One general limitation is that the decoder can only recognize concepts (“semantic features”) it has been trained for with the particular subject. The paper mentions that “we recorded responses while the subject listened to sixteen hours of naturally spoken narrative stories”. If the training material didn’t contain (say) the concept “grandmother”, the decoder probably won’t be able to decode that concept. And the more abstract the concept, the more training is probably needed. You likely also have to retrain each year or so because your brain changes.

Nevertheless, if it could somehow be commoditized into an affordable appliance, it could be very useful.

You just need to find a small spanning set of concepts for all of human experience and then just train the decoder on those.

(I'm being serious but do not have time to elaborate as I normally would on a serious comment).

In /Interface/ (https://www.goodreads.com/book/show/6563920-interface), they imagine kind of the opposite of this: They have a primitive mind-machine interface and probe all the different kinds of sensations it can transmit to a person's mind --the smell of apple pie is one example--and train them to associate it with some meaning (in this example, the guy is a politician and associates apple pie with "American values"). Then the campaign manager can press the "apple pie" button, which causes the politician to smell apple pie, indicating that he should should transition to talking about "American values" in his speech.

I would assume that those spanning sets would differ between individuals to a significant degree, so you'd still have to do the more extensive training to determine and verify a working spanning set in the first place. Maybe after training with a large number of people, certain typical patterns will emerge that can be used to speed up the process, but I suspect you'll still have to deal with a lot of individual variation.

Not a finite set of concepts per se, but a finite set of patterns that hides behind the near infinite set of all possible concepts.

A vector basis set for all human experience sounds like it’d be pretty neat.

Do you mind giving me an eli5 of what the indeterminacy of translation states? I don’t have much of a philosophy background.

" A vector basis set for all human experience sounds like it’d be pretty impossible and actually incoherent and anti-empirical."

Why would it be impossible? Everything about a human being can theoretically be deterministically simulated.