IMO he’s trying to say that if you observe a machine from the outside (ie only what (sequences of) inputs lead to what (sequences of) outputs), then the mere observation that it behaves as if it understands the problem is not sufficient to conclude that it understands the problem. This is because understanding is some property of the internals. The presence of understanding is not deducible from the outside, even given infinitely many maximally diverse observations.
The no free lunch theorems basically say that if you are unlucky enough with your prior, and the problem to be solved is maximally general, then you can’t improve on your efficiency beyond random sampling/brute force search, which requires you to examine every input, and thus you can’t get away with algorithms that don’t require you to examine all inputs like in brute-force search.
It’s closer to a maximal inefficiency for intelligence/inapproximability result for intelligence than an impossibility result, which is still very important.
IMO he’s trying to say that if you observe a machine from the outside (ie only what (sequences of) inputs lead to what (sequences of) outputs), then the mere observation that it behaves as if it understands the problem is not sufficient to conclude that it understands the problem. This is because understanding is some property of the internals. The presence of understanding is not deducible from the outside, even given infinitely many maximally diverse observations.
Only if you can’t examine all of the inputs.
The no free lunch theorems basically say that if you are unlucky enough with your prior, and the problem to be solved is maximally general, then you can’t improve on your efficiency beyond random sampling/brute force search, which requires you to examine every input, and thus you can’t get away with algorithms that don’t require you to examine all inputs like in brute-force search.
It’s closer to a maximal inefficiency for intelligence/inapproximability result for intelligence than an impossibility result, which is still very important.
I think Searle would disagree. But I also think this entire thought experiment is dumb.