The article is assuming that the only methods available are simplified, toy circuit like examples and carefully constructed thought experiments.
I did not see anything in there that couldn’t have been written in the early 1990s, although admittedly I skimmed after reading five or six very familiar examples.
A breakthrough in methodology took place with “deep learning”. …
…I still feel it as kind of a cheaters approach…
… but that’s the old reductionist in me reasserting itself.
The method of starting with a strong model and then approaching that model does make having good ground truth assessments more valuable because they are at the core.
(see dangerous facial recognition flaws of non-white people as an example of that approach that can lead to extreme biasing that looks all too human of a flaw)
I don’t think we will end up with an objectively perfect consciousness in this way.
but a functional consciousness? i think so.