Last time I asked who is allowed to test your Mark XIV. This time: what the test looks like when it is real.
In an experiment for DARPA — the Pentagon’s advanced research agency — a sensor built to detect approaching humans was set up in open ground, and eight US Marines were invited to try to beat it. They did not jam it, hack it, or spoof its electronics. Two somersaulted their way across the field. Two advanced underneath a cardboard box. One walked towards the sensor dressed as a fir tree. All eight reached the objective undetected.
Paul Scharre — a former Army Ranger who helped write the Pentagon’s rules on autonomous weapons — tells the story in Four Battlegrounds, and pairs it with the sentence it illustrates: AI is powerful, “yet its power is a brittle one.” The system had learned to recognize the humans it had been shown. Humans who walk. Nobody walked.
I laughed the first time I read it. The second time I stopped laughing, because two things in the story are not funny at all.
The first is the cost structure. The sensor represented years of funded research; the countermeasures were invented in an afternoon, for the price of a cardboard box. That asymmetry is not a punchline. It is the standing condition of any learning system placed in front of an adversary who is allowed to move.
The second is how the countermeasures were found. No methodology produced the somersault. What produced it was people who wanted to win, probing a rule-bound system until it gave way. Armies own that resource in unlimited quantity — and mostly aim it at the enemy’s equipment. The rare discipline is aiming it at your own, before fielding rather than after. That is the answer to the question from the torpedo letter: a real test is not a checklist administered by the office that built the thing. It is eight people with permission to be unreasonable.
The Marines did not out-compute the algorithm. They stepped outside its training data — and found the door was never locked.
So, this letter’s question: who in your organization is allowed to play the Marines against your own system — before someone else does? Reply if you have a story. The best ones come from people who were told to stop.
Since the last letter
America’s military revolution was named in Moscow first — Adamsky on Assault Breaker: a technology and its meaning are separate assets, sometimes held by different countries.
The missile was the least of the factors — Finkel on the Sagger shock of 1973, and why recovery time, not foresight, is the scarce resource.
The future-war genre was born selling something — Dorking, 1871, and the first question to put to any confident forecast.
One thing worth reading
Paul Scharre, Four Battlegrounds: Power in the Age of Artificial Intelligence (2023). The book the Marines come from, and the rare one on military AI written by someone who has both carried a rifle and drafted policy. Read it for the pattern rather than the anecdotes: every capability in it arrives with the adversary’s cheapest countermove already attached — which is the only way a military technology should ever be described.
Next letter in around two weeks. Reply if you disagree — that is what this is for.
Christoph