On Thu, Aug 20, 2009 at 20:10, Michael Petch <address@hidden>
This post doesn’t represent that of any other developer besides myself, and doesn’t necessarily reflect the views of the others.
There are things that you have said that are correct and others you draw conclusions on based on no evidence provided here. In some cases I will ask questions and then
SO you assert,on no evidence provided here. To this point. Reorder your exposiiton properly
if you expect to make such a cliam and them commit the same offense:
even if you rectify it lower down.
Crowing in advance of proof is a form of behavior that leads to incorrect results.
follow up based on your responses
“The interesting thing to note is that GNUBG also cheats.”
Okay, so you are going to prove that to us in your post. So lets not comment on this
No. I _asserted_ that in my posts, and was very specific in delineating how
completely honest code
well, written, bully documented
completely open in intention by the developer
with a completely benign function (to play backgammon extremely well)
has an unforeseen side effect or consequence 9apparent "cheating")
in subsequent program behavior
that is not predictable or antcipatbale by the game player and/or potentially the experienced
which in turn has as a consequence a characteristic in the "behavior" (note the quotes,
placed carefully, as an analogical meaning is being used: programs do not have behavior
in ANY psychological sense, no matter how well the program "mimics" or emulates it
(again, note the quotes carefully), and no matter how much modern researchers wish to
conflate or hold the two as "the same" or "fully equivalent" (they are not):
the usage is useful only in drawing a working analogy that exposits key features and criteria
(and only when it does do that)
that in turn illustrate a particular point of "behavior" not yet recognized as such
by the players or developers.
Or the society allowing the game to exist (note antisocial behavior at times stimulated and
reinforced by mass media mechanisms, from the 17th century on forward: a reall
concrete repeatable scientific psychological result.
(Just not in all the fear tactics-inspired politically inspired nonsense arguments about
exactly which ONES _do_ cause such behaviorally anomalies.
Much less about those that are INTENED to stimulate _exactly_ such a result
such as Hashish and harem in the society of assassins
such as Omerta or making your bones in La Cosa Nostra or Mafia
or "bonding ot the prophet" as practiced
by many fanatical zealotry based religious (dis)orders
or being "stooled tot he rogue" (society of thieves, apocyphally)
and so forth. These DO exist; it iis simply the case that much of the time
the circumstances CLAIMED to create such effects are NOT so:
the specific case cited does NOT cause such an effect.
You just have a fall guy to take a pot shot at.
yet. However, to declare the bot a cheater you would have to prove that the bot knows
No, I would not. But that particular discussion is off my point.
I declared the overall program PLUS THE DATABASE
(when run against an oppoenent! to be more saliently accurate:
you need to become more so yourself; and _read- More carefully)
a "cheater" (note the quotes again: analogical meaning in usage)
when playing against a _human_ specifically,
because the information content being used lies outside the span of the rules of the game.
I note the first was explicit; the second was implicit.
I also thought when you started your argumentatum that the second point had to be
inherently obvious to anyone who plays the game.
something that the human can not POSSIBLY know to make a decision.
NO: something the human was not expected TO know:
specifically the only way a game can be said NOT to cheat is
when ONLY the rules of the game itself and NO information outside of that
are/is used to draw a conclusion about what will "win" the game.
Trying to state that knowledge of a Mersenne twister
or any other PRNG or claimed TRNG
is in ANY historical sense
a part of the game of backgammon is well past idiotic:
it is prevaricative, and provocatively so.
That does NOT mean the _developers_ are such: they can be (and are in my opinion)
ENTIRELY of good will in this advent. They just a times may not have recognized
what actually IS going on at _some_ times within the play of GNUBG.
And all other NNP engines.
Also not all that unusual. Sometimes the originators of an idea that WORKS CORRECTLY
as describes may be the LAST to see where it also
WORKS IN OTHER WAYS as unforeseen.
By the way: this is the THIRD time i have posted this argument ot this group.
You are not the original responder.
You said: “The engine is scrupulously honest and there is NO hidden trapdoor in the code.”
I give you this point. Agreed.
You're not giving me a thing.
I presume you are using ordinary English idiomatic usages,
And do not have a problem with that. BUT
so far in your approach to this interaction
you appear to be less than clear on the difference between that and clear precise reasoning
using that same English language.
Until that becomes clear to me I have to denote when emotionalistic phrases come in that may "tilt" your reasoning.
You said:“What I am denoting is that the NNP "learns" the attributes of the specific roll generator you use”
More specifically _patterns_, not just an average statistical _bias_.
The two are entirely separate problems.
especially when a _pattern_ can be used against a _lattice-manifold_state_space_
whichw e call "winning a game in backgammon"
to shovel probaibltiies around
in ways that you (and I!) may not foresee
or even be able to foresee
or possibly even _capable_ of foreseeing
or in some circumsatnces not even capable of _learning_ to foresee them when given exactly the same inforation state as the program and the database in toto.
And THEN I stated very specifically that the SAME thing is true of the NNP state space of the human biological player, in that semiotic systems have a veyr complex set of patterns.
And I suggested that you might try to get a couple of world class players who have their own private set of "tells" to TELL you what those tells are, and get their complete and open opine about what the "tells" are
as a way to see if the assertionsa re in fact that case (the ones I made).
because, frankyl speaking:
most world class players do NTO always give you all thepointers they HAVE got.
For a very good reason: relative playing backgammon at that level:
they are worth quite a bit.
And money, while being an important part of THAT level of game play
is NOT the most precious part of it. Rather, the least.
This is true to a degree. If there is a bias in the PRNG with enough training the bot could inherently indirectly learn based on taking advantage of the bias. I mean lets say I create an PRNG that deliberately rolls a bias towards double 3, and train it with a billion (some arbitrarily high number) it does reason that the weights file could be skewed based on the biased view of randomness that was fed into it. So lets say for the sake of argument you are correct, and that Gnubg learned backgammon based on skewed dice and ultimately had a weighting file generated out of that bias.
You have seen the entry gates.
Now look at what is behind door #1-n out of a few million more:
1) take a picture in your mind's eye of the TREE of play within a game of backgammon.
I doubt that you can correctly predict ALL of the time when ther is a "pattern" in the roll
that con force a human gamer to errantly miss a play of equivalent probability at depth 6 or
or more which, in turn progrssively forces the opponennt player ever more porbably into
successively less stable situations, where the roll of the GNUBG DB selcts the more
In short, you not nly have tollok at the the FIRST moment (average probability)
and other less frequent moments (deviation, R value, etc.)
you have to look a the stochastic processes involved against the STATE spaces being
convolved. Or you will miss the "drift" involved.
I did an argument in progressive thirds that most 18th century gamblers would recognize
(stake, fold, or hold).
against a mere 3 level progressive analysis (not even to a 1% probability shift level).
No one got the point.
And for you to state it would take a billion training events to establish a bias that would be EFFECTIVE IN CHEATING has no basis in this argument: merely a counter assertion.
And that ain't proof.
Uou will have ot reaosn more clearly than that.
You said:“And does it probably better than a human can.”
Yup. Against the "patterns" inherent in a
_repeatable_result_state_space_ (PRNG Mersenne OR ANY other)
Are you telling me that you know someone who can perceive the difference (and statistically validate the claim) without the aid of software? I can tell you now
Yup. Exactly and only those who can.
Until you watch someone play at that level you ain't got a clue.
And: in order to be effectively at that level _apparently_ in wining against such a "cheat"
you would have to be able to see the pattern well enough to detect AND _circumvent_ the "gambit" that GNUBG is "playing" at the moment (snce the program does not consciously
“probably” is an understatement. In fact I think you may overestimate the humans
I acknowledged an estimate. And maintain its validity agains tthe game:
humans have a lot more ability than many anticipate.
You are not acknowledging that your estimate of my estimate is less well based.
On evidence in _this- statement, since that was the basis you started this on.
ability to recognize a pattern (by themselves) in most modern day random number generators (An there are a number of them now). It should read “And does it better than a human can”.
Nope. Noot until you show a mathematically given proof.
Or at least outline a proof process that can be objectified.
You are inverting your argument direction here: probably a type of rhetorical gambit,
conscious or unconscious:
by revrsing your field of play, you confuse the issue as to the validity ogf the other side's
argmunet, and can claim half corectly that you were not taking EITHER side.
(An oversimplified analysis, but I lack the patience to spell out all of the parliamentarian
tactics that can be involved: it gets very boring doing so).
You said:“It will also learn the average hidden "tell" in the pattern of a human player; and the full DB in all likelihood (I so assert) has several folded state transition spaces covering the panoply of the subtypes of human play available. (given enough games rolled in).”
This is true. The neural net wasn’t entirely trained by the bot playing against itself. World class players also manually trained it. Inherently GnuBG has been given to some degree knowledge of world class players (and their playing style). Had the best “Backgame” player in the world trained enough “back game” positions into the bot, the neural net would be better in that regard. But based on all available data (both self learning and from world class players) the bot is biased towards believing that backgammon is best played as a “race”. You could call the integration of World Class
And: it is os. IN al but end game positions, due ot the limits of eeven world class players in "tell"ing what the EXACT best play would be sttaically in a predetermiend pure probability way,
on simply and only the basi fo what is the maxcimum probabvility _win_.
But, unfortunately, the BIAS inherent in world class players play is that they do NOT do this:
their play is BIASED by CORRECT assessment relativ to the cube versus posiiton ina tournament pay against the syle of betting pool present AND ITS MAGNITUDE as assessed by the relative reisk factors assessed BY THAT PLAYER AND THE FRAME OF MIND DURING THAT STYLE OF TOURNAMENT PLAY. .
Which is NOT the average best probability move in apure uniform distribution sense.
As to exactly what that distribution IS:
if I knew it and could program it, I could beat GNUBG
.... and the world class players.
AND TEACH HOW TO DO IT.
Which THEY would not like very much.
Can you do that? When you can, demonstrate it by training people that way.
In mean mean time, the "tells" present that GNUBG has picked up from "watching" world class players play ARE most effectively used in a non-limited-move-situation (non-end-game, roughly equivalently)
as a _race_ strategy, to attempt to "force" the world class (and as a subsumed dominated subset, the average players that world class players easily routinely beat)
OUT of the phasing lock/sync step that they gain by forcing the OTHER player into UNstable positions agains a STABLE roll structure on any kind of patterned lock step basis.
OR ANY OTHER PATTERN "RECOGNIZED" BY GNUBG.
player training and subsequent absorption into the neural net to allow a bot to get a “tell” on the way the human may potentially play.
So I don’t dispute this either. But here is something I will say. It is not cheating to “exploit” the knowledge that it has learned about how humans play. I bully weaker
If the knowledge is outside of the rules of the game:
yes, it is. In the strictest sense that you are NOT paying the _game_:
you are playing the _opponent_
and as such, THAT information i soutside the nature of the game.
This also includes natural dice rols:
THEY ARE BIASED. THEY ARE NOT FAIR.
So, when an _average_ player CORRECTLY denotes that
GNUBG is not "fair", it is "cheating"
when it DOES play by ACTUALLY 'fair' dice rolls
the avergae player is CORRECT and you are WRONG:
from an information-theoretic point of view, it is "cheating".
I note that to gully effectively play the game, you mUST learn hoew to do BOTH:
but that is not the point I raised.
When the claim is made that GNUBG is ONLY Playing backgammon FROM THE RULES:
that is an actual lie: becasue you admit of it below. It is nOT doing so.
It IS playing the _player_. AND the _dice_ . AND the PRNG AGAINST the player.
Pleas note: I actually stated two type of "cheating":
one against the rules in the game (where dice rolss are probabilistically uniformly "fair"
and one against the expectation of the hisotrical humans playing against other humans
The one is contradictory to the other because THEY ARE NOT THE SAME TYPE OF
Just strongly related scenarios.
You have to keep the two straight as BEING different to corectl reaosn about this.
opps with cubes because I know some people have tendencies to play them badly. I capitalize on that to increase my chances to win. That’s not cheating, that’s just playing
Until GNUBG is trained on playing the cube, whether your point is acurate or not: it is not my point.
to the strength and weaknesses of the opponent. So on this point, if the bot has an inherent tell on human play, it doesn’t make it a cheater. So I move on.
Sorry, if the "inhehrent' tell is outside the information set implicit in only the rules of backgammon:
Yes GNUBG _plus_ the database is "cheating" making use of information that is nOT in the rules presented to the human player of the game as being a part of the rules of the game being played.
And it is a double hand parliamentarian dialectical maneuver wrapped ina rhetorical gambit for you to try to hide it in plain sight.
And NOT a problem with GNUBG or its developers or the people who play the game with GNUBG being ignorant of real world tactical and strategic considerations a a warfare game.
Just on the part of your reasoning and argumentatum.
Bunk, hohokum, and actually a direct lie on your part.
You just wan the part of the "tells' you have that you are NOT telling about to stay that way.
So, i move on.
You said:“You can tell alot about each player by their wn pattern;”
Agreed. See above. If the bot was never trained with manual training (from humans), then the bot will not have any tells on humans really play/perceive the game. Knowing your opps playing patterns is not cheating.
“and a true expert will be able to infer more about one player by how they react to another than you might believe.
Try again. Again, you miss the implicit inherently obvious point:
The most easily _recognized_ source of human "tells" is thei rdirect play.
It is NOT the only source.
Just as the NNP program of GNUBG will; pick up ANY information present:
so also does the real world.
And, as such, information travels from the game of backgammon (as well as the scenario it is played in, called "the human experience") back AND forwards
into, out of, through, and around
the state space you might call
where the developer of the software
plays the game of
playing the game of software development
and thus into the state space of the GNUBG database
EVEN WHEN MECHANICALLY SELECTING THE GAME PLAYS MADE
(even without world class player human input).
It WILL extract information from THAT set and collimate it.
Probably to POSSIBLY better than a human will in the same number of learning moves
or temporally time real world seconds of experience.
You said: “Basically, _everything_ goes into it. “
Eveyrting that is inherent in the phasic state space and catastrophe maifold spaces
implied by the actual real world scenario currently then present. You bet.
In what fashion, how organized and whether it will have anything to do with playing
is another topic.
For you to asert that it will NOT is NOT another topic:
you are committing the same error you accuse me of making.
Prove that it does NOT. Or accept that it MAY be so. And that I have _asserted_ it.
And likely, in the Venn diagram cleave of the real worrld of
(backgammon only info, rest of the world's info already present)
(will it help win, will it not help win)
I assert (and you would be a fool to make a bet against me here)
that ALL FOUR QUARTILES are in FACT _not_ empty.
the probability analysis against any real physical world state space is so close to zero
on THAT point of view, which is YOUR argument
as to be
(Please note; I just avoided falling intot he same trap you do:
probability/statistical inference is NOT causal implication; not ever:
even when it leads to wining a bigger share of the pot of a warfare game)
This is way too general. What goes into it is any knowledge that can be directly
Nope, it is intentionally specific to exactly the point I was raising.
measured and learned from. The direct inputs can be used, and bot alter its strategy
Nope. What goes into it is just that, when _trans_ferred to the zGBUBG_BGrules_DG game
and alos whatever ELSE, directly or indirectly, actually IS _tans_ferred in at the same time.
You cannot wipe out the smell of garbage in the food
simply because you did see it
because you did not LOOK for it
when you cooked the meal.
It is there, or it is not, depending on whether it WAS there or it was not.
And I have _asserted_ that it is VERY hard to get ALL of that out.
Care to take a bet, based on the Venn diagram argument above?
You would lose again.
Just possibly not ina n adversarial warfare game, as opposed to an actual rational argument based on mathematics, rwasoning, scientific application, and actual empirical results.
Oftimes barbarians with weapons (even just words as weapons) will "defeat"
the rational,ethical, scientific parties involved.
Until the civilization involved crashes from the damage involved.
based on additions/changes in the NN weights. It is possible for the bots to inherently garner knowledge and discover patterns that we may not even think about.
.... along with what can be inferred from all that INdirecltly TRANSferrd information that you attempted to rule out of the discussion as not being there.
You said: “And whether you fold the state space down from the infinite manifold caleed reality
into a mere 100,000 trained transition nodes”
Yes, but those limited number of trained nodes have to represent all of the data from the known universe (The data fed to the bot). Compared to the amount of data actually processed a lot of it has been thrown away as noise, in coming up with a final engine, and weights file. However some patterns may inherently learned and become part of those weights. That recognition is not in itself cheating (I know where you are
Sure it is, if it is not in just and only just the rules of backgammon the game itself.
By the way: the state space is way big enough.
How many states in a _1_ node transition network based on a 32 or 64 bbit number?
Quite a few. Take that to the 100,000 power and oyu have an idea of the power present in an NNP network of trivial size.
going with this, I will defer commentary until later on)
You said: “_everything_ goes into it.
And _everything_ in it that -can_ come out __will__”
Not necessarily. Even if there are patterns the bot may not have measured them as anything that affected the outcome and weights. Some will of course. Not all data that goes in necessarily was evaluated to be useful by the bot. We technically don’t know what hidden (and implied) knowledge is lurking in the weights file as seen by the NN.
UYou ignored the statement. If it CAN come out, then there IS some scenrio where it MUS come out OR ther is NO way that it CAN come out.
All you have done is sleect a carefully pruned SUBset of the WAYS it can come out
and proceeded on the (hidden in plain sight) assumption to prove
that it can nOT come out inthe scenarios that you set up for it NOT to come out.
Example: When the stock market crashed from programs that incorrectly bid into an unstable position,
the USG and SEC changed the rules to require human intervention before the trades were made.
But did that solve the problem, or only force the trqders into a position
where the trading programs
were actually training the TRADERS to _obey_ the most effective trading program?
Tell me which was the result?
I snipped a section out because it doesn’t add much to what has been said. We start
Or perhap it adds exactly the salient point you wish to ignore; I will not backtrack here.
You could have ignored it and left it in for other sto observe:
why didn't you?
getting into the heart of your argument here though:
The "heart" of my argument was in the first paragraph. The rest was elaboration and clarification.
You said: “then take it on faith that the program involved has found aparticular _weakness_ inherent in the play of either the human game platers loading the NNP state space or a hidden pattern in the PRNG series being used.” and previously you said “You said: “The engine is scrupulously honest and there is NO hidden trapdoor in the code.”
Yep. Both are correct.
And the two information spaces ARE DISTINCT and BOTH are available to the program PLUS the database PLUS the act of playing the game.
So lets say hypothetically you are right for argument sake. That Gnubg happened to stumble upon its limited weights file implicit understanding of some perceived bias
How many states are present in the weights file? Give that answer in full
and assert that it is NOT the case.
You cannot, because the reasoning involved would blow your argument out of the water.
I would estimate that probably 55% and possibly 75-80%of the "skill' set needed to win
consistently at world class at backgammon
is alreayd inherent in ordinary human experience.
And as suc the INVERSE is true:
ANY stae space comprises 80-100% of potential knowledge of backgammon
will cover 75-80% of THAT experience (giving 55-65% of human experience)
Not that GNUBG is that much of an expert. Possibly only 5-10% of that.
Which STILL is an impressive amount of human experience in that state space.
Albeit not ideally encapsulated, represented or activated by programs
that allow ACCESS to it in any form other thna playing backgammon.
(patterns) in the random number stream. Lets just say IT IS the case (I can’t prove it isn’t), but what I can say is this. There is no backdoor as you stated earlier, and the bots
Why can you not prove it? Is it because it cannot be disproved, since it is true?
Or some other reason?
NN isn’t unscrupulous. I can tell you that no inputs into the NN (as you probably are
I did NOT say the _bot_ was NOT unscrupulous.
I DID say that the developers and open _code_ were scrupulously honest
in the sense that there is no hidden trapdoor.
I _assert_ again
the bot and the program are entirely RUTHLESS
and not TRUTHless
but RUTHLESS: there is no mercy or backing off on the part og the bot.
It plays all out the way it was programmed to, every time.
And so, if you SET the goal to WIN< it WILL learn to WIN
assuming the developers were successful in encapsulating that imperative.
I think they were, fairly so, without hesitation, a compliment.
Does that mean the bot is scrupulously honest? In trying to win, yep.
By whatever means it CAN do so. Yep.
By the rules as stated for the JST and only just the gameof backgammon?
Hoo. Guffaw. laugh.
You need to understand that you may not understand.
already aware) include the source of the randomness. So all knowledge of any bias is inside the NN/Weights.
Yes, in a limited portion of the sense you are stating it.
In the sense that:
the game code ITSELF is an information manifold space:
it IS a transition network. A map. (Any Von Neumann machine is: look it up) .
So it ALSO has information in IT
that has nothing to do with just and only just the _intention_ of the devleopers
to win at backgammon.
It has all the REST of the information that went into it, what EVER that was.
And the same is true for the timing and play of the game in USING that state space:
although the DB does NOT learn unless you set the code to do so:
THE GAME ITSELF DOES DURING THE COURSE OF PLAY ACROSS ONE GAME.
That is what the definition of :"winning" the game IS (and more).
So given this above, and the assumption you are right. Care to tell us which random number generator(s) were used in the Training of Gnubg (I do know which one was used in the original 0.0 and 0.1 neural net trainers – the source Is available) . But here is food
I already denoted in the original post (and potential thew apostate and opposite in that post
as well as this one)\
that I might not be able to infer that, based on the rules of back gammon
or even with the knowledge of the formulas involved.
Albeit: I would bet you dollars to donuts
if i had the stake to make: I don't
that you would LOSE if I sat down and wrote a small analysis routine
to pluck the characteristics out
based on the list of PRNGs and the raining effects seen in the DB state space.
In which case, you might be able to say that "I" _could_ predict it.
I just od not take maximal advantage of all that extra knowledge
that you yourself in your "cube play weaknesses" and other argument sabove make
as not being "cheating". Which is not why I do not do it.
And if I _did_ do so, would it be a fair statistical test?
You know htat it owuld not.
Only across a selection of such PRNGs would it give a valid correlative result.
And more to the point:
It has no bearing on my poinyt. Only in your counterargument, doubly inverted again, here.
What differenc ewould it make about _me_: whne it was _GNUBG_ and the average __Player_
that we were discussing (contrary to counter examples you counter raised to counter points I was not making).
for thought. If I create a new Random number generator (lets say both a new real world RNG and a PRNG) and add support into GnuBG and don’t change the neural net
A new real wordl TRNG better than random.org
Whoo!. You must be a real expert to outdo the accepted norm.
Have you ever wondered why they do not log the sequences generated?
or the weights – can the bot possibly know patterns for random number generators it hasn’t seen. Well I say “No”. The information used during training did not include bias
And I say "Yes': based on the Ven diagram argument given above.
In fact, the real world subset called "accepted PRNG adequacy tests" is in fact subsumed in such scenarios, by the fact theat you only use accepted ones to train the game (again, outside of real world players of real world dice throws, which is in itself a PRNG).
or skew for random number generation (real world or PRNG) across all of space and time.
But it DID include bias or skew information ON THE AVERAGE about AVERGAE PRNG output.
And, whuile you can CLAIM there is no such "pattern", it is a far cry form mathematically DEMONSTRATING that here is no such pattern.
You owuld lose, because it is not so.
By the way, the fact I use the term "pattern" does NOT imply that a reducible denotatable characterisitc is present:
It does not have to be. )Cantor's diagonals argument precludes this being needed or even always possible).
All that it has to be is _convolutionally_ _convergent_ on the _goal_ of the _actual_ instance of the process in play ( and possibly weaker conditions than that:
such as players THINKING it will "win" and altering their behavior on THAT basis)
The actual real world is a bit vaster than that.
If someone believes there is a bias inside the neural net regarding randomness, then you should tell us which generators helped generate the bias because the simplest solution is to use a random number source that is different.
How does that remove a bias inherent in th act of using any PRNG?
Explain it to me please.random.org
ADMITS it is a PRNG (it remves sequences with "too much randomness":
It MAY do so, but you have not given a plausible reaosn to believe that
it WILL do so
in exactl and only
the cases where such a bias is PRESENT
those bieng the ones that my argumentatum is addressing.
and in NO way does that cover the gamut of how biases can crep in
much, many, much les ALL of the way nforeseen consequences can occur
that blow preconceptions
such as yours, _or_ mine
completely out of the water.
“Even random.org/com <http://random.org/com> notes that their random series are "improved" (made "more" effectively or "intuitively" random) by removing or reducing large series of improbable rolls (such as 0's and 1's in a row).”
Actually, you are wrong. It doesn’t arbitrarily throw away ones and zeros (it does but
You see? I note that specific statement, you deny it, and then you say it foesbut it doesn't
A confused mishmash here.
your view is too simplistic as stated here to give an accurate representation of what
I am not concerned with their _goals_
which i have not impugned, same arguments as with GNUBG developers
I _am_ concerned with unanticipated side effects (sigh: too complicated ot give the PRK here).
and yes, it DOES introduce a "bias" :: specifically admitted: no long runs of consecutive digits.
It _IS_ designed to do just that
to make the data more _effective_ towards _some_ types -f pursuits.
What are the extended effects on _other_ pursuits?
Such as a "fair:" backgammon game?
(Be sure to use my definitions when attacking here, not yours: improper logic when you do).
they really do). They do remove bits that are repeated but the algorithm does so for both 1’s and 0’s. It Is not biased in that regard. You can’t get less random data by applying a filter (like a noise filter,( to the existing data, as long as the filter applies the
Yes, actually you can. You are only considring a first moment PDF argument, not a Markhov chain state space, or any more convolved ysteretiv (memory/learning based) event sequence.
“SAME” logic to both 0 and 1 uniformly. On top of this something that also occurs is that data retrieved from the physical data source may not contain 100% entropy. So to begin with a lot of data retrieved from the real world is literally tossed away, and only the parts with the most entropy are used”.
Which in turn introduces another "bias"< intentionally set by the designers.
Relate the term :entropy' to "random". It is not an easy correlation, nor obvious.
And the relationship is not uniformly _random_ NOR evenly _entropic_.
As an example of this lets say I have a physical data source that is random. It just so happens that 3 bits of data are always 1. If I know the lower 5 bits of data contain the entropy I simply throw away the top 3 bits. Throwing away data that doesn’t add
The the data actually are NOT _uniformly- RANDOM. Think about it a minute.
And in the real world prior to human inventions and analog going digital:
where in Heaven, Earh, or Hell did you find _random_ data streams that had three bits alwas the same?
If that is not a clue to a hidden in the open source of NON randomness in THAT DATA STREAM, what will you accept as being so?
You are confused,in that you measure the randomnes of that data sream based only on yuor metrfication of it.
You have an argument of the form "I am that I am".
IF you definitions are CORRECTLY about randomness
THENA ND ONLY THEN your metrifications measuring it will be random.
And if your FRAMING for the data (three buits always the same)
is not random:
then it is highly probable that
your DATA has a non-random element
you jest ain't gotten to see yet.
anything to the randomness is valid. The guys at random.org over the years take the data from multiple sources and then XOR the data together (I think they use sounds cards still)
In other cases random numbers are generated from a pile of 1’s and 0’s (from physical sources) , but the actual ones and zeroes of the raw data are not random at all. This is the case if you have a fixed sample rate and the source of the randomness is in the time between successive transitions from 0 to 1, not the actual transitions themselves. So basically the randomness is temporal in nature. You effectively go down the data stream and look how long an event of 0’s and 1’s last and the time they appear is used to decide . This throws away a huge amount of data since the transitions are far less frequent than the total amount of data.
hich inherently palaces a functiohnal bias on the "trueness" of the TRNG you are looking at.
But you will not see that, I suspect.
Nor the potential for cross over causal linkages across disparate manifold state spaces.
Throwing away data uniformally with no bias doesn’t mean the data is less random, it
Nor more random. Nor the same. It is not rlevant to the question.
just simply means you are looking for the bits with entropy and the bits that aren’t noise.
“It is really difficult to predict what is actually in it, and whether it can "beat" the :"cheat" that the GNUBG database gives the engine.”
“Its difficult to predict whats actually in it”. So I will ask, did you do a statistical study
Sure is. State spaces code a lot of information in indirect fashion.
Try ot say "1+1=2" without 2,00 years of human though in it.
across a huge sample of matches to prove that there is an inherent difference between the results of matches done by random number generators known to Gnubg and a source of randomness that is unknown. Remember, you said ““The interesting thing to note is that GNUBG also cheats.”. That’s a definitive statement. You didn’t say ““The interesting thing to note is that GNUBG MIGHT cheats.””. If the bot has knowledge of
I sure did.
And I sued the word '"cheat"' ijnquotes to denote the exact sense of it.
Which you tried to define out and preclude.
how a player played (hypothetically) and used it to gain an advantage – that is not cheating. The human player can do the same thing when playing the bot – take
Sorry, those are YOUR dfinitions, not mine.
I reiterate: if knowledge outside of the rules of backgammon is involved:
then you have :"cheating' if this is not infact a part of the RULES of the game as stated.
As I have defined it.
And for a player who does not "get' tactics and strategy yet:
it is a form of "cheating" (using inacessible knowledge ongoing).
advantage of the bots weakness and exploit them. Same goes for 2 human players. If one player has no knowledge of his opp and has no idea about any potential “tells”, but the opponent has knowledge of the opponent and his tells – that doesn’t make the more knowledgeable player a cheater – it makes him more informed. If one player knows how to compute an accurate pipcount in his head and the opp can’t, that doesn’t make the player who can do it a cheater. If a player knows about match equity tables and the MWC at a particular score, and the opp has no idea what a MET is... It doesn’t make the player who knows – a cheater.
If GNUBG did not get it strictly from the rules of backgamon:
yep, it actually is "cheating:" as I have defined it.
And to give a more complete set of definitions in full form would take more space than you wqould bother to read or than I have time for.
Look up Cantor's Diagonalization argument (false)
Look up Law of the Excluded Middle (Theorem, actually, see "Laws of Form' George Spencer Brown) or the owrks of the Polish School of Logic, specifically Stansilaw Lesniewski.
And hten look up Godel's Theorem (correct, but badly misinterpreted)
((look up context free versus context sensitive languages)
And then see if you can folow a amthematical argument involving
a completely stated
PRK (primitive recursive kernel).
I am not saying you will concur with my conclusions: I am stating you may need some more tools to see what I am saying clearly.
IF a human is told a data source (Lets say a given PRNG) is random, and lets just say hypothetically Gnubg knew that wasn’t the case implied via the inputs, and the weight files – could that be cheating? Yes, one could say it is, the bot has been given a piece of information that needs to be equally shared by both players to be fair (What the bot
The bot has, in my argument, operationaly inferred a play based on the information present in the eigenstate of that.
NOT what you just said.
And that is "cheating" to the human player.
and human do with it is up to them). But the thing here is, no where in your post have you said how the neural net got its random data (which algorithm(s), and whether the match being played was using the same one. And if you feel that Gnubg has an unfair
Yes, actually I did. The human condition, over an dover.
As present in the eigenstate of the program code.
as present int he eigenstate of the exact order of presentation of each of the games played.
Ah! The DB will NOT be _exactly_ the same for different _orders_ of the games being played.
Just largely _similar_ in the effective _results)_ of the weighting file (which wil differ somewhat).
That is a state change. And implies information present and absent in each one from the other.
edge, then Gnubg developers have given you 2 methods - “External Dice” and “Manual Dice”.
Which in turn have their own eigenstate space, which is )Venn diagram argument above) convolved the same way again.
It sort of interesting. Lets say the NN was done with manually rolled dice (Hypothetically) and precision dice weren’t used and thrre was a known bias toward 5. Lets say Gnugb learned from all of that data, and starte dto play Backgammon, but its evaluation of a position and its value is based partially on those 5’s showing up more often (and the 2 showing up less often). Gnubg would learn to play BG in that environment but it could work against the bot if someone replaced the bad dice used during training with good precision dice used during matches played against the bot later on. Its possible that the implicit learned bias would work against the bot when different dice were used.
And in fact DID do that. Until earlier researchers learned how to get NNP networks to get that type of bias OUT.
Mainly by removing that particular moment (statistical mass) from their data.
Which Markhov states have NOT been removed that are stil there unseen?
I have seen nothing provided that proves Gnubg cheats. There is no evidence or data
Only what I showed: that it "cheats". Which you deliberately edited out. Or ignored.
Or _redefined_ (this is the typical _cheat" from apredisposed and _baised"
Of course, the value of GNUBG goes down a bit if
the actual effectivenes it has
is NOT based on the rules of BACKgammon
a brute force ruthless paradigm of a goal seking engine "win each time".
Because there are already a lot more effetcive engines doing that around.
supplied (If you have some I’d be happy to look at it) that there is any knowledge by the bot. You state ““It is really difficult to predict what is actually in it, “. So you must have done the difficult task to prove that patterns in the randomness do make a difference,
Nope. But you already knew that, in that I -asserted_ my stance.
Which you proceeded to ignore and use emotionalisms
cloaked as logical argument.
I wonder if this will posted ... I wil not repeat this type of exposition.
ther is no point to it.
When yuo catcall and call the catcall mthematics or science or reasoning.
If not posted I will yahoo group it instead.
and I wish to see that data. Until that’s done I don’t buy your first statement “The interesting thing to note is that GNUBG also cheats.”. And if you want to believe the bot cheats because it has an inherent accumulated knowledge of patterns of randomness in how it evaluates the equity in a position then I say use a different generator! The Code is open source, we’d be glad to see you add a new generator to the mix that you know wasn’t used during training.
Already addressed all those false reasonings above.
My opinion Is that you have given a flimsy argument to say the bot cheats because of some knowledge about PRNG’s that it learned indirectly during its training with no evidence to support the position. I am going to say that knowledge is power. GnuBG
Assertions a=on your part.
I note that NO One has actually pulled up and done a true random unfirom analysis
in exactitude, precisely, mathematically
of what optimal play WOULD be
based on unfirom random dice
and ONLY the rules of backgammon
the specific subcases
I _partially_ delineated above.
Until that is actually done:
your are being specious at obversive at best.
Look,man, if a HUMAN player can learn a "tell"
who are you to say the prgoram can NOT do so without proving it?
And your shift of my _assertion_ (at the beginning of your argument
to saying I must have _"proved" later on
was as adriotly maladroit an argument ad hominem as I have seen.
But way too typical.
knowledge doesn’t make it a cheater. It has a match equity table built in. That doesn’t make it a cheater, it makes it more informed. The bot can evaluation 100s of thousands of possibilities very quickly, humans can’t. Does that make the bot a cheater? No. It makes it knowledgeable. I am beginning to think (And I suggested this with Murat) that I’d be curious to see how one is defining “cheater”.
I already did, Michael"" which is why you adroitly avoided using _my_ definition
in your responses, and proceeded to skip back and forth through out mixmastering
form start to finish.
And why you had to tr y ot imply I did NOT do so at the end of your "logical" ''mathematical'
"fair" 'exposition' here.
And: it doe snot sund throughout like you were really curious about what I aas
saying in the first place, so:
sounds like you were being "truthful" again at thend.