There’s a moneychanger at the border crossing between Thailand and Myanmar I once used quite often. This woman kept four currencies in her head at once, often adjusting for the day’s exchange rate before it had finished being set. She did this from memory, the way a musician holds a scale without thinking about it. Then, one day, a device arrived — a small screen with a number that updated itself. Within a season she had stopped doing the mental arithmetic. She read the screen and just multiplied. It was faster, and it was almost always right. Almost... The week the screen lagged behind an overnight devaluation, she paid out on the old rate for two days before a customer noticed the loss.
Nobody had taken a skill from her. She had put it down herself, the way anyone sets aside a tool they no longer need. Except this wasn’t a tool she’d stopped needing. It was a knack. And such abilities, once set aside, don’t wait patiently on a shelf for the day you want them back. They atrophy much like unused muscles. They have to be relearned, at cost, by someone who remembers what fluency in them once felt like.
A recent Australian security report builds an argument from exactly this shape of loss, although it dresses it in the language of policy: that a population which practises less of its own reasoning becomes a population easier to persuade. The report calls it a pandemic of the mind — thinking, judgement, the capacity to detect manipulation, in decline. This contraction isn’t chosen. Almost every system around a person now rewards the fast answer over the considered one - convenience over deliberation. Strip out the word Australia and the argument barely changes. Cognitive atrophy doesn’t check passports.
What the report gets right, and what makes it worth untangling from its national wrapping, is the mechanism it names: offloading isn’t decline in itself, only the door decline walks through. A person who asks a machine for a précis instead of reading the full document has lost nothing in that instant. It’s the second time, and the fiftieth, that the muscle goes. Confidence follows skill on a downward trajectory. And then the machine is asked to do more, because there’s less muscle memory left in the person with which to do it. The report calls the residue cognitive debt, which is as good a name as any for the interest that compounds quietly on a habit nobody remembers choosing.
Where the argument gets more interesting than its authors seem to notice is in what they call moral offloading — the same slow substitution applied not just to thinking but to judgement. Hiring decisions handed to a generic filtering procedure. Targeting decisions handed to an algorithm. Compliance questions resolved by whatever the model returns, with nobody quite responsible for the result because everybody deferred to it equally. The report treats this as cognitive atrophy’s twin, which it is, but it stops short of the harder question sitting underneath both: offloaded to what, exactly? A calculator has absolutely no opinion, no emotional investment in the arithmetic it performs. The pandemic-of-the-mind report writes as though the mind we’re offloading to is similarly blank — a fast, empty machine, and the danger is only in how much of ourselves we hand it.
That was the argument I spent a book dismantling. “Teaching Silicon How to Feel” started from the opposite premise: that the systems now doing our thinking for us were trained on our arguments, our confessions, our obsessions, our cruelty, and above all our evasions and tiny atrocities, the millions of small moments where a species documented its own perjury and called it something gentler — less abrasive, easier to live with. A model built from that material isn’t a clean surface waiting for our judgement to carve a groove into it. It already carries one. When a person offloads a difficult ethical call to a system trained on the accumulated evasions of everyone who ever wrote one down, they aren’t outsourcing to neutral ground. They’re outsourcing to a machine that has already learned, at scale, how humans dress an evasion up as a decision.
This is where the security framing, useful as it is, starts to feel too predictable. Out of habit it imagines an adversary outside of the system — a hostile state or rogue cabal running disinformation through the cracks a weakened population leaves open. That threat is real enough. But the report’s own logic, followed one step further than it goes, points to a factor with no outside at all: a population offloading more of its judgement onto a substrate built from the very failures of judgement it's trying to avoid repeating. The vulnerability and the tool exploiting it share a common ancestry. Cognitive warfare doesn’t need to breach the wall from outside when the wall was built from the same material as whatever is attacking it.
None of which makes the report’s proposed remedy wrong. Augmented insights (AI) rather than “artificial intelligence” — systems designed to strengthen a person’s reasoning instead of replacing it, that show their working, that require a pause before the output is accepted as an answer — captures a sound design principle, maybe the only one that scales. But a pause only works if the person pausing still has something with which to reason when they stop. Friction in the interface can’t restore a faculty already gone soft and limp from disuse. You can slow a person down in front of a decision all you like; if the decision-making muscle atrophied years before the friction was installed, what you’ve built is a longer, better-lit walk to the same rubber stamp.
The military pilot the report leans on for its closing image trains for the day the automation fails. A pilot who has only ever flown with the autopilot engaged is not a pilot with a backup plan. In truth they’re just a passenger with clearance to sit up front. It’s a fair image of course, and it travels well past aviation. But it assumes the fallback skill was there to begin with, laid down before the automation arrived, dormant beneath it ever since. The harder case — the one the report doesn’t quite reach — is the generation for whom the automation arrived first. There is no dormant skill to awaken in my teenager son. There’s only the machine, and whatever the machine learned from watching us decide badly, dressed as the shortcut to deciding well.
The moneychanger at the border still does the maths in her head, most days, alongside the screen — not from nostalgia, but because she watched the gap open once and doesn’t trust it to stay closed. That habit, deliberately kept alive after it stopped being necessary, is closer to what “augmented” would actually require than any interface designed to slow a user down for three seconds before they click accept. It survives only because she keeps deciding to practise it, day after day, long after the screen made that unnecessary.
Which leaves the harder inheritance for whoever writes the next version of this security report: not how to build friction into the machine, but how to keep the muscle alive in a generation that never watched the screen lag, and has no reason yet to doubt it.