Basically, he explained that shared-memory systems simply don't make sense for distributed programs.
It was years ago, but I believe he said something to the effect of "imagine that there's a server in Europe, and a server in America. Where does that shared memory exist?? Shared memory violates the laws of physics!"
He then explained that since Erlang processes don't share any memory, and the only way for them to communicate is via message-passing, there's no need for locks or anything like that.
