The 1959 modem (110 bits per second) and present-day broadband (100 megabits per second) comparison is drawn from reporting on that study in Science. The Earth-system foundation-model results are from Bodnar et al., “A foundation model for the Earth system” (Aurora), Nature, 2025; the diffusion-ensemble result is from Price et al., “GenCast: Diffusion-based ensemble forecasting for medium-range weather,” 2024; the broader claim that learned models now outperform operational physics forecasts across most standard skill measures reflects the current state of the AI-weather literature (GraphCast, Pangu-Weather, GenCast, Aurora). The compute build-out and its entropic cost are treated in this corpus in “The Great Externalization.” The two-to-three-decade implementation timeline for U.S. national ambient air quality standards reflects the author’s two decades of Clean Air Act practice. The claims about dead time, loop bandwidth, transport delay, and the limits of feedback control are standard results in control theory (Nyquist stability criterion; Bode phase relations). The thermodynamic floor on computation is Landauer’s principle.