On: Amend copyright licences to halt AI misuse and reassert human control

25 August 2026.

A call to amend copyright licences so that machines may not devour what they cannot credit. My first instinct is sympathy; my second is the surveyor’s habit of asking what, precisely, is being measured before agreeing to regulate it.

Who compiled this proposal? An editorial, I gather - an opinion wearing the dress of a finding. Very well, but let us be honest about its provenance: no observations, no instrument, only argument. Argument is legitimate work; I have done my share. But argument about the behaviour of these engines requires data on that behaviour, and here the gap inventory is startling. Nobody can say, with any rigour, how much of any given output derives from any given source in the training mass. The claim “this machine copied my work” rests on a black box whose interior has not been charted by any systematic sweep. We are legislating over an unexplored sky.

And there is the old confusion between precision and accuracy. A licence clause drafted to six subclauses of legal exactness gives an appearance of control that the underlying measurement - whatever “derivation” means for a statistical engine - simply does not support. Better a blunt, honest rule than an elaborate one calibrated to nothing.

Still, I take the editorial’s point with feeling. When I catalogued the southern heavens, every observation carried my name and the conditions under which it was made, because provenance is what makes knowledge usable by others. If these engines produce text without traceable descent, they produce something cheaper than knowledge - plausible paste. Insisting that human authorship remain legible is not sentimentality; it is the maintenance of the catalogue itself.

My position: amend the licences, yes, but first demand the maps. Regulate what you can measure, and mark clearly as unknown all the rest. Otherwise we shall pass laws against weather.