I suppose a true alpha-zero-bg-bot would learn to play
through random checker and random cube decisions,
without any human introduced bias.

Even though we have much more CPU power on a PC
today than the computers TD-Gammon v1 was trained
on, training a true alpha-zero-bg-bot would still require
supercomputers that are still only owned by IBM, etc.

So, what about a practical, temporary compromise..?

We can easily train a TD-Gammon v1 like bot to play
1-pointer "money games" 1,000 times better, even if
proportionately much less improvement compared to
it which was already as strong as any human player
at the time.

Now, for cube decisions, what if we don't use any
jackoffski formulas but simply make them crudely
on cubeless equities..?

Similar to experiments that I have suggested/done
and what Axel assumedly is still doing. Except that
in this case, we can experiment by pitting our bot
against Gnubg and XG, with or without further training.

In other words, we can either train our bot further, and
long enough, by making cube decisions on crude winning
chances based on cubeless equities and then play shorter
experimental sessions against other bots; or skip the
additional training and play much/enough longer sessions
against other boths, in a way combining the two.

I think all current bots have been trained by cubeless
1-pointers, then MET's and doubling window formulas
are inserted to play cubeful or cubeless matches.

So, it should be trivially easy to modify Gnubg to just
not use doubling window formulas but to use experimental
(preferably user selectable) doubling/taking points.

I have a feeling that such a compromise bot will be good
enough to beat the current bots and debunk most checker
and/or cube dogmas perpetuated by mentally ill gamblers
(especially with math degrees).

Any thoughts? I would be willing to contribute considerable
time and effort towards such a project.

MK

On 12/21/2021 8:42 AM, MK wrote:
> Now, for cube decisions, what if we don't use any
> jackoffski formulas but simply make them crudely
> on cubeless equities..?

More in the AlphaZero spirit would be to train a separate net
for each match score. Start by training neural nets for DMP,
and 2-away Crawford. No cube here to worry about. You can
then figure out what the match-winning chances are at 2-away
Crawford, by doing Monte Carlo simulations.

The first score where you really need to consider the cube
would be 2-away post-Crawford. If you're willing to cut
corners then you could cheat and use human intervention here,
but it would be interesting to see if the neural net can
figure out the cube on its own. In particular, does it
play on for the gammon if it's the 2-away player and wins
the opening roll with 31 and the opponent rolls 62 or 63?

In any case, once you have 2-away post-Crawford worked out,
you can then train a net for 3-away Crawford, and then finally
move on to the more interesting score of 2-away/3-away. It
would be interesting to see a neural net try to figure out
the cube strategy on its own here. (Of course if you're willing
to cut corners, you could use the "standard" match equity table
and jump to 2-away/3-away directly, without training separate
nets for the smaller scores.) And then obviously, you can
work your way up the chain of match scores, training separate
nets for each score.

---
Tim Chow

If Deepmind will ever do something like DeepBG I'm very confident that they won't come up with such a success as in Go and chess.

- chess and go are two dimensional games. The idea to use NN that are used in picture recognition was the key idea. BG instead is only a 1-dimenional game so the achievements of deep learning don't achieve much

- if you take into account what a amount of resources deepmind throws in, they would surely come out with something better, but not washing the floor with any of XG, BGBlitz or GnuBG. The alphago team was 18 people IIRC. When Aja Huang was asked on a keynote how to parallize learning he said that they had other people that did this (so the 18 were only involved with Go). Compare this to leisure time programmers (BGBlitz, GnuBG) or a single full time programmer (Xavier?). The hardware that learned "chess in a couple of hours" is not something anyone of us have. IiRC when you run a fast desktop PC you'll need about a year of training to have the same computing power. Money were no issue for them.

- deep mind always tries to do thing that haven't been done yet.Unless one of their Cxx get's mad on BG they wont do BG.

- There's no reason to bring DeepMind into this discussion. The basic
ideas are out there, and other people can more or less replicate the
results, with much less effort than was needed to come up with the
ground-breaking ideas in the first place.

- The money/computing power is probably also not an issue, because the
amount needed for BG is probably going to be much smaller.

- I don't understand why so many people seem to be obsessed with the
winning rate from the starting position. Surely it's obvious that the
point is to understand superbackgames and wild containment games and
other positions which the current bots have less than zero clue about.

On 12/21/2021 6:12 PM, Frank Berger wrote:
> If Deepmind will ever do something like DeepBG I'm very confident that they won't come up with such a success as in Go and chess.
>
> - chess and go are two dimensional games. The idea to use NN that are used in picture recognition was the key idea. BG instead is only a 1-dimenional game so the achievements of deep learning don't achieve much
>
> - if you take into account what a amount of resources deepmind throws in, they would surely come out with something better, but not washing the floor with any of XG, BGBlitz or GnuBG. The alphago team was 18 people IIRC. When Aja Huang was asked on a keynote how to parallize learning he said that they had other people that did this (so the 18 were only involved with Go). Compare this to leisure time programmers (BGBlitz, GnuBG) or a single full time programmer (Xavier?). The hardware that learned "chess in a couple of hours" is not something anyone of us have. IiRC when you run a fast desktop PC you'll need about a year of training to have the same computing power. Money were no issue for them.
>
> - deep mind always tries to do thing that haven't been done yet.Unless one of their Cxx get's mad on BG they wont do BG.

On December 21, 2021 at 7:54:15 AM UTC-7, Tim Chow wrote:

> More in the AlphaZero spirit would be to train a separate net
> for each match score. Start by training neural nets for DMP,
> and 2-away Crawford. No cube here to worry about. You can
> then figure out what the match-winning chances are at 2-away
> Crawford, by doing Monte Carlo simulations.

I understand the basic concepts of net training barely enough
to talk about them but know very little about technicalities of
various ways of going about it. I this dicussion will mature to
educate/benefit all with the participation of more people.

I didn't know that you could train separate nets and combine
them without starting over. This will make it much easier to
do with home PC's.

Personally, I would be more than happy enough to settle for
cubeful, 1-point money games and cubeless 5 and 7-point
cubeless matches which I consider the "real/classical" BG.

I suppose N-away will also need to be proportionate to the
match length. For example, 4-away in a 5-point match means
you still have 80% more to go but in a 7-point match it means
you only have 57% more to go.

> The first score where you really need to consider the cube
> would be 2-away post-Crawford. If you're willing to cut
> corners then you could cheat and use human intervention
> here, but it would be interesting to see if the neural net can
> figure out the cube on its own. In particular, does it
> play on for the gammon if it's the 2-away player and wins
> the opening roll with 31 and the opponent rolls 62 or 63?

I am proposing "human intervention" only for the puprposes
of poking a hole in the so-called "cube therories" and also
"checker therories" or unified "backgammon therories"...

After that, the ultimate goal is of course to not teach bots
how to play like humans but for humans to learn from bots.
So, no "human intervention" should be allowed at that stage.

> It would be interesting to see a neural net try to figure out
> the cube strategy on its own here.

This is the entire reason behind my arguments on this subject.

> (Of course if you're willing
> to cut corners, you could use the "standard" match equity table
> and jump to 2-away/3-away directly, without training separate
> nets for the smaller scores.) And then obviously, you can
> work your way up the chain of match scores, training separate
> nets for each score.

Again, cubeful or cubeless, I'm not so sure that N-away/N-away
is the same in any match length. I'll need to digest this and may
need to be convinced whether by others or by myself.

MK

On December 21, 2021 at 4:12:07 PM UTC-7, Frank Berger wrote:

> if you take into account what a amount of resources deepmind
> throws in, they would surely come out with something better,
> but not washing the floor with any of XG, BGBlitz or GnuBG.

On the opposite end, I bet that an alpha-zero-bg-bot would run
XG and GnuBG through the garbage disposer. (BTW: BGBlitz is
not in the same class as the other two.)

> The alphago team was 18 people.... Compare this to leisure
> time programmers (BGBlitz, GnuBG) or a single full time
> programmer (Xavier?).

All notable BG bots since TD-gammon v2 are offshoots of it.
So, it's easy for even a single amateur/leisure programmer or
a petty-crook/professional programmer to copy and release
a slightly improved version of a previous BG bot.

The alphago team, on the other hand was trying something
new. And, you may be right that to try something new in BG
bot development, a team of 18 people may also be needed.

But the question here is whether a hole can be poked through
the fancifully elaborate, pretentiously so-called "theories" of
BG. A single person who is willing to go against the dogmas
and who has the technical capabilities needed to do, can do it
with no need for a team of 18 people.

I believe this is brewing to happen right right now, right here.

> deep mind always tries to do thing that haven't been done yet.
> Unless one of their Cxx get's mad on BG they wont do BG.

I think they or most similar people with similar efforts are
driven by the urge to do better. "Debunking bullshit" does, in
fact, gives the greatest pleasure and feeling of achievement.

Personally, I think "they" are avoiding BG because of how it's
*infested* by gamblers.

Tesauro, on the other hand, had sought validation for his bot
from the gamblers and by doing so has done great harm to
BG bot development. Personally, I'm mentioning this opinion
of mine at every occasion to put him to some sort of shame
and guilt, in order to drive him to do something to redeem
himself, which he could easily do with the super hardware
and other resources available to him.

MK

On Friday, December 24, 2021 at 6:11:16 AM UTC-5, MK wrote:

> Again, cubeful or cubeless, I'm not so sure that N-away/N-away
> is the same in any match length. I'll need to digest this and may
> need to be convinced whether by others or by myself.

I'm not sure I understand exactly what you're saying here. If you're
saying that 3-away/3-away is different from 4-away/4-away, then
yes, they're certainly different. That's why I was suggesting training
separate neural nets for different scores. (This is not a new idea;
bot developers have considered it before, but have decided that it
takes too much effort for too little gain. But if you're a purist, it is
certainly the logical way to proceed.)

On the other hand, if you're questioning whether a score of 2-2 in
a match to 5 points is the same as a score of 4-4 in a match to
7 points---they are the same. The history of what has happened
before in the match cannot influence your strategy; you are now
playing a match to 3 points and have to win it.

You said that you would be satisfied with training something to
play a single cubeful money game, but what I'm suggesting is
that the way to approach it is via match play. I mentioned before
some of my colleague's preliminary experiments. If you try to do
this kind of training directly, the unlimited cube values will make
it very difficult for the neural net to learn from experience. I tried
to explain this to you before, but if you don't believe me, the only
way to convince yourself is to try it and see for yourself.

---
Tim Chow

On December 24, 2021 at 8:23:56 AM UTC-7, Tim Chow wrote:

> On Friday, December 24, 2021 at 6:11:16 AM UTC-5, MK wrote:

>> Again, cubeful or cubeless, I'm not so sure that N-away/N-away
>> is the same in any match length. I'll need to digest this and may
>> need to be convinced whether by others or by myself.

> I'm not sure I understand exactly what you're saying here.....
> On the other hand, if you're questioning whether a score of 2-2 in
> a match to 5 points is the same as a score of 4-4 in a match to
> 7 points---they are the same. The history of what has happened
> before in the match cannot influence your strategy; you are now
> playing a match to 3 points and have to win it.

Yes, of course. Even if played for money, a match is a match.
Sorry for wasting anyone's time with this obscure question. I
don't know what led me to it but I was trying to compare in my
mind the difficulty if winning a single 7-point match vs. 2 (out
of 3) 5-point matches.

In the "old world", the standard "quicky" is a 5-point match. If
they have time for it, a more serious challenge is 2 out of 3
5-point matches, all cubeless of course. Often a single 7-point
match is substituted as being almost as good of an alternative
and I always thought that this was done to save time. But then,
I was wondering, how could they be even roughly equal?

So, maybe you guys can offer opinions on this instead: how
hard would it be to win 2 out of 3 5-point matches vs a single
7-point match?

Then, if you want to get more elaborate about it, you may offer
opinions on N-away/N-away in the first 5-point match of a set,
the second match and the third/tie breaker match if it's needed,
compared to the same N-away/N-away in a single 7-point match?

Who cares? Well, the question just occurred in my mind and I
don't know the answer, which I would like to figure out somehow.
Mind exercise to fill free time...

> You said that you would be satisfied with training something
> to play a single cubeful money game, but what I'm suggesting
> is that the way to approach it is via match play.

Am I wrong to think that cubeful money game is what would
bring out the best/most out of the players' "cube skills", if there
is indeed such a skill??

> I mentioned before some of my colleague's preliminary
> experiments.

Yes, but they were just incomplete attemps that failed to show
anything even weakly decisive.

> If you try to do this kind of training directly, the unlimited cube
> values will make it very difficult for the neural net to learn from
> experience.

May be difficult but not impossible, yes? You can tacke this in
small, gradual steps just as you suggested for various N-aways
in match play. Start with limiting the cube at low values at first.
Then gradually increase them and try get as close as you can to
unlimited cube values.

I believe that you will get your answer much before needing to
try very high cube values, let alone unlimited cube values, that
the so-called "cube skill" is not what it is at all...

> I tried to explain this to you before, but if you don't believe me,
> the only way to convince yourself is to try it and see for yourself.

I would believe you if you had gone at least far enough with your
experiments even if you had not been able to complete them.

I already did convince myself. But there is no way for me to
convince you people. That's why I'm trying to make you all
convince yourselves.

I can't offer a replacement for what you guys believe in now.
So, I'm trying to cause credible doubts in your minds, in order
for you to feel a need to renew your faiths and hoping that you
will fail at doing it and instead become heathens as a result...

MK

On Sunday, December 26, 2021 at 12:28:24 AM UTC-5, MK wrote:
> Am I wrong to think that cubeful money game is what would
> bring out the best/most out of the players' "cube skills", if there
> is indeed such a skill??

The conventional view is that match play is a better test of
cube skill because the cube handling varies with the match score.

> May be difficult but not impossible, yes? You can tacke this in
> small, gradual steps just as you suggested for various N-aways
> in match play. Start with limiting the cube at low values at first.
> Then gradually increase them and try get as close as you can to
> unlimited cube values.

That is almost exactly what I was proposing---start with lower
match scores where there is a natural cap on the size of the cube,
and work your way up.

> I already did convince myself. But there is no way for me to
> convince you people. That's why I'm trying to make you all
> convince yourselves.

The reason you can't convince us is because you operate outside
the realm of facts and logic. You convince yourself about how
AlphaZero would perform without understanding the first thing
about deep learning. Of *course* you can't convince others who
actually know something about the subject.

---
Tim Chow

On Sunday, December 26, 2021 at 3:18:29 PM UTC-5, I wrote:
> > May be difficult but not impossible, yes? You can tacke this in
> > small, gradual steps just as you suggested for various N-aways
> > in match play. Start with limiting the cube at low values at first.
> > Then gradually increase them and try get as close as you can to
> > unlimited cube values.
> That is almost exactly what I was proposing---start with lower
> match scores where there is a natural cap on the size of the cube,
> and work your way up.

I realize that I didn't express myself clearly here. If you start with a
money game, but then impose a limit on the cube, that is almost
exactly the same as playing a short N-away/N-away match. (It's
really more like playing for table stakes, but it comes down to
essentially the same thing, with a slightly different match equity
table.) So once you start talking about capping the cube, your
proposal is not materially different from mine.

---
Tim Chow

Tim Chow schrieb am Mittwoch, 22. Dezember 2021 um 04:46:14 UTC+1:
> - There's no reason to bring DeepMind into this discussion. The basic
> ideas are out there, and other people can more or less replicate the
> results, with much less effort than was needed to come up with the
> ground-breaking ideas in the first place.
Because many people in similar discussion expressed the belief that if Deepmind will do BG they will wash up the floor with the current bots, just because they learned chess in a couple of hours (on a fast PC it was about a year of training, if you don't have Googles ressources). I disagree.

> winning rate from the starting position. Surely it's obvious that the
> point is to understand superbackgames and wild containment games and
> other positions which the current bots have less than zero clue about.
I don't agree completely on that. You see that here GnuBG is completely off whereas BGBlitz plays reasonable ( http://bgblitz.com/olympiad_2016.html ) and on Dailygammon (dailygammon.com Discussion search for: threads like Position #25 that are posted by Zorba) a very deep backgame was discussed intensively recently. In some positions XG doesn't find a reasonable move whereas BGBlitz founds it with 1-5 ply. Not always that extreme but for the small sample I have (please send me extreme Backgames and containment games to study) BGBlitz seems not to fulfill "zero clue about". I'm very thankful if someone could send me some positions where BGB is way of.

MK schrieb am Freitag, 24. Dezember 2021 um 12:42:27 UTC+1:
> On December 21, 2021 at 4:12:07 PM UTC-7, Frank Berger wrote:
>
> > if you take into account what a amount of resources deepmind
> > throws in, they would surely come out with something better,
> > but not washing the floor with any of XG, BGBlitz or GnuBG.
> On the opposite end, I bet that an alpha-zero-bg-bot would run
> XG and GnuBG through the garbage disposer. (BTW: BGBlitz is
> not in the same class as the other two.)
Too bad we won't see it. I'm quite sure that Deepmind would only invest resources, if they see a possibility for a breakthrough.
Naturally I would be very curious if they do it. If you see what AlphaGo does with Stockfish than maybe Backgames are a much better game plan :)

> > The alphago team was 18 people.... Compare this to leisure
> > time programmers (BGBlitz, GnuBG) or a single full time
> > programmer (Xavier?).
> All notable BG bots since TD-gammon v2 are offshoots of it.
> So, it's easy for even a single amateur/leisure programmer or
> a petty-crook/professional programmer to copy and release
> a slightly improved version of a previous BG bot.
I'm not sure. there are a lot of small tweaks to improve and not every idea does work, so to competet it might take some work.

> Personally, I think "they" are avoiding BG because of how it's
> *infested* by gamblers.
As I mentioned, they probably don't see a real breakthrough. Let's say that they do a bot that has a PR of -0.5. If they do a real match and they don't have a playing style to do backgames all the time and confude the opponents they may easily loose (dice, you know). That doesn't seem a good invetsments for a 6 to 7 digit investement.

I don't understand the "gambler" thing. BGBlitz learns through selfplay, I don't know where there is influence.

On 12/30/2021 6:48 AM, Frank Berger wrote:
>> winning rate from the starting position. Surely it's obvious that the
>> point is to understand superbackgames and wild containment games and
>> other positions which the current bots have less than zero clue about.
> I don't agree completely on that. You see that here GnuBG is completely off whereas BGBlitz plays reasonable ( http://bgblitz.com/olympiad_2016.html ) and on Dailygammon (dailygammon.com Discussion search for: threads like Position #25 that are posted by Zorba) a very deep backgame was discussed intensively recently. In some positions XG doesn't find a reasonable move whereas BGBlitz founds it with 1-5 ply. Not always that extreme but for the small sample I have (please send me extreme Backgames and containment games to study) BGBlitz seems not to fulfill "zero clue about". I'm very thankful if someone could send me some positions where BGB is way of.

Well, there are some standard candidates. Like what does BGB
think the cube action is below? Or how about, can BGB roll a
prime of 15 checkers all the way around the board from the
opponent's home board to its own home board?

XGID=-eEeEeE-------------------:1:1:1:00:0:0:0:0:10
X:Player 1 O:Player 2

Score is X:0 O:0. Unlimited Game
+13-14-15-16-17-18------19-20-21-22-23-24-+
| | | |
| | | |
| | | |
| | | |
| | | |
| |BAR| |
| | | X O X O X O |
| | | X O X O X O |
| | | X O X O X O | +---+
| | | X O X O X O | | 2 |
| | | X O X O X O | +---+
+12-11-10--9--8--7-------6--5--4--3--2--1-+
Pip count X: 60 O: 330 X-O: 0-0
Cube: 2, X own cube
X on roll, cube action

Analyzed in 4-ply
Player Winning Chances: 89.46% (G:88.08% B:29.87%)
Opponent Winning Chances: 10.54% (G:0.00% B:0.00%)

Cubeless Equities: No Double=+1.969, Double=+3.937

Cubeful Equities:
No redouble: +2.019
Redouble/Take: +3.848 (+1.829)
Redouble/Pass: +1.000 (-1.019)

Best Cube action: Too good to redouble / Pass
Percentage of wrong take needed to make the double decision right: 35.8%

eXtreme Gammon Version: 2.19.207.pre-release

---
Tim Chow

Tim Chow schrieb am Donnerstag, 30. Dezember 2021 um 14:33:11 UTC+1:

> XGID=-eEeEeE-------------------:1:1:1:00:0:0:0:0:10
not yet. It does understand to roll home the prime, but it does not yet understand to slot at the front of the prime. You may try it yourself.
And between getting the most extreme position perfect and totally lost is a continuum. At least my definition of totally lost is not met, your mileage may vary.
And at the Kauder paradox we as Humans have an edge but what about XGID=---a--A--BBB--ABa---BbAaAA:0:0:1:41:5:6:1:7:10 ? XG has, after 15 hours of rollouts b-24,10-6 as first and the obvious(?) b-24,20-16* only on pos 6, -0,235 worse. BGBlitz got this move right on 1-5-ply. (see the discussion and more data here: http://www.dailygammon.com/bg/forum2/main/read/56208 ) . But I think this not so difficult for humans.

But what about that XGID=---A-Aa--BBCA-----A--d-BB-:0:0:1:21:5:6:1:7:10 ? I wouldn't trust anyone who "knows" what is right here....

On 12/30/2021 6:29 PM, Frank Berger wrote:
> Tim Chow schrieb am Donnerstag, 30. Dezember 2021 um 14:33:11 UTC+1:
>
>> XGID=-eEeEeE-------------------:1:1:1:00:0:0:0:0:10
> not yet. It does understand to roll home the prime, but it does not yet understand to slot at the front of the prime. You may try it yourself.
> And between getting the most extreme position perfect and totally lost is a continuum. At least my definition of totally lost is not met, your mileage may vary.

What does BGBlitz think the cube action is?

> And at the Kauder paradox we as Humans have an edge but what about XGID=---a--A--BBB--ABa---BbAaAA:0:0:1:41:5:6:1:7:10 ? XG has, after 15 hours of rollouts b-24,10-6 as first and the obvious(?) b-24,20-16* only on pos 6, -0,235 worse. BGBlitz got this move right on 1-5-ply. (see the discussion and more data here: http://www.dailygammon.com/bg/forum2/main/read/56208 ) . But I think this not so difficult for humans.
>
> But what about that XGID=---A-Aa--BBCA-----A--d-BB-:0:0:1:21:5:6:1:7:10 ? I wouldn't trust anyone who "knows" what is right here....

This is my point. There's still plenty of room for improvement
in backgammon if you allow for "prop" play. Choose a position
and play 100 games as White and 100 games as Black, and see who
comes out ahead. It's totally possible that a new bot could
crush current bots under those circumstances.

---
Tim Chow

Tim Chow schrieb am Freitag, 31. Dezember 2021 um 05:13:27 UTC+1:
> > But what about that XGID=---A-Aa--BBCA-----A--d-BB-:0:0:1:21:5:6:1:7:10 ? I wouldn't trust anyone who "knows" what is right here....
> This is my point. There's still plenty of room for improvement
> in backgammon if you allow for "prop" play. Choose a position
> and play 100 games as White and 100 games as Black, and see who
> comes out ahead. It's totally possible that a new bot could
> crush current bots under those circumstances.

100 Games each side is probably not enough when it's close, but I think the classical Kauder Paradox is an excellent Guinea pig to test the understanding of Containment games and outfield primes and maybe even 50 games each side will be enough to get a interesting number. I'll do that in 2022 with a freshly breeded AI (March? April? My schedules are always ridicously optimistic) and I'm very curious. The good thing is: we know the equity numbers exactly and one side is trivial to play, so we have a measure of understanding (equity in the starting position) and how it handles it (final result)..

Too bad that so few people are interested in BG-AI compared to e.g. chess :( but probably there are some reasons for it.

On December 26, 2021 at 1:18:29 PM UTC-7, Tim Chow wrote:

> On Sunday, December 26, 2021 at 12:28:24 AM UTC-5, MK wrote:

>> Am I wrong to think that cubeful money game is what
>> would bring out the best/most out of the players'
>> "cube skills", if there is indeed such a skill??

> The conventional view is that match play is a better
> test of cube skill because the cube handling varies
> with the match score.

I have two counter arguments.

First, based on the logical arguments I repeatedly
provided here, the cube will go up to the match
length during the first game and perhaps during
the first few rolls of the first game. Only two, three
or four cube actions is all needed for the cube to
go over match lengths of 7, 15 and 25 points.

Second, what about beavers, raccoons, etc.? Don't
they also require better cube skill? Why aren't they
allowed in match play? The obvious answer is that
even fewer matches will get played past the first
game than in my above argument.

>> Start with limiting the cube at low values at first.
>> Then gradually increase them and try get as close
>> as you can to unlimited cube values.

> That is almost exactly what I was proposing---start
> with lower match scores where there is a natural
> cap on the size of the cube, and work your way up.

That's not the same thing. Most short matches will
by end in the first game. You'll need to go up to very
long matches before you can call them "matches",
i.e. "lasting more than one game". By that time, you
will practically be playing very short sessions of
money games.

>> I already did convince myself. But there is no way
>> for me to convince you people. That's why I'm trying
>> to make you all convince yourselves.

> The reason you can't convince us is because you
> operate outside the realm of facts and logic. You
> convince yourself about how AlphaZero would
> perform without understanding the first thing
> about deep learning. Of *course* you can't convince
> others who actually know something about the subject.

No, the real reason is that you can't or don't want to
understand my arguments that are indeed based on
facts and logic. The problem is that you guys are in
denial of them. You are "believers". Debating with you
gets as tiring and useless as arguing with creationists
who insist that dinasours could fit in Noah's Ark; well
at least baby dinasours or maybe even dinasour eggs...

I don't need to understand the detailed mechanics of
"deep learning" in order to predict what the results are
likely to be. If the machine can't learn by itself without
human interference that would inject human bias, let
it be so. Unlike you, I'm not agreable to "tweaking" it.

Also, unlike my being willing to accept what AlphaZero
bg bots may come up with, you are not willing to give
up the existing inaccurate formulas, MET's etc. and can
only conceive that AlphaZero bg bots will only be able
to improve upon the existing garbage, rather than starting
over from a clean slate.

MK

On December 26, 2021 at 1:23:24 PM UTC-7, Tim Chow wrote:

> If you start with a money game, but then impose a
> limit on the cube, that is almost exactly the same
> as playing a short N-away/N-away match... So once
> you start talking about capping the cube, your
> proposal is not materially different from mine.

I proposed it only as a temporary compromise to get
you started to gradually solve your difficulties of deep
learning, with the ultimate goal being unlimited cube.

MK

On December 30, 2021 at 5:02:20 AM UTC-7, Frank Berger wrote:

>> Personally, I think "they" are avoiding BG because of how
>> it's *infested* by gamblers.

> As I mentioned, they probably don't see a real breakthrough.

How can anyone see a breakthrough without looking?

Even the developers of existing bots may be interested
in pursuing a different way if they could be convinced
that the current way is flawed. And that's what I'm trying
to do here: to raise doubts in your minds, enough to run
my proposed "compromised/limited" experiments which
would be trivially easy to do but may be effective in poking
a hole in your current bg dogmas.

It would be like finding a loose thread on a sweater to pull.
After that, the entire current "bg theories" will come undone
and you all will need to come up with something better by
necessity.

> Let's say that they do a bot that has a PR of -0.5. If they
> do a real match and they don't have a playing style to do
> backgames all the time and confude the opponents they
> may easily loose (dice, you know). That doesn't seem a
> good invetsments for a 6 to 7 digit investement.

This is exactly you guys' problem. You can't see past what
you have now. PR is only meaningful within the scope of
the existing pile of garbage. Otherwise there is no use for
it for any other purpose. You won't be able to measure a
fondamentally different bot by how current bots rate
themselves through masturbation...

The purpose of a new bg bot would not be to find ways
to beat the current bots which have obvious flawa that
can be exploited but it would be to make progress towards
a universally optimum bg play.

So, yes, I agree that simply beating the existing shitty bot
would not justify "6 to 7 digit investement". It would take
a higher ideal.
> I don't understand the "gambler" thing. BGBlitz learns
> through selfplay, I don't know where there is influence.

You are measuring its PR against other bot and human
players of "gamblegammon" world, aren't you?

MK

On December 31, 2021 at 8:38:51 AM UTC-7, Frank Berger wrote:

> I'll do that in 2022 with a freshly breeded AI (March? April?
> .....
> The good thing is: we know the equity numbers exactly

Ha ha! No less than "exactly" at that. You are hallucinating.

Here are a few friendly suggestion for you to make your
bot more useful and desirable, even if not the strongest.

Don't waste time trying to improve it as it exists. Small
improvements won't win over any devout bot-kissers.

Even if you are not interested in developing a new bot
from scratch, you can add simple features/tools to your
existing bot to make it uniquely useful.

Make all contants used by your bot selectable by the user,
such as your selectable various cube strategies. People
are curious and like experimenting with things. Allow them
to use your bot to do that.

I'm not sure if your bot can play against itself already? If
not, enable it to do that and with each side's settings being
separate, so that a user can make your bot play against
itself at different skill levels, cube strategies, etc. in order
to see the difference.

While at it, add the capability for it to play against other
bots. I don't know if Dueller was open source but it was
already capable of having JW, SW and old Gnubg play
against each other. Axel has a tool to make the latest
Gnubg play against itself. Hopefully you can get some
help from others to not reinvent the wheel all over. But
even so, it will be worth your effort. You bot playing a
few selected odd positions better than other bots isn't
enough to convince anyone.

With these features added, even I would buy it and use.
In fact, I may also promote it.

I took the time and effort to give you hopefully helpful
advice in goodwill but I doubt that you will appreciate.

> Too bad that so few people are interested in BG-AI compared
> to e.g. chess :( but probably there are some reasons for it.

If you yourself aren't interested in doing anything different,
why should anybody else should do? BG has become too
much focused on gambling. There are no amateur bg events
like "bg olympics".

Over the years, there were times I really wanted to play in
tournaments to measure myself against others but I just
didn't want to shake hands and mingle with a bunch of
sick gamblers... :(

MK

On 1/5/2022 11:47 AM, MK wrote:
> Debating with you
> gets as tiring and useless as arguing with creationists
> who insist that dinasours could fit in Noah's Ark; well
> at least baby dinasours or maybe even dinasour eggs...

So why do you do it?

I debate with because I find you hilarious. But you don't
find me hilarious.

---
Tim Chow

On January 5, 2022 at 7:37:02 PM UTC-7, Tim Chow wrote:

> On 1/5/2022 11:47 AM, MK wrote:

>> Debating with you gets as tiring and useless as
>> arguing with creationists who insist that dinasours
>> could fit in Noah's Ark; well at least baby dinasours
>> or maybe even dinasour eggs...

> So why do you do it?

Partially because others read what I write to you here.

Like Axel, for example, who has at least a "cracked open"
mind to run experiments similar enough to what I have
proposed.

It doesn't even matter if he or others can convince you.
Why do you think you matter that much to anybody??

BTW: was Axel your secret colleague who had taken a
shot at developing a Alpha-Zero Hypestgammon bot?

I'm beginning to believe he was and he aborted that
project because he had already seen where it would
end but didn't want to face it.

He may be in the same predicament with his latest
experiment with "Murat mutant". I see four possible
reasons why he hasn't yet announced his findings:

1) He is still genuinely trying to do the best he can
to find out the truth about "cube skill".

2) He has already seen where an honest experiment
would end at and he keeps trying to tweak it, hoping
to achieve the results that he wishes for.

3) He has completed his experiments but he can't
bring heimself to accept the results and publish to
the rest of us.

4) He has already seen where an honest experiment
would end at and conveniently aborted it so that he
can keep claiming indefinitely that the "jury is still out".

But no worries folks. It's just a simple matter of time...

> I debate with because I find you hilarious. But you
> don't find me hilarious.

No, I find you a miserable, pathetic idiot with inferiority
complex. Educated/trained/"indogtrained" doesn't mean
intelligent.

As of 2020, an estimated 98.2% of the world population
believe in some sorts of gods/creator, hells/heavens,
angels/spirits, etc... (interestingly and unexpedtedly up
from 98% in 2010 and 95.5% in 1970).

As a believer in statistics, I would bet that you are among
those 98.2% insect brained creatures who believe in all
sorts of illogical shit.

Yet, you have the audacity to request "logic" from me!?

You couldn't recognize logic if it hit you on the forehead. :(

I don't but could find you hilarious depending on your
answers to some questions if you want to fuck with me.

Do you believe Noah had dinosaurs in his Ark, even if
baby ones or even if just eggs?

Do you think Jessus was a bastard born child, out of a
slut/whore who delivered him in secret/shame in a barn
witnessed by donkeys, sheep, etc.?

Do you think Jesus grew up to become a faggot who
never married or had offsrings?

I haven't invented these questions which are only a few
that have been going around for millenia. But I thought
I could use them to see if I can get anything hilarious
out of you just for the fun fuck of it... :))

MK

On 1/6/2022 6:22 AM, MK wrote:
> On January 5, 2022 at 7:37:02 PM UTC-7, Tim Chow wrote:
>
>> On 1/5/2022 11:47 AM, MK wrote:
>
>>> Debating with you gets as tiring and useless as
>>> arguing with creationists who insist that dinasours
>>> could fit in Noah's Ark; well at least baby dinasours
>>> or maybe even dinasour eggs...
>
>> So why do you do it?
>
> Partially because others read what I write to you here.

So it's not useless after all? You contradict yourself.

> Why do you think you matter that much to anybody??

I don't matter to anybody. What gave you the impression
that I think that?

> BTW: was Axel your secret colleague who had taken a
> shot at developing a Alpha-Zero Hypestgammon bot?

No.

---
Tim Chow