In grad school I was for a year part of a small study group. We shared notes, on paper and aloud. It made each of us far, far better at the subject than we could have been on our own. And we were at most 7 people (sometimes only 3).
When developing Hode[1], and when helping develop Semantic Synchrony[2], I imagined some kind of mind-meld utopia where superbeings would not just frolic in but also tend each others' knowledge gardens. (Brandon Toner on Twitter[3] even introduced me to a hashtag for them ... which I can't remember ...)
Reading nonlinearly is a faster way to accumulate knowledge than reading linearly. That's is why people who really care prefer newspapers to watching the news on TV. And a format that lets lots of people contribute their input can potentially convey more than a format that locks you into reading a single author. Alas, the potential for noise is higher too -- but still, this is why Wikipedia has been so wildly successful.
One might ask, "We've already got Wikipedia; what possible improvement on that do you see?" The thing about Wikipedia is it presents what the author(s) consider(s) settled knowledge. It is a place for record, not a place for debate. A good shared knowledge base would by contrast also resemble Twitter -- but more organized, more navigable.
In practice the personal knowledge bases I've seen look much more like silos. I once frolicked for an afternoon through Andy Matuschak's[4], which was quite enjoyable. But it was mostly about ... can you guess? ... how to build a knowledge base.
If someone finds a scheme that scales, I believe it could be much more powerful than what we currently call AI.
[1] https://github.com/JeffreyBenjaminBrown/hode
[2] https://github.com/synchrony/smsn/
[3] https://twitter.com/brandontoner
[4] https://notes.andymatuschak.org/Knowledge_work_should_accret...