Some of your examples of paired input-recreation are curiously reminiscent of conductive dysphasia. The gist is preserved, but the reproduction is done at a semantic level, instead of a detail level.
Great post. very interesting. Has anyone started using this tech to identify the exact sub-regions in the brain where the individual phrases and words are being generated? i.e. where that little voice in our head is located. I can't see how you could accomplish what is described in this article without being able to locate what the machine-learning is learning from.
It is very cool. I still hate the possibility of it
Maybe you need to open your mind 🥁 *Ba dum tss*
This is super cool! Keep writing this amazing work!
Some of your examples of paired input-recreation are curiously reminiscent of conductive dysphasia. The gist is preserved, but the reproduction is done at a semantic level, instead of a detail level.
Great post. very interesting. Has anyone started using this tech to identify the exact sub-regions in the brain where the individual phrases and words are being generated? i.e. where that little voice in our head is located. I can't see how you could accomplish what is described in this article without being able to locate what the machine-learning is learning from.
In a word: yes. That's what Gallant's 2016 paper was about, trying to create a "semantic map" of the brain. See Figures 2 and 3 here: https://pmc.ncbi.nlm.nih.gov/articles/PMC4852309/
The one-liner is that semantic representations of language are spread across the cortex.
Interesting interesting; so it essentially boils down to costs - capitalism at its best.