Thought of the morning: people often say "we have variables and comments because programming languages are for humans, not computers". But if computers were really at the point where they were able to program themselves, and I mean really do it (even invent and code new things, and I don't mean genetic algorithm bullshit, I mean thinking about design... so this also means code as more than just "learning", but actually planning and programming something new), would they need variable names and comments?
My thinking is: yes, or they'd need something like it. If you don't have this, this means you're effectively reverse engineering "purpose" in the codebase all the time, which can be both expensive and faulty. I think any AI that's not some ~dumb application of known heuristics will need to be able to "think" about the code at point, and knowing the reason for a code change is important. So of course relevant information should be recorded in that portion of the code.
Now, does that mean something as messy as English will be used (as the majority of present code is written in English)? I doubt it. Probably something like Lojban will be used. Maybe it will not even be plaintext code: it could be machine-readable, machine-contemplatable code with "byte compiled lojban", or similar.
Relatedly, in the (glorious???) future where machines can think and design programs, assuming enough resources exist to keep said machines running, humans will have to interface with computers on a computer's level more often. Will Lojban as a second language be mandated in schools?
(On that note, maybe I should really learn Lojban... :))