Not really "human". Doing no-carry addition is much easier for a machine to do as well, as that's basically what XOR does, i.e., SIMD. Carry introduces dependencies between the digits, potentially as long as the whole string goes. So that's pretty hard to understand, also for a machine.
At first I thought you were saying that doing arithmetic by carrying is not really a human trait, but on reflection, I think you are saying that carrying methods are inherently mistake-prone, regardless of who or what is using them.
I feel it would be a very big deal if GPT-3 (or this variant) was carrying, even if imperfectly, but other comments here seem to be suggesting that, on account of the way all input is tokenized, consistently doing arithmetic by carrying would simply be outside of the set of transformations it could perform (though some results that look like it might arise by chance.)
