[dhsmegmi] Chess opening theory investigated by computers

Chess opening theory is perhaps one of the oldest "distributed computing" projects of mankind.

Opening theory has developed though thousands of grandmaster games.  But nowadays, we can quickly generate billions of grandmaster-level games by computer, so opening theory is no longer interesting.

How much of opening theory can be recovered "from scratch" by allowing a computer to play itself and applying the UCT algorithm (or something similar)?  The opening table can be restricted (albeit to gigabytes or terabytes) to nodes whose parents have high probability mass.