If you would copy my brain but make it twice as large that copy would be as "lonely" as I would be and this would remain after arbitrary doublings. Single individuals can be extended in space without communicating with other individuals.

The "extended wire" thought experiement doesn't specify enough how that physical communication line is used. It's plausible that there is no "verbalization" process like there is an step to write an email if one replaces sonic communication with ip-packet communication. With huge relative distance would come speed of light delays, if one twin was on earth and another on the moon there would be a round trip latency of seconds which probably would distort how the combined brain works. (And I guess with doublign in size would need to come with proportionate slowing to have same function).

I think there is a difference between a information system being spatially extended and having two information systems interface with each other. Say that you have 2 routers or 10 routers on the same length of line. It makes sense to make a distinction that each routers functions "independently" even if they have to be able to suggest each other enough that packets flow throught. To the first router the world "downline" seems very similar whether or not intermediate routers exist. I don't count information system internal processing as communicating thus I don't count "thinking" into communicating. Thus the 10 router version does more communicating than the 2 router version.

I think the "verbalization" step does mean that even highbandwidth connection doesn't automatically mean qualia sharing. I am thinking of plugings that allow programming languages to share code. Even if there is a perfect 1-to-1 compatibility between the abstractions of the languages I think still each language only ever manipulates their version of that representation. Cross-using without translation would make it illdefined what would be correct function but if you do translation then it loses the qualities of the originating programming language. A C sharp integer variable will never contain a haskel integer even if a C sharp integer is constructed to represent the haskel integer. (I guess it would be possible to make a super-language that has integer variables that can contain haskel-integers and C-sharp integers but that language would not be C sharp or haskel). By being a spesific kind of cognitive architechture you are locked into certain representation types which are unescaable outside of turning into another kind ot architechture.