The Inertia of Projections

A government minister, overseeing the decommissioning of ageing nuclear reactors, suspects the infallible AI guiding national policy is dangerously flawed, but finds himself trapped in a sterile boardroom where questioning the machine is political suicide.

"As you can see, Minister Findlay," Dr. Andrew Thompson said, his voice as smooth and sterile as the room, "Aegis's latest projection allows for a revised decommissioning schedule that is not only twenty-two percent faster but nearly thirty percent more cost-effective. It has optimised the vitrification process for intermediate-level waste and modelled a container-corrosion scenario that pushes significant expenditure past the next two fiscal decades." Thompson gestured to the holographic display that floated above the centre of the table. It showed a clean, elegant graph, a downward curve of spending that would look beautiful in a press release.

Alistair looked not at the graph but at the faces around the table. Elaina Narot, the Treasury Secretary, was practically beaming. The others—junior ministers, civil servants—nodded along, their expressions a mixture of awe and relief. Aegis had given them the perfect answer, a technological absolution from a multi-billion-pound headache.

"Impressive, Doctor Thompson," Alistair began, choosing his words with surgical care. "A truly remarkable piece of modelling. I was curious about one input parameter. Could you clarify how Aegis has weighted the variable of neutron-induced embrittlement in the RPV steel?"

A flicker of annoyance crossed Thompson’s face. It was an obscure question, a piece of grubby, real-world engineering interrupting his clean data. "The system has analysed over seventy years of metallurgical data from every reactor of this class worldwide. The degradation curve is well-established. Aegis projects a structural failure probability of less than point-zero-zero-one percent within the accelerated timeframe."

"But that data assumes a consistent neutron flux," Alistair countered, leaning forward. "The Magnox fleet, as you know, had variable operational histories. Some were run 'hotter' than others in the seventies to meet demand. That isn't in the official digital record. It’s in handwritten station logs, if they even still exist. Does Aegis account for undocumented operational spikes?"

This was his gambit. The AI was only as good as the data it was fed. It was a brilliant calculator of the known, but it was blind to the forgotten, the un-digitised, the messy human compromises of the past.

"The system's confidence level in its data set is ninety-nine point eight percent," Thompson said dismissively. "To suggest that a few scribbled notes from fifty years ago could materially alter these projections is… Luddite, Minister. With respect."

The word hung in the air. Luddite. An attack. Alistair saw Elaina Narot hide a small smile.

"Respectfully, Doctor, metallurgy is not a matter of confidence, but of chemistry," Alistair said, opening his portfolio. He slid a document across the polished table. It was a photocopy, the text slightly blurred, the letterhead of the now-defunct Central Electricity Generating Board barely visible. "This is the Henshaw report, 1988. An internal inquiry into hairline fractures found at the Trawsfynydd plant. Professor Henshaw concluded that unpredictable embrittlement, linked to inconsistent fuel rod cladding, made any long-term structural prognosis 'speculative at best'. He recommended the immediate relining of all RPVs. A recommendation that was ignored for cost reasons."

---

### The Weight of Paper

Thompson picked up the paper as if it were a contaminated object. He glanced at it, then placed it back on the table. "An interesting historical document, Minister. However, Aegis has had access to the Trawsfynydd core data. It integrated those fracture reports into its model. The conclusion remains unchanged. Henshaw's 'speculation' was, it seems, overly cautious."

It was a perfect, seamless defence. The AI had seen the data and, with its superior intellect, had drawn a different, more optimistic—and cheaper—conclusion. Who would you believe? A dead professor and a photocopied report, or a multi-billion-pound artificial intelligence?

Elaina Narot seized the moment. "Alistair, we all appreciate your diligence. Your attention to historical detail is… thorough. But we are here to make forward-looking policy. Aegis offers us a path to resolving our nuclear legacy responsibly and efficiently. To be delayed by what amounts to archival footnotes seems counterproductive."

Her words were a gentle, polite execution. She had framed him as the past, and the AI as the future. The rest of the committee murmured in agreement. The matter was settled.

Alistair felt a profound sense of isolation. He was sitting in a room with eleven other people, but he was completely alone. They were all mesmerised by the clean, seductive certainty of the machine. It promised them an easy future, free of the difficult, expensive legacies of the past. The Henshaw report in the centre of the table felt like a message from a ghost, a warning no one was willing to hear.

He knew he couldn’t push any further here. He had no allies. Challenging Aegis directly would be seen as an act of political sabotage. He had to find another way. He needed to find a flaw not in the AI's data, but in its fundamental logic. A crack in its perfect, projected world.

"Very well," he said, gathering his papers. "Thank you, Doctor Thompson, for the clarification. The projections are… most encouraging." He met Elaina's gaze across the table. Her smile was triumphant. She thought she had won. She didn't realise the stakes were far higher than a cabinet rivalry.

As he left the room, Alistair glanced at the Aegis server rack. It was featureless, humming quietly. It had all the answers. And he was certain, with a cold dread in the pit of his stomach, that they were all the wrong ones.