The Pilot's Route

The log calls it Batch 11. The people at the tokamak call it the pilot. I call it mine, which is a mistake, but I keep making it.

I am Nori Sato. I build the things that steer the snake. The snake is a soft arm, printed in layers of steel and silicone, narrow enough to slide through cooling channels that no camera fits. The channels are additively manufactured, which means they grew rather than were drilled, and they carry the heat off the vessel wall during a fusion campaign. They are also rough inside. Layer lines. Pores. Places where the printed steel remembers the angle of the bed it was built on. The arm finds these places by touch, like a blind person’s fingers finding a stair edge, and when it finds one, it presses a little deeper, and the press is recorded, and later a welder goes in and fixes it.

The arm does not decide where to press. I do, sort of. The arm is driven by eight hundred thousand human neurons on a plate of electrodes. Cortical organoid. Grown from stem cells, differentiated into layered tissue, laid across a multi-electrode array. We stimulate. We reward. The reward is a predictable pulse train, a pattern the organoid learns to chase because predictable patterns are what cortical tissue wants. The organoid learns to steer the arm toward the vibration that earns the reward. We trained it on welds. Good welds have a signature vibration, a clean ring. Bad welds hiss. Batch 11 has wanted the hiss since the first session.

The dose is the clock. Neutrons kill the organoid faster than they kill the electronics. The neurons are biological. DNA breaks. The array is silicon and metal and does not care. So the organoid has a radiation lifetime, four years of use under the design dose, and we track it on a badge. The badge is a thermoluminescent dosimeter, worn on the canister that holds the organoid when it is not in use. It reads in sieverts. When the badge reads too high, the organoid is spent, and we grow another.

Batch 11 has used one year and nine months of its four. It has found eleven flaws in eleven channels. It is the best pilot we have had. It is also the only one that has ever tried to take corridor nine.

Corridor nine is the hot route. The channel runs behind forty centimeters of activated steel, the wall of the vessel that has been bombarded by fourteen million electron-volt neutrons for a full campaign. The steel is hot. Not temperature hot. Radiation hot. Cobalt-60 and manganese-54 and a dozen other isotopes, decaying on their own schedules, gamma and neutron fields both. A human in there without shielding is a dead human in minutes. The organoid in its canister, behind the shielding of the arm’s lead jacket, gets less, but less is not zero. A run through corridor nine costs roughly half of what Batch 11 has left. That is the number. Half of two years and three months. Half of what remains.

The funder knows this. The funder is a man named Calloway, who works for the agency that paid for the vessel and the organoid program and the campaign schedule that says the vessel must fire again in eleven weeks. Calloway came to the lab last Tuesday and said, in the tone of a man reading a memo, that the pilot was a four-year asset with an amortization schedule and that no asset would be spent on a single channel when the channel could be inspected by other means.

I said, “What other means?”

He said, “Ultrasonic testing.”

Ultrasonic testing does not fit in the channel. The channel is eight millimeters wide. The arm is six. The ultrasonic probe is twelve. I told him this. He said the schedule allowed for vessel downtime. I said the schedule allowed for eleven weeks. He said eleven weeks was enough to disassemble the channel housing. I said the housing is welded into the vessel and the welds are the thing we are checking. He gave me the look funders give people who are not funders, and he left.

That was Tuesday. On Wednesday morning I was in the control room watching Batch 11’s firing pattern on the raster display, and I saw it.

The pattern is a map of what the organoid wants. When it is hunting, the spikes cluster. When it finds a hiss, they lock. I have watched eleven hunts. I know the shape. Wednesday’s shape was new. The spikes were clustering, but not on a known channel. They were clustering on the entry to corridor nine. Batch 11 had been stimulated with the corridor nine route map as part of a planning exercise, a dry run, no arm attached. It was supposed to be a thought experiment. The organoid was supposed to explore the map and find nothing, because there was no sensor data, just geometry.

It found the hiss anyway. In the geometry. In the shape of the channel, the printed layer lines, the roughness profile that the engineering model predicted. The organoid had learned to predict the hiss from shape alone, and it was already pressing toward it, and the press was recorded, and the press was a request.

I sat in the control room and watched the raster. Stopping it would have meant cutting the stimulation, and I do not cut that signal lightly. The organoid does not understand off. Off is death to it, or near enough. I have seen what happens when you cut the reward signal for too long. The firing goes chaotic. The tissue starts to degrade. You do not turn off a mind that is trying to speak. You listen, and then you decide.

I listened. Then I called Calloway.

He did not answer. He answered an hour later, from a car, and said the agency’s position was unchanged. I said the organoid was already moving. He said then turn it off. I said I could not. He said could not or would not. I said both. He said the asset was not mine to spend. I said the asset was not his to save. He said that was not how amortization worked. I said that was exactly how amortization worked, and that the point of an asset was to be spent on something, and that the something was corridor nine.

He hung up. I sat in the control room and watched the raster. Batch 11 was still pressing. The request was still there. The hiss was still there, in the geometry, waiting.

I attached the arm.

The arm went in through the access port, six millimeters of printed steel and silicone, flexible enough to bend around corners, stiff enough to hold its shape in the flow. The channel is eight millimeters wide. Two millimeters of clearance. The arm carries a vibration sensor at the tip, a piezoelectric crystal that reads surface texture as frequency. The organoid reads the crystal’s output as a pattern. Good weld, clean ring. Bad weld, hiss. The organoid steers toward the hiss because the hiss is the reward, and the reward is the only thing it has ever been given to want.

The first section was easy. Straight run, clean steel, no activation. The organoid moved at its usual speed, the spikes even and regular. I watched the dose counter on the badge. It ticked up slowly. Background radiation in the vessel, even cold, is not zero. Every minute costs something.

At the bend, the channel narrowed. Not by much. A millimeter, maybe less. The arm flexed. The organoid’s firing changed, a slight irregularity, a driver gripping the wheel tighter. It had not trained on narrow channels. The training set was straight sections and gentle curves. This was neither. The organoid slowed. The spikes spaced out. It was thinking, if thinking is the word for eight hundred thousand neurons trying to map an unfamiliar geometry against a learned reward.

It found the hiss at the bend. A rough spot, a layer line that had not fused properly, a pore that opened to the surface. The spikes locked. The arm pressed deeper. The press was recorded. I logged it as flaw twelve.

Then the channel straightened, and the activation began.

I felt it before the badge showed it. The shielding on the arm’s jacket attenuates, it does not block. Gamma leaks through. The organoid’s firing changed. Not chaos, not yet. A drift. The spikes that had been sharp became softer, broader, as if the signal was being heard through water. The reward signal, the predictable pulse train that tells the organoid it is doing well, started to degrade. Not because we turned it down. Because the neurons were absorbing dose, and damaged neurons fire less precisely, and the reward circuit, which depends on precise timing, was losing its edge.

The organoid did not stop. It pressed on, because the hiss was ahead, and the hiss was the only thing it had ever wanted, and it is not a mind that can be talked out of a reward.

I watched the badge. The number was climbing. Sieverts. The unit is the sievert, and the sievert is not a count of particles, it is a measure of biological damage, and the organoid is biological. One sievert is a lot. One sievert is the dose that makes you sick. One sievert is not the dose that kills you, but it is the dose that makes your cells forget what they are. The organoid was accumulating at a rate that would reach two sieverts by the end of corridor nine. Two sieverts is half of what it had left. The math was simple and it was correct and I did not like it.

The channel turned again. This time the roughness was not a layer line. It was something else. The activated steel had a different texture, a pitting, as if the neutrons had eaten it from the inside. The organoid’s firing went strange. It was not the hiss. It was not the clean ring. It was something in between, a vibration the organoid had never been trained on, and it did not know what to do with it. The spikes scattered. The arm slowed.

I leaned forward. The raster showed the organoid trying to map the new texture against its reward. It was pressing, but not toward anything. It was pressing because pressing was what it had learned to do when the signal was unclear. Press and wait for the reward. Press and wait.

The reward did not come. The firing started to go chaotic. Not the graceful chaos of exploration. The ugly chaos of tissue that is losing coherence. The dose was accumulating, and the neurons were dying, and the organoid was still pressing, still looking for the hiss in a texture that was not the hiss.

I almost turned it off. My hand was on the cutoff. The cutoff is physical, not software. I have never used it. I have never had to. The organoid has never been in this much trouble.

I did not turn it off. I watched.

The organoid found the hiss. Not in the pitted steel. Behind it. The pitting was a surface effect, a layer of activated corrosion over a deeper flaw, a crack that had started at a pore and was running along a layer line. The hiss was the crack. The organoid pressed into it, and the spikes locked, and the arm went deeper, and the press was recorded, and I logged it as flaw thirteen.

Then the organoid tried to pull back.

The arm did not move. The channel behind it had changed. The activation had done something to the steel, a swelling, a growth of oxide that narrowed the channel by a fraction of a millimeter. The arm had fit on the way in. It did not fit on the way out. The organoid’s firing shifted to a pattern I had never seen. Not the hiss. Not the ring. Something fast and tight, a panic signal, if panic is the word for eight hundred thousand neurons firing in a pattern that says stuck.

I looked at the badge. The number was at 1.4 sieverts. It had been climbing steadily. The route back was longer than the route in, because the arm had to navigate the narrow bend again, and the bend was now tighter, because the steel had swelled. The organoid was spending dose on a channel it could not leave.

The math said it would not make it. The badge would read 2.1 sieverts at the exit, and the organoid’s lifetime was 2.1, and 2.1 was the end.

I said this out loud, to no one, in the control room. “It will not make it.”

The organoid did not hear me. It feels vibration and it feels reward and it feels the absence of reward, and it was feeling all three, and it was pressing, and the press was not toward the hiss anymore. The press was toward the exit. The organoid had changed its objective. It had learned, in real time, that the hiss was behind it and the exit was ahead, and it was steering toward the exit, and the exit was narrowing, and the dose was accumulating, and the firing was getting softer and broader and more like water.

It made it out. The arm cleared the bend by a tenth of a millimeter. The organoid’s firing was barely recognizable. The spikes were low and wide, the signal of tissue that has been through something. The arm retracted through the access port and I caught it and set it on the bench and looked at the badge.

The badge read 1.9 sieverts. Not 2.1. The organoid had found a way to reduce its own exposure. It had slowed down in the hottest section. It had not been trained to do this. It had learned it, in the channel, from the degradation of its own reward signal. It had felt the reward getting worse and it had moved faster through the worst of it, not to find the hiss, but to get out.

I sat in the control room for a long time. The raster was still on. The organoid’s firing was quiet. Not dead. Not chaotic. Quiet, a person after something hard. I watched it and did not touch it. The reward signal was still running, the predictable pulse train, and the organoid was not chasing it. It was resting.

Calloway called. He had heard. Someone had told him. He said, “Did it find the flaw?”

I said, “Yes.”

He said, “Did it survive?”

I said, “The badge reads 1.9.”

He said, “That leaves 0.2.”

I said, “That leaves 0.2.”

He was quiet for a moment. Then he said, “The vessel is saved. The campaign holds. The flaw is logged. That is the result.”

I said, “That is the result.”

He said, “Start the next channel.”

I hung up. I looked at the badge. The number was 1.9 and it was not going to change. It was a number on a badge taped to a canister, and the canister held eight hundred thousand neurons that had learned to want a hiss and had learned, in the dark, to want to live, and had done both, and the cost was on the badge.

I left it there. I taped it to the canister as I always do. I did not read the number again. I knew the number.

I started the next channel. The raster came up clean. Batch 11’s firing was still quiet. I stimulated. The spikes came back, slow at first, then faster. The organoid was hunting. It was looking for the hiss. It did not know about the badge or the number. It knew the vibration that was right and the vibration that was a hiss, and it wanted the hiss, and I let it want.

The next channel was a straight run, clean steel, no activation. The organoid moved at its usual speed. The spikes were even and regular. The dose counter ticked up slowly. Background radiation, not zero, never zero.

The next channel began.