>>27
Inferno, colorForth, Singularity, SharpOS, JX, JNode, JavaOS, etc.
I don't know all of these, but the ones I know are either not serious, or not based on a truly high-level language.
Modern computers can run Lisp programs a lot faster than Lisp machines.
I know, that's why I say this. I want it to work like they wanted Lisp Machine OSes to work. The only barely used OS like that I know at the moment is Sugar for the OLPC.
Artificial intelligence isn't really useful for very many things.
Several AI techniques provided as a core API or used by the operating system functionality would improve the user experience in things such as dialog default action prediction, UI object placement, rule-based scripting, etc.
Sure, fuse (http://fuse.sourceforge.net/) + the Plan 9 filesystem is a lot more ambitions than the Plan 9 filesystem by itself.
What I'd implement includes functional directories that you could define on the fly as a function (think of an executable file that acts like a directory and gets loaded and daemonized when you reference it), such as an HTTP directory to browse the web or a directory exposing a lazily-evaluated fibonacci sequence in subdirectories. I'd also like to have globally and locally indexed metadata, a reverse link of hard links, transactions for transactional filesystem controllers, and other features supported by the filesystem API.
You can do that on any operating system.
But it's not done. If I had magic time to work on an OS, I'd have to provide it with the basic tools and applications for its use, and while I'd import/adapt much of the GNU system, the standard editor would be this.