Why not simply estimate it yourself? These sorts of things aren’t very hard to do. For example, you can estimate typing as follows: peak at 120 WPM; words are average 4 characters; each character (per Shannon and other’s research; see http://www.gwern.net/Notes#efficient-natural-language ) conveys ~1 bit; hence your typing bandwidth is 120 4 1 = <480 bits per minute or <8 bits per second.
Do that for a few modalities like speech, and sum.
I’ve just noticed he said “an arbitrarily fast mind running in a human body”, not an actual human being, so I don’t think it would be much slower at typing uuencoded compressed stuff than natural language (at least with QWERTY—it might be different with keyboards layouts optimized from natural language such as Dvorak, but still probably within a factor of a few).
The 120WPM is pretty good for the physical limits: if you are typing at 120WPM, then you have not hit the limits of your thinking (imagine you are in a typing tutor—your reading speed ought to be at least 3x 120WPM...), and you’re not too far off some of the sustained typing numbers in https://en.wikipedia.org/wiki/Words_per_minute#Alphanumeric_entry
La Wiki is apparently not using the entropy estimates extracted from human predictions (who are the best modelers of natural language). Crude stuff like trigram models are going to considerably overestimate matters.
Why not simply estimate it yourself? These sorts of things aren’t very hard to do. For example, you can estimate typing as follows: peak at 120 WPM; words are average 4 characters; each character (per Shannon and other’s research; see http://www.gwern.net/Notes#efficient-natural-language ) conveys ~1 bit; hence your typing bandwidth is 120 4 1 = <480 bits per minute or <8 bits per second.
Do that for a few modalities like speech, and sum.
I’ve just noticed he said “an arbitrarily fast mind running in a human body”, not an actual human being, so I don’t think it would be much slower at typing uuencoded compressed stuff than natural language (at least with QWERTY—it might be different with keyboards layouts optimized from natural language such as Dvorak, but still probably within a factor of a few).
The 120WPM is pretty good for the physical limits: if you are typing at 120WPM, then you have not hit the limits of your thinking (imagine you are in a typing tutor—your reading speed ought to be at least 3x 120WPM...), and you’re not too far off some of the sustained typing numbers in https://en.wikipedia.org/wiki/Words_per_minute#Alphanumeric_entry
My point was that 1 bit per character is an underestimate.
La Wik says 8 bits per word, FWIW.
La Wiki is apparently not using the entropy estimates extracted from human predictions (who are the best modelers of natural language). Crude stuff like trigram models are going to considerably overestimate matters.