I don't agree. The typical undergrad CS program doesn't prepare people for AI research, but that's because it dedicates lots of time to discrete/combinatorial stuff and low level systems knowledge. Drop/condense architecture, OS, compiler, exotic data structures and algorithms, networking, programming languages. Add more calculus, linear algebra, probability, optimization, statistics, symbolic AI. That undergrad would be equally good at AI work as a traditional undergrad would be at compiler work. AI field is rarefied now but will not be so forever.