Post by Tim Chow
Obviously, I don't expect to be able to change your mind any more than I
expect to be able to change Murat's mind. But Paul Epstein has shown evidence
of an open mind, so it's for the benefit of people like him that I articulate
contradictions and biases in what you say, that might cause confusion.
Let me elaborate a bit on this. Here's how I would imagine someone---let's
call her Twiggy---who is consistently trying to apply the "just make the DMP
play" maxim might analyze this position.
Twiggy says, "First of all, what's the DMP play? It might not be clear to
everyone, but it's clear to me---play safe, 8/6 8/2. Now, what about the
banana split? That might win more gammons at the cost of some wins. How
do I assess the tradeoff? It looks murky. So forget about trying to assess
it. Just make the DMP play. 8/6 8/2 it is." When the bot verdict comes in
and says to hit, Twiggy says, "Well, let's see...was I wrong that the DMP
play was clear? No. A DMP rollout puts 8/6 8/2 ahead by about 0.018, which
is clear by DMP standards. Looks like the gammon tradeoff was favorable
after all. Well, in some positions you just have to evaluate the win/gammon
tradeoff if you want to master them. But I did the best I could with the
knowledge I had."
Alternatively, Twiggy might say, "First off, what's the DMP play? Well, it's
not clear to me. Usually it's to play safe, but if O's board is crunched
enough, the banana split can be better at DMP. I'm guessing that playing
safe is better at DMP, but maybe not by a lot. Now what about gammons?
Hitting must win more gammons. If it's close at DMP then I'm going to go
with the gammonish play." Upon looking at the bot results, Twiggy says,
"Well, I guess I got the right answer, but I'm slightly surprised that 8/6
8/2 is as far ahead at DMP as it is. I need to study this sort of position
at DMP more."
Either of these would be a principled approach. Of course I've used a lot
of words and it takes effort to type out so many words. Who has the time?
But the point is that if you advertise an approach but don't practice what
you preach by following it in a principled way, and then point out when it
runs into difficulties (and what to do when difficulties are encountered),
then the approach loses credibility.