On October 1, 2021 at 5:44:33 AM UTC-6, Tim Chow wrote:
On 10/1/2021 1:26 AM, MK wrote:
What if Ray had asked the same question about Hypergammon?
My friend/colleague did manage to do a few experiments with this
before he put the project on hold. What he said was that the bot
would double very aggressively, because there was no easily
perceived penalty for doing so. So the cube would get exponentially
high. But that would mean that games with high cube values would
be disproportionately weighted, making it hard for the bot to tell
what was luck and what was skill.
This an interesting bit of information. Not much yet but better
than nothing we ever had and hopefully will lead to more.
If I understand correctly, the experiments were 1-point money
games?
Interestingly, here you talk about the bot as if it could perceive
things as a human, like you all do when discussing positions
and "ask the bot to whether pay now or later".
"Easily" or not, the bot doesn't need to nor can "perceive penalty"
for playing in any manner that it does. It just makes the moves
that it learned to be most likely to win.
Similarly, the bot needs not to "tell what was luck and what was
skill" either. It has played 4 billion random games. Luck evened
out. And the "best" moves have bubbled up. It doesn't make any
decisions so to speak but just pulls up the best move from its
bag at each position.
I don't understand what you mean by "games with high cube
values would be disproportionately weighted". The bot can't
know nor cares what humans perceive and interpret how. It
just executes. The results are the results. That's it.
So it would never learn how to play well. We knew that it was
playing poorly because it would get trounced by the "solved" bot.
How well did you train the bot? How many games they played?
How did you "know!" without completing the experiments? You
are just trying to confirm your denial. I will argue that if the bot
is sufficiently trained and allowed to play a sufficient number
of games against your "solved bit", it will trounce it instead.
The way around this would be to implement match scores so that
there was a limit on the cube size, but that was around the time
my friend put the project on hold.
From even this little experimenting, it looks like things are going
in the direction that I had predicted. It's just a matter of time
that I will be proven right. Here are a few examples of what I had
said in the past:
1) ==================================================
https://groups.google.com/g/rec.games.backgammon/c/-Y9AWXOFwrc/m/gkJ0UUXBajQJ
Murat clearly plays an aggressive cube against the Bot.
I never liked the cube but once I submitted to the fact
that I had to live with it, I came to believe/practice
that you should use it, almost as another checker, not
just let it sit there.
--------------------------------------------------------
"Use the cube almost as another checker"? What a novel idea! :)
And "Don't just let it sit there". In fact, I remember making a pun
using the word "incubate" to say don't sit on it if you have access
to the cube. Your colleague's alpha-bot must have heard me. ;) =====================================================
2) ==================================================
https://groups.google.com/g/rec.games.backgammon/c/4VxyU97HmO4/m/92tzL48MVx4J
Let's cut it short and jump to where we lose contact... At that point
I'm ahead by 1 pip and I'm on roll, (i.e. I'm ahead by about 9+ pips).
By the coin tossing logic and strategy, I already won, right...??? :))
I mean, after 4 billions of games starting at this very position, I
would win more and on top of that, the cube is yet at a measly 4
in a looong 64-point match...
.................
Remember, if we you are playing with the cube, "use it"...! Don't
just let the silly thing sit there...
......
Not necessarily... I played so many series of long matches for
so many purposes... I'm not ashamed to admit that I even played
making random cube decisions, similar to the experiments
they had done with old ladies in quilting-clubs and monkeys to
play the stock market...
If some monkeys did just as good with the cube as your much
adored book-writing-world-class-clowns, you guys would have
been heart-broken much more than me... :)) --------------------------------------------------------
Yes, I had even experimented with playing like Alpha-Murat... :)
I guess it has always been easier for me to be more open minded
and have the courage to speak out my mind here because I have
been free of the peer pressure that most of you may be crippled by. =====================================================
3) ==================================================
https://groups.google.com/g/rec.games.backgammon/c/j1zzL5M67QQ/m/Kwtuy5-OepsJ
Albert Silver wrote:
I recall your writing this elsewhere but don't understand
it. Although one could easily double at 51%, it would be
a colossal blunder to drop such a cube, since the classic
.....
P.S. Just to clarify that my above arguments apply only
to reducing the effect of "cube skill" passively, by
dampening it. Otherwise, my view of cube ownership is
almost the opposite based on the fact that your opponent
has access to and thus "owns" the cube anyway. So, don't
let it just sit there but use it as soon as you can, as
often as you can. After early cubes, there is still lots
of checker play left to favor the stronger player...
--------------------------------------------------------
These ideas of mine have in time evolved to argue that the player
who gains the first/earliest advantage in a game will never lose
it (that is statistically, in 4 billion trials).
I had predicted that an alpha-bot would do just that: perhaps
double as soon as after the first roll and as often thereafter, so
that the cube would reach the match limit only after a few moves.
After that, I had even suggested that people playing backgammon
for gambling shouldn't bother playing past the opening roll, since
winning the opening roll creates and advantage that will never be
lost for the rest of the match (that is statistically, in 4 billion trials). And so, I had also suggested that they may save time and be more
efficient by giving up backgammon for taking up coin tossing to
gamble for money.
=====================================================
Anyway, this subject had been discussed long and hard over
many years. I don't wnat to go forever with reminding you of
what was said. I'm sure you'll remember enough of it to see
you all are just in demial and only delaying the obvious to come.
MK
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)