Not sure there's a definitive answer, but I believe pg saw an ideal evolutionary path for Lisp that was not being followed. I believe the plan, with arc, was to correct the foundation such that this hopeful evolutionary path could happen.
This is just what I get from reading a bunch of his post, code comments and articles. It's probably a good idea to read pg's articles from years ago during the initial design phase to get an understanding of what choices were made and why.
"an ideal evolutionary path for Lisp that was not being followed"
"much easier to learn"
These are my reasons in a nutshell.
I've always liked the philosophy of programming in whatever language was most suitable to the task, and everywhere I looked indicated that lisps emphasized that philosophy. It made sense, since s-expressions seemed pretty ideal as a way to use a single generic parser to handle all kinds of custom languages.
As I was looking into it, I read about Paul Graham's plans for Arc and liked the idea of a lisp with that kind of open-to-the-programmer philosophy (and avoidance of extraneous parentheses). And then I found out Arc existed, and I gradually made my way over to Arc from Groovy.
Arc doesn't do everything I'd like it to: No modules, less polymorphism than I'd like (but way more than Scheme), not enough consistency for the kind of metaprogramming I have in mind. Still, it's pretty close to what I want, and it's the kind of foundation that's allowed me to get closer and closer.
