This is not a claim about how many states we can be in. This is a claim about the ability exercised in a single case, the ability to think in a form that is sum-giving for every sum, a definite thought form distinct from every other. When a person has acquired such an ability is not always transparent from successful answers, and it can be exhibited even by mistakes. [Emphasis added.]
But when he talks about machine addition, he counts any error, even a potential error many years in the future, as evidence that the machine doesn't truly add.
This is a blatant double standard. Logically, if a mistake is evidence that X is not performing the function, then that is true whether X is a human or a machine.