Acquire program AI
Steve Okonski
implementation of Acquire. Both of us are moderately good players and
will often beat the computer. However, the program is challenging, and
sometimes we get blown out. We were wondering about what this says
about 1) the game itself, and 2) the computer implementation.
By examining the source code, I can see that in many places the AI makes
decisions on only the simplest of factors. Why is it that such simple
algorithms can produce challenging play, and sometimes crushing
victories? Several possibilities came to mind:
1) Acquire is highly dependent on luck.
2) There are not many decisions to make in the game.
3) The decisions have simple solutions.
4) Conversely, the solutions are not simple; instead, the AI coding is
very elegant.
5) Other possibilities?
We kinda agreed on reason #1, but it was getting too late for us to
discuss this in detail. Comments welcome.
David desJardins
> Several possibilities came to mind:
> 1) Acquire is highly dependent on luck.
> 2) There are not many decisions to make in the game.
> 3) The decisions have simple solutions.
> 4) Conversely, the solutions are not simple; instead, the AI coding is
> very elegant.
> 5) Other possibilities?
Another possibility (category 5) is that there's still a lot of room for
you to play better. More than one person has told me that after playing
a whole lot of Acquire against the computer, they got significantly
better and could win much more consistently. You don't mention any
improvement in your skill over time, which makes me think that perhaps
you are underestimating how much you can still learn.
The more players there are, the harder it is to win consistently. Not
just because your prior chance of winning is 1/6 instead of 1/4, but
because with six players instead of four, you only get to play 1/6 of
the tiles instead of 1/4 of the tiles, which substantially reduces your
control over how the game develops. So a reasonable challenge is to try
to win consistently against a smaller number of players, and then to
increase the number of opponents over time.
I don't know if your computer program makes it easy to examine the
holdings of the computer players. Many people who seem to consider
themselves good Acquire players think that playing with hidden holdings
makes the game significantly different or more interesting. This is
more or less conclusive evidence that they aren't very good. Competent
Acquire play requires always knowing what everyone has, either by
looking or by remembering. If you don't remember all of the stock
holdings of the computer players (and if the program doesn't let you
check them, or you don't use that option) then that's definitely a
significant area for improvement.
David desJardins
Vampire~D
Luck is a significant part of the game.
>2) There are not many decisions to make in the game.
Not when compared to a "complex" game. I consider Acquire to be relatively
low on the complexity scale. That isnt to say it cant be fun or
challenging.
>3) The decisions have simple solutions.
This is definately true.
>4) Conversely, the solutions are not simple; instead, the AI coding is
>very elegant.
I cant comment not having seen the code.
>5) Other possibilities?
The other possibility is that you are not very skilled at the game mechanics
yet. The fact that you have such a severe swing in outcomes, is a testament
to one or both of the following: The game is highly lick based, or Your
play is less than consistent and therefore in need of further development.
Computer AI is a difficult thing. You can usually find a hole in the code
somewhere that will allow you to defeat a computer player regularly. The
computer given the same exact set of givens at different times will always
chopose the same thing based on its coding. You as a human player do not
have to. The computer version of 1830, I consider to be very good! Mind
you the CPU players seem to play as a team for the most part, but
neertheless, I find the game to be challenging time and time again. This is
a somewhat mroe complex game. Severely more complex than Acquire. The
simplest of decisions can have a drastically different outcome based on all
the other players moves.
Also, if you want to see the luck factor reduced or at least get some ideas
on it, see the thread abouot Double Acquire.
Mike
Bagheera
David desJardins <da...@desjardins.org> wrote:
> Another possibility (category 5) is that there's still a lot of room
for
> you to play better. More than one person has told me that after
playing
> a whole lot of Acquire against the computer, they got significantly
> better and could win much more consistently. You don't mention any
> improvement in your skill over time, which makes me think that perhaps
> you are underestimating how much you can still learn.
Assuming you were a much better player, what sort of thought process
do you go through when playing?
This could help generate a much better AI for future versions.
BTW, I beat George's computer 90% of the time against 5 players.
--
Bagherra <jae...@frenzy.com>
http://www.frenzy.com/~jaebear
"What use is it to have a leader who walks on water
if you don't follow in their footsteps?"
Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't.
David desJardins
> Assuming you were a much better player, what sort of thought process
> do you go through when playing? This could help generate a much
> better AI for future versions.
I don't think I would assume that I am a much better player than someone
else. (Except perhaps if that other player doesn't count all the shares
sold, and keep at least a rough count of other players' available cash.)
I also don't think that studying the thought processes of human players
is all that likely to impove the play of computer opponents. Studying
Kasparov's thought processes is almost irrelevant to producing a strong
chess program.
But here are some examples of things that I think about:
1. Planning ahead several turns, to anticipate which players are
likely to buy shares in what chains, and how many. Especially when the
shares are likely to become sold out soon, and the timing of who gets
the last shares is critical.
2. When considering a merger, figure out what every other player will
do in the event of a merger. If all chains are on the board, anticipate
who is likely to get to found a new chain, and how that will affect the
dynamics. (Sometimes it can be to your advantage for another player to
found a new chain, as that gives that player a place to invest instead
of competing with you.) Anticipate whether the opportunity for the same
merger will still exist in a turn, and whether it will be more or less
favorable. If a merger that one controls is likely to be unfavorable
for some time to come, how can one play elsewhere in the affected chains
in order to change the dynamics of that merger so that it will
eventually be a positive opportunity?
3. Postulating that certain players hold certain key tiles (generally
tiles that are or could become merger tiles), anticipating whether they
will play them if they have a chance to do so, and planning ahead for
that event.
4. Trying to deduce which key tiles players have from their other
plays and purchases. For example, if a player suddenly starts buying
stock in a chain that player previously wasn't interested in, perhaps
the player just drew a key tile relating to that chain. Or if a player
expands a chain in a way that doesn't clearly benefit that player right
away, perhaps the player has a tile which becomes a merger tile as a
result.
5. Judging the "pace" of the game: how long is it likely to be before
more cash is generated by mergers? This is crucial to the decision of
whether to buy expensive shares, cheap shares, or perhaps even no
shares.
6. Managing the flow of tiles through my hand. When I adopt a
strategy of handicapping a chain by not playing certain tiles that would
help that chain, how much is my choice of plays restricted? Is it
likely to become more restricted in the future? Is the advantage of
holding back other players by retaining those tiles more or less than
the advantage of cycling them through my hand? (Often, when one decides
not to play certain tiles early, one ends up in a position where one has
to sit on them for almost the whole game, or at least a large fraction
of it.)
I don't really see how to translate any of these general concepts into a
computer algorithm. Indeed, if I could figure out how to translate my
general knowledge about Acquire strategies into an algorithm for playing
the game, then the game itself probably wouldn't be very interesting to
me any more.
David desJardins
Bagheera
David desJardins <da...@desjardins.org> wrote:
> Bagheera, the jungle scout <bagh...@my-deja.com> writes:
> > Assuming you were a much better player, what sort of thought process
> > do you go through when playing? This could help generate a much
> > better AI for future versions.
>
players
> is all that likely to impove the play of computer opponents. Studying
> Kasparov's thought processes is almost irrelevant to producing a
strong
> chess program.
Actually, Kasparov plays similar to most grand master players.
They memorize thousands of patterns and just try to set the board
up into those patterns in order to consolidate a win. This is
trivial to simulate in a computer program...well up to the bit
where you have to figure out how to store several million possible
patterns. That's basically how Deep Blue works, it does regressive
pattern matching on an insane number of board positions until it finds
the most beneficial pattern. Many chess programs start out by moving
"scripted" moves. This is a random set of moves pre-programmed in order
to determine what type of pattern the opponent is moving towards.
Many patterns are named (like Blitzkrieg, and several openers named
after famous players), which lends towards easy identification of
what the pattern is, and types of game play that is most beneficial
towards that type of game play.
> But here are some examples of things that I think about:
<snip strategies>
> I don't really see how to translate any of these general concepts into
a
> computer algorithm. Indeed, if I could figure out how to translate my
> general knowledge about Acquire strategies into an algorithm for
playing
> the game, then the game itself probably wouldn't be very interesting
to
> me any more.
>
> David desJardins
That is why we are the programmers and you are the game players.
A good programmer could easily take what you wrote and turn it into
an AI. I think I might snip the speculative strategies you use as
they are "fuzzy" in nature and won't really lead to any real intelligent
judgement on best decisions. However, one in particular, noting how
other players buy on certain chains, especially if they were
unattractive before, could lead to a good algorithm.
Another measure you could use is to see if a player is hesitant to play
certain tiles. These tiles could be integral merger links that would
hurt that player OR tiles that add to chains that player doesn't have
majority stock in.
AI is not an easy business, but "logical" thinking is very easy to
pattern on a computer.
As for Crenshaw's version of Acquire, after reading through the
code, it seems as though the computer always trys to play it's
rightmost tile. I might be reading this wrong, but that's what
it seems like. I haven't programmed in BASIC in over 10 years...
so, I reserve the right to be wrong.
--
Bagherra <jaebear @ frenzy.com>
http://www.frenzy.com/~jaebear
"There's a snake in my boot!"
Bruno Wolff III
>
> That is why we are the programmers and you are the game players.
Generally to get good AI you need good programming and good game playing
together.
> A good programmer could easily take what you wrote and turn it into
> an AI. I think I might snip the speculative strategies you use as
Not quite. David's strategies didn't specify values, just general guidelines.
You might be able to parameterize it and use trial and error to get some
decent paramter values. However this doesn't work in general.
> they are "fuzzy" in nature and won't really lead to any real intelligent
> judgement on best decisions. However, one in particular, noting how
> other players buy on certain chains, especially if they were
> unattractive before, could lead to a good algorithm.
Unfortunately noting things doesn't help in itself. After you note this
you need to use it when making decisions. And that part is the hard part
(at least if you want good decisions).
> Another measure you could use is to see if a player is hesitant to play
> certain tiles. These tiles could be integral merger links that would
> hurt that player OR tiles that add to chains that player doesn't have
> majority stock in.
This one will probably be hard for an AI to use. Since usually all they
get to know is what tile was played and how long it took a player to
move. The AI isn't going to know what exactly took a player so long
to make a particular move. Perhaps the player is just trying to make it
look like they have a tile that is nearly as good as the one they ended
up playing.
> AI is not an easy business, but "logical" thinking is very easy to
> pattern on a computer.
That I agree with.
David desJardins
> That's basically how Deep Blue works, it does regressive pattern
> matching on an insane number of board positions until it finds the
> most beneficial pattern.
This is absolutely nothing like how Deep Blue works. Deep Blue uses a
modified alpha-beta search, like all strong chess programs, with various
enhancements such as move ordering. This is entirely different from how
Kasparov and any other human player plays.
> That is why we are the programmers and you are the game players.
> A good programmer could easily take what you wrote and turn it into
> an AI.
"We"? Whom do you include in that illustrious group? I don't know you,
but I'm certain I'm a much better programmer than you are. I'm also a
much better programmer than I am a game player. I also know that it's
vastly more difficult to write decent computer players for such games
than you imply. That's why there aren't any. If it were really so
easy, then how would you explain the fact that none exist?
David desJardins
Nick Wedd
>I also don't think that studying the thought processes of human players
>is all that likely to impove the play of computer opponents. Studying
>Kasparov's thought processes is almost irrelevant to producing a strong
>chess program.
The methods that have worked so well for chess don't work nearly as well
for Go. People are trying to improve their Go programs by studying the
thought processes of human experts - with little success.
Nick
--
Nick Wedd ni...@maproom.co.uk
Bagheera
David desJardins <da...@desjardins.org> wrote:
> Bagheera, the jungle scout <bagh...@my-deja.com> writes:
> > That's basically how Deep Blue works, it does regressive pattern
> > matching on an insane number of board positions until it finds the
> > most beneficial pattern.
>
> This is absolutely nothing like how Deep Blue works. Deep Blue uses a
> modified alpha-beta search, like all strong chess programs, with
various
> enhancements such as move ordering. This is entirely different from
how
> Kasparov and any other human player plays.
break down what a pruned alpha-beta search is...I think you'll find
that's exactly how many humans think...
> > That is why we are the programmers and you are the game players.
> > A good programmer could easily take what you wrote and turn it into
> > an AI.
>
> "We"? Whom do you include in that illustrious group? I don't know
you,
> but I'm certain I'm a much better programmer than you are. I'm also a
I'd be willing to make that wager.
Should it be an experience flaunt?
education penis contest?
award show and tell?
my compiler can whip your compiler?
While I don't doubt your programming skills, I think you would
be hard pressed to say if you are better than me, or vice versa.
I walked into a few programming competitions and Demo Coding parties
thinking I was the most fragrant terd in the room, IYKWIM. And in
some cases I was, and others I wasn't. *I* do happen to be pretty
good at writing AI algorithms from just explanations of how things
should be, and I feel comfortable doing so in short periods of time.
I will be the first to say that I am not a programming GURU,
I just feel like one.
> much better programmer than I am a game player. I also know that it's
> vastly more difficult to write decent computer players for such games
> than you imply. That's why there aren't any. If it were really so
> easy, then how would you explain the fact that none exist?
it really isn't that hard. You just have to have intimate knowledge
on how to play, and more importantly, win the game.
How do I explain the lack of good AI?
It isn't profitable.
Kevin J. Maroney
>I'd be willing to make that wager.
>Should it be an experience flaunt?
>education penis contest?
>award show and tell?
>my compiler can whip your compiler?
I think that you started this pissing match and it is incumbent upon
you to acknowledge that you were both ignorant and arrogant, but I
suspect that will be far from coming.
--
Kevin J. Maroney | Crossover Technologies | kmar...@crossover.com
Games are my entire waking life.
David desJardins
> break down what a pruned alpha-beta search is...I think you'll find
> that's exactly how many humans think...
You are mistaken. I know what an alpha-beta search is, and I know a
fair amount of what is known about how humans play games, and they
aren't at all similar. I don't think you can find a single reference to
any reputable source that suggests that they are.
> While I don't doubt your programming skills
Then why did you say that you are a programmer and I'm not?
> I walked into a few programming competitions and Demo Coding parties
> thinking I was the most fragrant terd in the room, IYKWIM.
No, I have no idea what you mean, except you seem to be a denizen of
some bizarre subculture, and perhaps that subculture gave you some
really strange ideas about computer algorithms. I think reading some
standard texts on algorithms and AI instead would benefit you more.
> it really isn't that hard. You just have to have intimate knowledge
> on how to play, and more importantly, win the game.
So it's not just me that is outside the class of people who can program?
It's all the AI researchers in the world, who have tried and failed to
produce such methods for any games except that small class that can be
solved by brute force? You seem to think very highly of yourself, for
someone who isn't willing to identify yourself. At the very least, if
you could do a fraction of what you claim you could have the ACM Turing
prize.
David desJardins
Bagheera
> So it's not just me that is outside the class of people who can
program?
self conceit, how nice.
You implied you didn't know how to write an ai to emulate
assumed human thinking, therefore, I gathered that this
was a field you were inept at. If I am mistaken, I apologize,
I was just running off your lead.
> It's all the AI researchers in the world, who have tried and failed to
> produce such methods for any games except that small class that can be
> solved by brute force?
there are many algorithms out there which are very well done,
(chinook, the c-4 solution, and others) that are not brute
force solutions. While I admit, that these solutions are far
above my abilities, general AI programming in itself is relatively
simple in comparison.
Take non-solved connect four for example.
If you told me to write an agent that was capable of stale-mating,
and for my search algorithm, I would determine positions which
are either two in a row to stop a three in a row, or a three in a
row cornered for a block, and to search out positions which are dually
beneficial by a) blocking a tri-line as well as b) building on an
existing line the agent was working on...
Well, while it sounds like a mouthful, that is pretty trivial to code.
It's a simple matter of looking for positions where the opponent can
gain a line, and block that path. If there is no danger, build on your
own lines, or deter future paths. This is something a "good" player
might do while playing connect four. What makes this hard to program?
> You seem to think very highly of yourself, for
> someone who isn't willing to identify yourself. At the very least, if
> you could do a fraction of what you claim you could have the ACM
Turing
> prize.
My identity is freely available.
I even have a nice photo with my Deja-user page.
Shoot, you could even go to my website which is
at the URL included with every single post I have
ever made.
And for the record, I've sent two entries to the Turing contest,
and though I haven't won yet, I plan on being in the group that wins
it :)
I must admit, I am not doing it alone, though. I'm not very good
at Natural Language Processing, but my partners are. I focus more
on Expert Systems, Fuzzy Logic, and Machine Learning. They use me
for my experience in Fuzzy Logic.
On a similar topic, I have won in 3 other ACM programming contests,
and gotten an honorable mention at the ACM world finals programming
competition. Again, another goal of mine is to win that competition.
Unfortunately, I am only going to get one more try :(
David desJardins
> human thinking, therefore, I gathered that this was a field you were
> inept at.
No one knows how to write computer AIs that emulate any aspect of human
thought, much less human game playing. So do you gather that all AI
researchers are "inept"? Just maybe, the problems are harder than you
suggest.
> there are many algorithms out there which are very well done,
> (chinook, the c-4 solution, and others) that are not brute force
> solutions. While I admit, that these solutions are far above my
> abilities
In fact, these are perfect examples of brute force solutions, often
cited as such. And this is just the sort of project that my expertise
is in. I can do that very well---but you can't use those techniques to
play Acquire well. At least, no one has ever demonstrated any such
capability, and doing so would be a major advance in computer science.
> My identity is freely available. Shoot, you could even go to my
> website which is at the URL included with every single post I have
> ever made. ( http://www.frenzy.com/~jaebear )
You seem to go to great lengths to keep it out of your postings. I went
to the URL and it isn't there either, or else it's hidden on some
sub-sub-page I didn't have the time to search for. Clearly you don't
want to stand behind your postings, or you wouldn't post anonymously
with a pseudonym.
> And for the record, I've sent two entries to the Turing contest,
The A. M. Turing Award is awarded by the ACM each year for the best work
in computer science ("lasting and major contributions to the computer
field"). Some well known winners include Marvin Minsky, John McCarthy,
Donald Knuth, Ken Thompson, Dennis Ritchie, Richard Karp, John Hopcroft,
Robert Tarjan, and many others. If you could do what you claim, you
would certainly receive it. Unfortunately, there's no Nobel Prize in
computer science, or you would get that too.
The good news is that you have to do much less than you say you can do
in order to be recognized as a major figure in computer science. I, for
one, will be glad to read your research papers and study your work when
you succeed.
David desJardins
Bagheera
David desJardins <da...@desjardins.org> wrote:
> "Bagheera, the jungle scout" <bagh...@my-deja.com> writes:
> > You implied you didn't know how to write an ai to emulate assumed
> > human thinking, therefore, I gathered that this was a field you were
> > inept at.
>
> No one knows how to write computer AIs that emulate any aspect of
human
> thought, much less human game playing. So do you gather that all AI
> researchers are "inept"? Just maybe, the problems are harder than you
> suggest.
I'll let the people on comp.games.ai argue that one.
And I don't gather any AI researchers as inept...just you.
> > there are many algorithms out there which are very well done,
> > (chinook, the c-4 solution, and others) that are not brute force
> > solutions. While I admit, that these solutions are far above my
> > abilities
>
> In fact, these are perfect examples of brute force solutions, often
funny, I don't find the c-4 solution cited except in one document
in a game ai competition research paper. And then, it is not a "brute
force solution" in that it does not search EVERY path, but the few paths
that follow the algorithm of "stalemate, win, initial move"
> cited as such. And this is just the sort of project that my expertise
> is in. I can do that very well---but you can't use those techniques
to
> play Acquire well. At least, no one has ever demonstrated any such
> capability, and doing so would be a major advance in computer science.
well, no, you cannot use the algorithm that plays backgammon, checkers
or c4 to play acquire, but that is because acquire is a different game.
The thing that makes acquire truly unique in this situation is that
almost every factor in the game is random. It's a wonder the game even
gets any noteriety because of this fact.
For a "well designed" agent for Acquire, you would have to write
a "fuzzy logic" AI. There is no brute force solution, and indeed,
unlike C-4 there is no solution at all. You can't solve chaotic
games. It is incomprehenisbly infeasible.
To get back to the original issue at hand, I was asking for people
to give suggestions for tactics on playing Acquire so that *I* or
one of my contemporaries could generate a "fuzzy logic" powered
agent that could play an "intelligent" game of Acquire that could
challenge an average or better player. I feel fully capable of
adapting REAL strategies into code for an agent. While I may not
be able to enter "zen" aspects such as hunches, by using fuzzy logic
I could simulate that aspect of it. The point is, there are people
out there who play Acquire masterfully, despite the fact that the game
is almost purely random. They use techniques that are solid, and
predictable. It is this predictability that you encode into the game.
Yes, it is true, there will probably not be anything like a real AI
for games in the near future, but there are agents for games like
CivNet and Quake that are virtually unbeatable. I postulate that
given a set of constraints, a similar agent could be generated for
any chaotic board game.
> > My identity is freely available. Shoot, you could even go to my
> > website which is at the URL included with every single post I have
> > ever made. ( http://www.frenzy.com/~jaebear )
>
> You seem to go to great lengths to keep it out of your postings. I
went
> to the URL and it isn't there either, or else it's hidden on some
> sub-sub-page I didn't have the time to search for. Clearly you don't
> want to stand behind your postings, or you wouldn't post anonymously
> with a pseudonym.
ahh, but it is there...you just didn't care to find it.
It's in everything I did there. And I post anonymously
to avoid the deluge of Spam that usually ensues. Even though
it really isn't anonymous, because if you took the time to look
up my user profile on deja.com, you'd find that I filled out 95%
of the information they requested on the profile page.
> > And for the record, I've sent two entries to the Turing contest,
<snip ACM award stuff>
we are obviously talking about two different Turing contests here.
The one I'm talking about is held annually and is designed such that
entrants create AI agents that are tested against turings test for
perfect AI. (indistinguishable from a human under normal interaction).
The contest, of course, is imperfect in that it uses an IRC like
platform to do the test and therefore only focuses on the cognitive
aspect of AI, and not the pattern recognition or speech recognition
portions. (it is also hard to test emotion over IRC).
Geenius at Wrok
> "Bagheera, the jungle scout" <bagh...@my-deja.com> writes:
>
> > You implied you didn't know how to write an ai to emulate assumed
> > human thinking, therefore, I gathered that this was a field you were
> > inept at.
>
> No one knows how to write computer AIs that emulate any aspect of human
> thought, much less human game playing.
That's not true. I've done it. And I'm not even a professional
programmer, just a hobbyist. I've written two game AIs, one that played
Mille Bornes and one that plays Hot Death Uno (change of verb tense
because the Uno program is still floating around, whereas the Mille Bornes
one went the way of the Apple //e I wrote it on). The Mille Bornes
program used the same heuristic that I myself apply when I play. The Uno
program is a little more fudgy, to account for special card effects that
can't really be quantified, but it still plays fundamentally the same way
that I do, and I even programmed the computer players to hold grudges and
to do certain irrational things that the real people I've played with do,
such as getting into Reverse wars and "one good Draw 2 deserves another"
behavior.
Both programs were written in BASIC (Applesoft and Visual, respectively)
rather than in actual firing neurons, but from a logic standpoint, they
evaluate(d) play decisions the same way I do. And I think I'm fairly
qualified to speak on whether I'm human.
Actually, I think your premise -- the way you stated it -- is way off.
Programming a computer to emulate human game playing, at least for certain
games (very abstract games such as chess and go not included among them),
is much easier than programming it to emulate ordinary undirected human
thought.
--
"I wish EVERY day could be a shearing festival!" -- The 10 Commandments
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+
Keith Ammann is gee...@albany.net "I notice you have a cloud of doom.
Live with honor, endure with grace I must admit it makes you seem
www.albany.net/~geenius * Lun Yu 2:24 dangerous and sexy."
David desJardins
> I even programmed the computer players to hold grudges and to do
> certain irrational things that the real people I've played with do,
> such as getting into Reverse wars and "one good Draw 2 deserves
> another" behavior.
For once I think you have a good point. There are some aspects of human
behavior, especially the more "irrational" ones, which one can emulate
pretty easily. What I should have said is that the "rational"
approaches to problem solving that people use, especially the most
skilled players of various games, are far beyond what computers can do.
> Programming a computer to emulate human game playing, at least for certain
> games (very abstract games such as chess and go not included among them),
> is much easier than programming it to emulate ordinary undirected human
> thought.
There's little question that programming computers to play games is much
much easier than many other aspects of AI. When one thinks about the
carefully defined problem domain of something like Acquire, or Texas
Holdem, or Squad Leader, and the power of modern computers, it's
actually pretty astounding that computers can't do what humans can do
even in such a limited domain. But that's why the extremely slow
progress on even that relatively easy problem is so discouraging for the
general AI problem.
David desJardins
the Dave
Geenius at Wrok <gee...@albany.net> wrote:
>On 15 Jul 1999, David desJardins wrote:
>> "Bagheera, the jungle scout" <bagh...@my-deja.com> writes:
>> > You implied you didn't know how to write an ai to emulate assumed
>> > human thinking, therefore, I gathered that this was a field you were
>> > inept at.
>>
>> No one knows how to write computer AIs that emulate any aspect of human
>> thought, much less human game playing.
>
>That's not true. I've done it. And I'm not even a professional
>programmer, just a hobbyist. I've written two game AIs, one that played
>Mille Bornes and one that plays Hot Death Uno (change of verb tense
>because the Uno program is still floating around, whereas the Mille Bornes
>one went the way of the Apple //e I wrote it on).
Part of the reason this has become a Pissing Contest is because you're
all talking about 3 different things (I won't speculate on other
reasons....). DdJ is talking about how we humans actually think when
we play games (which, he is right, nobody really knows how to
program), Jae is talking about how to implement certain AI algorithms
(which is well-understood--after all they're algorithms), and Geenius
is talking about how to write a program that plays particular games at
least competently (it's known how to do this for some games, but not
others).
--
Bay Area Games Day VII is Sat. July 31 : http://www.io.com/~dk2/gamesday
David desJardins
I'm confident you aren't going to find a single other person who will
seriously argue that interviewing an expert human game player, and using
what the human tells you to write an algorithm that plays the game that
way at an expert or even competent level, is remotely within the state
of the art.
> funny, I don't find the c-4 solution cited except in one document
> in a game ai competition research paper.
Not only is Victor Allis's master's thesis on the subject widely
available, but his PhD thesis is available in book form (Searching for
Solutions in Games and Artificial Intelligence, ISBN 90-9007488-0). I
don't think you can find anything in his work that supports your ideas.
> And then, it is not a "brute force solution" in that it does not
> search EVERY path, but the few paths that follow the algorithm of
> "stalemate, win, initial move"
Proof number search, like alpha-beta search, is a perfect example of
(intelligent) brute force solution. No one but you limits the term
"brute force" to mean exhaustively listing every possible position in
the solution, instead of just the ones necessary. You can obviously
refine search techniques, but you still aren't playing the game anything
like how humans do. (Again, you are not going to find anyone to support
your notion that alpha-beta search is a close model for how humans
reason.)
> To get back to the original issue at hand, I was asking for people
> to give suggestions for tactics on playing Acquire so that *I* or
> one of my contemporaries could generate a "fuzzy logic" powered
> agent that could play an "intelligent" game of Acquire that could
> challenge an average or better player. I feel fully capable of
> adapting REAL strategies into code for an agent.
I gave you many such suggestions. So I look forward to your results.
I'll give very long odds that you won't remotely approach being able to
challenge a decent human player. I also think that with a smaller
amount of work one could do a much better job of developing a computer
algorithm using different techniques, such as neural nets---but this,
again, isn't going to play anything like how humans do.
> Yes, it is true, there will probably not be anything like a real AI
> for games in the near future, but there are agents for games like
> CivNet and Quake that are virtually unbeatable.
You can't write a player for CivNet that is even at the level of a
competent beginner, much less "virtually unbeatable".
David desJardins
David desJardins
>
Either you identify yourself, or you don't. Putting your name on your
postings wouldn't increase or decrease the amount of spam that you get.
It would just imply that you want to be taken seriously, and don't mind
standing behind what you write. I'm not surprised that you don't want
to do that.
David desJardins
The Maverick
>
> Clearly you don't
> want to stand behind your postings, or you wouldn't post anonymously
> with a pseudonym.
What is "anonymous" about posting as Bagheera the Jungle Scout? Check
your dictionary on that one! If he was posting under a name never seen
here or "unk...@nowhere.net" I could see your point... but the
implication that someone's ideas are better or more sincere because they
happen to be posting under the grand and mighty title of David
desJardins (an "anonymous" pseudonym for all we know or care) is just
plain dumb.
the Mav
--
Cliffhanger Serials, Boardgames, Videogames, and Red Baron I
http://www.volcano.net/~themaverick
The Classic Microgames Museum
http://www.angelfire.com/ca/themav/micind.html
The Macho Women with Guns Homage Page
http://www.geocities.com/TimesSquare/Lair/6747
David desJardins
> What is "anonymous" about posting as Bagheera the Jungle Scout? Check
> your dictionary on that one!
anonymous adj. 1. with a name that is not known or not made public.
It's pretty funny to see the anonymous "Maverick" defending the
anonymous "Bagheera"! Hey, you guys obviously have the right to post
without admitting who you are. But it takes a lot of gall to post under
a pseudonym, refuse to give your name, and then claim you aren't hiding
your identity!
> David desJardins (an "anonymous" pseudonym for all we know or care)
You're welcome not to care, but clearly plenty of people agree about the
dubious contributions of posters who hide their identities rather than
stand behind what they write.
If you were to care, it's easy to check that I am who I say I am.
David desJardins
Stephen Tavener
To answer a number of points, ...
In a lot of two player games, I think players do something similar to a
pruned alpha-beta search: "If I do this, then he'll play that, and then
I play..." - the "human" element is not the basic algorithm, but the way
players decide which move their opponent will make. This could be
decided by knowledge of your opponent's playing style, pattern
recognition, a library of openings, or even the compluters's "what's
their best move?"
Many multi-player games encourage players to choose their best move for
*this* turn, without looking ahead - otherwise the downtime between
turns tends to get long and painful. In this case, players don't
minimax as such, but an element of pattern recognition creeps in which
serves a similar purpose.
One of the best computer players for a 2-player game of pure information
(that I have encountered) is Doby III - a Trax playing game written by
Dr. Donald Bailey. On levels 1-7, the computer uses algorithms for
simplifying a board position plus a lookup library. Only on the highest
level - level 8 - is there any lookahead. It can beat most human
players consistently on level 7. (See http://www.traxgame.com/ for more
information on Trax and Doby 3).
TTFN,
Stephen
--
Stephen Tavener | There is no such thing
Games bought,sold,traded,played | as "just a cat"
http://www.scat.demon.co.uk/ | - Tanya Huff
Geenius at Wrok
> Either you identify yourself, or you don't. Putting your name on your
> postings wouldn't increase or decrease the amount of spam that you get.
> It would just imply that you want to be taken seriously, and don't mind
> standing behind what you write. I'm not surprised that you don't want
> to do that.
Even though I put my name at the bottom of my posts rather than at the
top, what more, exactly, does that tell you? To my knowledge, there are
two Keith Ammanns in this country, whereas there is only one Geenius at
Wrok. Unless you're planning to look me up in the phone book and make
harassing calls to my house or report me to local authorities every time I
get out of line, I'm no less accountable under a pseudonym than I am under
my legal name. You still know it's the same guy posting every time. The
only true anonymity -- in which you have no way of knowing whether the
poster is one who came before or not -- is to use an account with a
generic name that lots of other people use. Or a long number that people
can't keep track of easily, like anon.penet.fi used to do, so that the
identities all blur together.
Bagheera
David desJardins <da...@desjardins.org> wrote:
> "Bagheera, the jungle scout" <bagh...@my-deja.com> writes:
>
> I'm confident you aren't going to find a single other person who will
> seriously argue that interviewing an expert human game player, and
using
> what the human tells you to write an algorithm that plays the game
that
> way at an expert or even competent level, is remotely within the state
> of the art.
it may not be "state of the art" but it works.
> Proof number search, like alpha-beta search, is a perfect example of
> (intelligent) brute force solution. No one but you limits the term
> "brute force" to mean exhaustively listing every possible position in
> the solution, instead of just the ones necessary. You can obviously
> refine search techniques, but you still aren't playing the game
anything
> like how humans do. (Again, you are not going to find anyone to
support
> your notion that alpha-beta search is a close model for how humans
> reason.)
one already posted that.
I'm sure others would agree.
I think you need to start thinking "outside of the box".
Take for example a game of chess. You could implement the
game play as an alpha-beta search. And quite surely, some
intermediate level players play this way.
The player would make a set of decisions in his head about what
would happen if he moved this piece here or that piece there, and
he would prune the "tree" in his head based on moves that gave
his opponent a stronger position. From this pruned tree, he would
make a decision on the remaining moves open to him.
I don't know about you, but:
a) this is the way a lot of average players play
b) this is TRIVIAL to implement in programming
c) this is a mapping of human decision process to computer logic
While it may not be the way a human "thinks", it is a fair simulation
of the process involved. In fact, it is possible that by reducing the
algorithm to this simple decision process you are actually creating
something superior to human thought. But that is speculative.
> > To get back to the original issue at hand, I was asking for people
> > to give suggestions for tactics on playing Acquire so that *I* or
> > one of my contemporaries could generate a "fuzzy logic" powered
> > agent that could play an "intelligent" game of Acquire that could
> > challenge an average or better player. I feel fully capable of
> > adapting REAL strategies into code for an agent.
>
> I gave you many such suggestions. So I look forward to your results.
what you gave me were a serious of "hunches". You didn't actually
state a hard-fast strategy. Some points could be boiled out of your
statement, but all in all, what you described was the "feeling" of
playing the game, not the overall strategy involved. For instance,
you never described how you selected the tile to play, nor did you
explain your basis for determining the worth of purchasing a particular
stock. What you gave would make a purely fuzzy agent, and it wouldn't
be worth playing against even a chaotic opponent.
> I'll give very long odds that you won't remotely approach being able
to
> challenge a decent human player. I also think that with a smaller
> amount of work one could do a much better job of developing a computer
> algorithm using different techniques, such as neural nets---but this,
> again, isn't going to play anything like how humans do.
again, this is because your description of a strategy was inferior.
A neural net could certainly do a wonderful job of playing the game,
but I don't personally know anyone who would be willing to invest the
time to write one for a non-commercial product.
> > Yes, it is true, there will probably not be anything like a real AI
> > for games in the near future, but there are agents for games like
> > CivNet and Quake that are virtually unbeatable.
>
> You can't write a player for CivNet that is even at the level of a
> competent beginner, much less "virtually unbeatable".
>
> David desJardins
Me? No, I can't. But, I have played against superior agents on CivNet.
I admit that there is not an unbeatable agent for CivNet to my
knowledge, but my statement also extended to Quake, where I know there
are "virtually" unbeatable agents (assassin, Knife, black wind, and
stilleto to name a few I know of) implemented and being used. Maybe
this is because you can write the agent script directly in the game
unlike CivNet where you have to write it offline and run it through
a script (at least that's how I have to do it, others may have found
a better way).
And as to the question of my anonymity...
To quote my good friend the Doctor, "My name is Smith. John Smith."
Like I said, you can look it up under deja-news, go to my website
(where it is on ALL new pages created. It's in the copyright
information on all copyrighted stuff, it's everywhere. Just because
you are too inept to dig for it doesn't mean it isn't there).
Just because you post under the name David desJardins
doesn't mean that is your legal name. For all we know, it is a
pseudonym and your real name is Theodore Kzinsky. How do we know?
If it makes your nuts so bunched up, just finger my email address.
You can do that you know. I don't have a plan up or anything, but
it is there. And it has my real, legal name on it. Then there is
the issue, do you want my REAL name, the name that people call me by,
the name that my family uses, the name I publish under, my pen name,
my stage name, my canvas name, my delta name, my coder name, my irc
handle, my ICQ handle, the name I give the IRS, the name on my paycheck,
the name on my diploma, or the name I get called by the general USENET
community?
Be careful what you ask for. Some people are more complex than you
wager for.
For a short reference:
I have been published in the Strategist three times I forget what name I
told them...email varies.
I have two publications from university (quoted as assitant, but I wrote
them both):
Genetic Algorithms and rule based genesis. Apr 97
Machine Learning using Genetic Algoritms and rule based genesis. Dec 98
I have 6 demos out on the web somewhere:
Bumblbee.exe WaterKAD.exe BedBugs.exe
SpcCmdr.exe RRBlues.exe MskSmpl.exe
These are all under the pseudonym Polar Bear Productions
I have 2 impulse trakker songs
TheMission.it BountyHunter.it
Credited to MadDog of PolarBearProductions
I have a book that is currently under production for print run under
my pen name, Will S. Grant.
I could go on and on, but I really don't think it's necessary.
If you really want to know more about me, please scour my website
thoroughly.
Bagheera
In article <7mnfpi$gsd$1...@nnrp1.deja.com>,
David Damerell
>"Bagheera, the jungle scout" <bagh...@my-deja.com> writes:
>>I'll let the people on comp.games.ai argue that one.
>I'm confident you aren't going to find a single other person who will
This is why our resident gasbag should be less prone to exaggerate...
>seriously argue that interviewing an expert human game player, and using
>what the human tells you to write an algorithm that plays the game that
>way at an expert or even competent level, is remotely within the state
>of the art.
As far as I know, there is quite a relationship between the quality of a
Go AI and the quality of the Go players involved in writing it; it'll
always be awful at Go, but how awful does depend on the human players
involved.
--
David/Kirsty Damerell. dame...@chiark.greenend.org.uk
CUWoCS President. http://www.chiark.greenend.org.uk/~damerell/ Hail Eris!
|___| You bought a mask: I put it on: you never thought to ask me if I wear
| | | it when you're gone. The Sisters of Mercy: When You Don't See Me.
Bruno Wolff III
>
> Yes, it is true, there will probably not be anything like a real AI
> for games in the near future, but there are agents for games like
People and bots aren't playing on a level field. In quake the clients are
expected to hide information from the players. Bots would have access
to more information then a typical player would. Also the speed at which
you can perform actions seems to be how fast you can tell the server what
you are doing. This puts humans at a serious disadvantage.
If you changed the system to a client server one where actions took minimum
times comparable to human reaction times and allowed humans and bots access
to the same information, the contest would be more interesting. If you
increased minimum action times to where people actually had time to think,
the strategic incompetence of bots would probably make them poor players.
The Maverick
>
> "The Maverick" <thema...@volcano.net> writes:
> > What is "anonymous" about posting as Bagheera the Jungle Scout? Check
> > your dictionary on that one!
>
> anonymous adj. 1. with a name that is not known or not made public.
>
> It's pretty funny to see the anonymous "Maverick" defending the
> anonymous "Bagheera"!
Hell, I see two names there... perhaps you should check your posts for
accuracy after you send them to your typist! Or maybe your AI just
needs tweaking...
David desJardins
>> I'm confident you aren't going to find a single other person who will
>> seriously argue that interviewing an expert human game player, and
>> using what the human tells you to write an algorithm that plays the
>> game that way at an expert or even competent level, is remotely
>> within the state of the art.
>
> it may not be "state of the art" but it works.
As long as you have zero examples of this, and no one has ever
successfully done it, and your argument is simply that it's very easy to
do but that no one has ever bothered, I think you are going to have a
hard time convincing anyone to take that position seriously.
>>> To get back to the original issue at hand, I was asking for people
>>> to give suggestions for tactics on playing Acquire so that *I* or
>>> one of my contemporaries could generate a "fuzzy logic" powered
>>> agent that could play an "intelligent" game of Acquire that could
>>> challenge an average or better player. I feel fully capable of
>>> adapting REAL strategies into code for an agent.
>
> what you gave me were a serious of "hunches". You didn't actually
> state a hard-fast strategy.
Exactly. Humans don't play games by applying a fixed algorithm which a
computer can then apply step by step. If, when you interviewed an
expert human player, and asked how the human decides what moves to make,
the human gave you a step-by-step deterministic algorithm for making the
decisions, then of course it would be easy to translate that algorithm
into computer code and make a computer play just as well as the expert.
Furthermore, it would also be easy for any reasonably intelligent human
to listen to the algorithm, apply the same algorithm, and play just as
well as the expert. That's precisely why that's never the case for real
games. Humans don't play games where their decision-making process
consists of just applying a small fixed set of rules, because first of
all playing like a (very slow) computer is boring, and secondly because
there's no skill in such games: anyone can play just as well as anyone
else. That's why there's no tournament blackjack, or any other game
where you can just write down the strategy. It's also precisely why
your whole plan of interviewing human experts, asking them how they
play, and then translating their answers into a computer algorithm,
doesn't produce satisfactory computer game players.
> Me? No, I can't. But, I have played against superior agents on CivNet.
No. You haven't. There aren't superior computer players for CivNet, or
even competent ones. If you think that there are, then your standards
for competence are ridiculously low.
(I think you are probably confused and talking about computer players
that in fact get a big advantage over the humans, and then gang up on
the human players to boot. And these are still easily beatable. The
only interesting question in playing against computer players is how
fast one can win and with how high a score.)
> my statement also extended to Quake, where I know there
> are "virtually" unbeatable agents (assassin, Knife, black wind, and
> stilleto to name a few I know of) implemented and being used.
Quake isn't a strategy game. In any game where reflex speed is the
important factor, of course computers can beat humans. I can design a
computerized boxer that can beat Mike Tyson, because when he hits it
he'll break his hand. But this doesn't have anything to do with AI.
> I could go on and on, but I really don't think it's necessary.
> If you really want to know more about me, please scour my website
> thoroughly.
No thanks. You can decide whether to put your name on your postings or
not. If you choose to post anonymously, I don't have any interest in
playing detective to undo that decision. It's just that your
contributions will receive the respect that that choice deserves.
David desJardins
David Damerell
>People and bots aren't playing on a level field. In quake the clients are
>expected to hide information from the players. Bots would have access
>to more information then a typical player would.
[And can react faster].
The roguelike computer game 'Angband' has an automatic player which does
not have access to more information than a human player does [1] - in fact,
although it is compiled up as part of the Angband executable, it is
designed to be potentially operable as an entirely separate process - it
attempts to parse the screen output rather than sucking the information
directly out of the mechanics of the game (it cheats on some things like
the inventory, but only for speed; it doesn't use this ability to gain any
extra information). In roguelikes, reaction times are not an issue; these
games wait as long as you please for you to take a move.
Nevertheless, the 'Angband borg' is a moderately competent player -
largely because Angband rewards slow cautious play over risk-taking, and
the borg is incredibly craven but can play literally [2] hundreds of times
faster than a human being.
[1] It does have a perfect knowledge of the properties of every monster,
item, artifact &c; but a human player _could_ acquire such knowledge.
[2] Yes, I know what 'literally' means.
David desJardins
> Nevertheless, the 'Angband borg' is a moderately competent player -
> largely because Angband rewards slow cautious play over risk-taking
I wouldn't call it competent. In any actual contest of skill (as
opposed to just patience) between a human and the AI, the human is far
superior. There's a long history of such programs, going back to
'rogomatic'. Rogomatic was a substantial accomplishment in my view: it
could actually play Rogue. But I wouldn't call it in any way comparable
to a human. I'm not sure what its chance of winning the game is, but I
think it's hundreds or thousands of times less than my own. Perhaps the
Angband robot has a higher ratio, but it's not a great accomplishment to
be able to win a game that a careful and patient human can win with
probability 0.999.
Rogue is a pretty good example of a game where, to some extent, one can
study what a competent human does, extract that information into
automated strategies, and then apply them to a computer algorithm, with
results which are far inferior to the human, but not embarrassingly
bad. But that's also why it's not played much any more, and certainly
not taken seriously as a contest or test of skill: the fact that the
techniques for playing the game can all be described in a methodical way
makes it not much of a contest. It's nothing like the board game
(Acquire) that this thread was about.
> [1] It does have a perfect knowledge of the properties of every monster,
> item, artifact &c; but a human player _could_ acquire such knowledge.
Yeah, in fact I do have such knowledge. It's all in the source.
David desJardins
David desJardins
The difference isn't some sort of legal notion of accountability. It's
that the people who choose to hide their identity have a reason for
doing so: they don't want to be associated with what they say. While
anyone can of course have the potential to make useful contributions,
the people who carefully keep their name from being associated with
their postings have a much worse track record. That's hardly surprising.
> The only true anonymity -- in which you have no way of knowing whether
> the poster is one who came before or not -- is to use an account with
> a generic name that lots of other people use.
In practice, people post under their own name or they don't. In theory,
some of those posting under real names could be constructing elaborate
fictional identities. In practice, that's not the case: the people who
think that identifying themselves with their postings is worthwhile are
the same ones who don't feel any reason to hide their identity.
David desJardins
David desJardins
> In a lot of two player games, I think players do something similar to a
> pruned alpha-beta search: "If I do this, then he'll play that, and then
> I play..."
I guess our difference is in our interpretation of "like". Any analysis
of a two-player game must necessarily involve some notion of "if I do
this then he'll do that and then I will do this". Just because
alpha-beta search involves such analysis, and humans also use such
analysis, doesn't make the two methods seem similar to me. Alpha-beta
search uses precise numerical cutoffs: if a value is a fraction higher
than another value, then a whole subtree will be cutoff, while if it's a
fraction lower, then a large subtree may be searched. It also uses a
fixed ordering: once a move is considered, then all analysis of that
move is completed before considering other moves; humans consider many
moves simultaneously, and go back and forth in their thinking between
different plans. Humans also consider similar positions very
differently from computers using alpha-beta search; computers can use
hash tables to detect transposition, but they can't efficiently discover
that certain lines of analysis apply to many very similar but not
identical positions, while this is crucial to almost all human analysis.
And, of course, humans only consider a very tiny fraction of the
positions that any alpha-beta search would consider: this pruning (which
isn't done using any sort of alpha-beta technique) is essential to their
succcess.
> the "human" element is not the basic algorithm, but the way
> players decide which move their opponent will make.
See above for a strong disagreement.
> One of the best computer players for a 2-player game of pure information
> (that I have encountered) is Doby III - a Trax playing game written by
> Dr. Donald Bailey. On levels 1-7, the computer uses algorithms for
> simplifying a board position plus a lookup library.
Computer algorithms can definitely play some games very strongly in this
manner. Backgammon programs using neural nets for training are also
very strong, even without any lookahead (but still much stronger with
lookahead, of course). The fact that these algorithms don't "think" the
same way that humans do doesn't mean they can't play the games well.
David desJardins
David desJardins
> Many multi-player games encourage players to choose their best move for
> *this* turn, without looking ahead - otherwise the downtime between
> turns tends to get long and painful.
Actually, there's another reason that players don't do this, which von
Neumann discovered a long time ago. You can construct game trees for
multiplayer games using lookahead analysis: find the end of the game,
then just before the end of the game find the best move for the player
who is about to end the game, then assuming that that player will make
the best move, find the best move for the player before that, and so on.
Von Neumann discovered two things. First of all, humans don't play
games this way. Secondly, algorithms that do work this way get terrible
results against real (human) players, because their assumptions about
how the humans will play are all wrong. This led to this approach to
the analysis of multiplayer games being almost completely discarded.
David desJardins
Kevin J. Maroney
>In a lot of two player games, I think players do something similar to a
>pruned alpha-beta search: "If I do this, then he'll play that, and then
>I play..." - the "human" element is not the basic algorithm, but the way
>players decide which move their opponent will make.
But a human will generally quickly learn how to prune the tree in a
way that no computer program has yet learned to do. Even a novice
chess player won't consider moving her Queen's Rook Pawn when her
Queen is under attack, but a computer has to be explicitly told to not
bother down that road.
Geenius at Wrok
> Geenius at Wrok <gee...@albany.net> writes:
>
> > I'm no less accountable under a pseudonym than I am under my legal name.
>
> The difference isn't some sort of legal notion of accountability. It's
> that the people who choose to hide their identity have a reason for
> doing so: they don't want to be associated with what they say.
That's the sociological definition of accountability. And in any case,
using a pseudonym doesn't hide one's identity the way anonymity does --
there is still an identity to pin the words on. Does my calling myself
"Geenius at Wrok" prevent me from being associated with what I say? No
(and it wouldn't even if I didn't put my real name down at the bottom).
What it does do (or rather would do) is keep the stalkers away. Which can
be kind of nice if, for example, "Tristan Schaut" turns out to be a
celebrity.
To quote another pseudonymous poster, "Jesse Garon" of
alt.society.generation-x:
> How to establish an identity:
>
> 1. Create it.
> 2. Articulate it.
> 3. Propagate it.
In contrast, you can establish anonymity in just one step: Never allow
your words to be associated with anything people will see (or recognize) a
second time. That's hardly what Bagheera, Maverick et al. are doing.
Kevin J. Maroney
>> Yes, it is true, there will probably not be anything like a real AI
>> for games in the near future, but there are agents for games like
>> CivNet and Quake that are virtually unbeatable.
>
>You can't write a player for CivNet that is even at the level of a
>competent beginner, much less "virtually unbeatable".
Specifically, the AIs in Microprose's _Civ_ flat-out cheat: They don't
play by the same rules that the humans do.
The better bots in _Quake_ outperform humans by acting faster and more
precisely than humans, not "smarter". They can target and fire faster
than any human player can. But they can't solve puzzles, and to the
best of my knowledge, they don't do things like decoy firing,
rocket-jumping, or other weird HI tricks.
Kevin J. Maroney
>Hell, I see two names there... perhaps you should check your posts for
>accuracy after you send them to your typist! Or maybe your AI just
>needs tweaking...
David is making the point that, because he has chosen to post under
his real name, his claims about his non-Usenet activities can be
investigated by anyone curious enough to do so. He has a website at
his domain which provides material he has written about his various
fields of interest, including his computer science degree.
You, Maverick, have made your "true name" fairly well-known and don't
hesitate to provide evidence of your non-Usenet activities. You're in
the middle--you have a handle.
"Bagheera", on the other extreme, has not made his real name known on
Usenet and, to the best of my knowledge, has not provided any evidence
to back up his claims to AI programming expertise. (Checking his
website, I deduce that his real name is *probably* Jason Cordes, but
it took me a fair amount of digging to determine that.)
Bagheera
I should try that tactic ;)
so mr David desJardins, I'm so special I have my own ISP
to hide my real identity...
In article <vohaesw...@yuban.berkeley.edu>,
David desJardins <da...@desjardins.org> wrote:
> "Bagheera, the jungle scout" <bagh...@my-deja.com> writes:
> >> I'm confident you aren't going to find a single other person who
will
> >> seriously argue that interviewing an expert human game player, and
> >> using what the human tells you to write an algorithm that plays the
> >> game that way at an expert or even competent level, is remotely
> >> within the state of the art.
> >
> > it may not be "state of the art" but it works.
>
> As long as you have zero examples of this, and no one has ever
> successfully done it, and your argument is simply that it's very easy
to
> do but that no one has ever bothered, I think you are going to have a
> hard time convincing anyone to take that position seriously.
ai programmers model human decision processes all the time.
maybe you need to take a non-undergrad course in AI sometime...
The original chess programs used alpha-beta pruning techniques,
which both Glenn and I have demonstrated can be used by humans
in game playing, and can be used in effect as a fair player against
average opponents. This is a form of human intellect modelling, while
it does not ACCURATELY MODEL THE PRECISE THINKING PROCESS, it is A GOOD
APPROXIMATION.
I could discuss other implementations, but varies NDA's prohibit that.
If you can convince the game industry to life the NDA's they have, I
will gladly discuss games with you that use human-like decision making.
> >>> To get back to the original issue at hand, I was asking for people
> >>> to give suggestions for tactics on playing Acquire so that *I* or
> >>> one of my contemporaries could generate a "fuzzy logic" powered
> >>> agent that could play an "intelligent" game of Acquire that could
> >>> challenge an average or better player. I feel fully capable of
> >>> adapting REAL strategies into code for an agent.
> >
> > what you gave me were a serious of "hunches". You didn't actually
> > state a hard-fast strategy.
>
> Exactly. Humans don't play games by applying a fixed algorithm which
a
> computer can then apply step by step. If, when you interviewed an
> expert human player, and asked how the human decides what moves to
make,
> the human gave you a step-by-step deterministic algorithm for making
the
> decisions, then of course it would be easy to translate that algorithm
> into computer code and make a computer play just as well as the
expert.
You really are that stupid aren't you?
Most humans don't consider their thought processes linear, and
therefore cannot expound their ideas in that fashion.
In the case of another chaotic game (a game which relies more on luck
than skill), Can't Stop, the player must decide which set of rolls
she wants to use to advance markers on the board. Well, it's extremely
obvious that the player is going to choose the option which has the
highest odds for successful repetition (please see rules for Can't Stop,
if you can't find them, email me and I will post a summary on my
website). So you will see that she is choosing combinations which
include 6,7, and 8 over other pairs. So in her mind, she is selecting
the move which gives her the highest possible return value. This is
easily implemented, as per your instructions (which I convieniently
snipped). Well, now we get to the "zen" or "fuzzy" decision.
Does she press on, or stop rolling in fear that she will bust?
Well, most people do this on a hunch. When I play, I do it on
probability. Let's take the hunch player.
Okay, let's say the hunch player only rolls again if he has at least
one marker on 6 7 or 8. Or maybe "just once more" if he doesn't.
So let's say, it's a 80/20 toss up. Well, that's easy to code. But
what about the "sure thing" row? Well, the hunch player decides he will
roll again if he's real close to getting the row...okay, after half
point, increase chances. Before half point, consider position. If
other players are ahead, the hunch player will press his luck to the
halfway point, or as near the other players as he can get.
So we set up a fuzzy enumeration, "no, bad, dontwantto, maybe,
notperfect, go, perfect" no is enumerated if you have 2 3 and 12,
perfect is if you have 6 7 AND 8. Perfect rolls until one row hits
end give or take a 20% waiver factor.
Well, I personally believe this would make a damn good Can't Stop
player. It instills a queasy factor, and has some "gut-feeling" in it.
if the gut feeling is closer to 0, we don't go on. If the "gut-feeling"
is closer to 1, we go on. Shoot, we can even add a factor that reduces
the gut feeling after every roll to "emulate" the uneasy feeling
gamblers get after a long run of successes.
Again, I ask you think "outside the box". You obviously have no
faith in your skills (or obviously those of your contemporaries),
or just don't know good design practices.
> > Me? No, I can't. But, I have played against superior agents on
CivNet.
>
> No. You haven't. There aren't superior computer players for CivNet,
Just because you haven't eaten the maroon carrot doesn't mean it don't
exist. (maroon carrots can be requisitioned from Texas A&M commissary
College Station, TX).
> (I think you are probably confused and talking about computer players
> that in fact get a big advantage over the humans, and then gang up on
> the human players to boot. And these are still easily beatable. The
> only interesting question in playing against computer players is how
> fast one can win and with how high a score.)
You obviously have limited experience with CivNet. I think you play
CivII gold maybe? It is net playable, but is hardly CivNet. The group
of people I play CivNet with have clients (much like mud clients) that
allow them to run "botscripts" that play the game. A couple of them
are good enough to trounce the best Civ player I happen to know (whose
high score is 144% on Emperor level of original Civ, 160% on Deity of
CivII).
> > my statement also extended to Quake, where I know there
> > are "virtually" unbeatable agents (assassin, Knife, black wind, and
> > stilleto to name a few I know of) implemented and being used.
>
> Quake isn't a strategy game. In any game where reflex speed is the
> important factor, of course computers can beat humans. I can design a
> computerized boxer that can beat Mike Tyson, because when he hits it
> he'll break his hand. But this doesn't have anything to do with AI.
hardware and software are two different things.
And Quake is a strategy game, you just don't know how to play right
if you think otherwise. Yes, it does have some dexterity attribute
to it, but a smarter player will more often than not, beat an agile
player because they know where to hide, and when to hit hard, and when
to run away (and how to run away). Sniping isn't just for eBay anymore!
> No thanks. You can decide whether to put your name on your postings
or
> not. If you choose to post anonymously, I don't have any interest in
> playing detective to undo that decision. It's just that your
> contributions will receive the respect that that choice deserves.
>
> David desJardins
okay,
Jae, aka "astromutt", aka "polar bear", aka "Jake", aka "maddog", aka
"bagheera", aka "jaebear", aka "poohbear", aka "Malfoar", aka
"Steelgut", aka "Spock", aka "Splinter", aka "Jayce", aka "smartass",
aka "Prince Wilhelm", aka "Dzyeson Ivanovich", aka "Yakov Ivanovich",
aka "Will Grant", aka "baggy", aka "shaggy", aka "Shagster", aka
"Thunder", aka "BARD!", aka "MEDIC!", aka "GrUnT", aka "El Perro
Rabido", aka "AlphaDog", aka "Pack Leader", aka "21 of 42", aka
"ParserMan", aka "Code Guru", aka "Debugger Guy", aka "Evil One", aka
"Mama's little sugar", aka "Ambassador Jae", aka "Translator", aka
"TK-421", aka "James", aka "hunk", aka "lunk", aka "Henri"
Cordes.
there, happy, now you know who I am and the whole world can see who I am
and anyone who has ever run into me in my lifetime can identify me to
you. I don't think I missed any on that list. If I did, I'll be sure
to let you know, k?
David Vander Ark
windbags would have an accurate enough urine stream to put out the flame
war. That you haven't might be related to the difference between a
carbine and a rifle...a carbine is shorter and less accurate. If one of
you had pissed out the flame war it might suggest something about your
length, or accuracy. Please do the rest of us a favor and keep future
posts on the esoterics of AI in the AI group. Unless you're developing
a board game in which a computer controls bladder drainage through some
form of artificial intelligence.
Feel free to flame me too since I had the stupidity to enter into this
thread. In fact, I'll do it for you. I was a stupid jerk to enter this
worn out thread, and an even stupider one to call you guys windbags
(some of you I know to be OK people in real life), and it was childish
of me to comment on the perceived size of your genitalia.
Now save the boardgamers the trouble and this newsgroup the bandwidth
and go fight in the other group.
Thanks
David desJardins
> The group of people I play CivNet with have clients (much like mud
> clients) that allow them to run "botscripts" that play the game. A
> couple of them are good enough to trounce the best Civ player I happen
> to know (whose high score is 144% on Emperor level of original Civ,
> 160% on Deity of CivII).
We know this is not true.
David desJardins
Bagheera
David desJardins <da...@desjardins.org> wrote:
> Bagheera, the jungle scout <bagh...@my-deja.com> writes:
> > The group of people I play CivNet with have clients (much like mud
> > clients) that allow them to run "botscripts" that play the game. A
> > couple of them are good enough to trounce the best Civ player I
happen
> > to know (whose high score is 144% on Emperor level of original Civ,
> > 160% on Deity of CivII).
>
>
> David desJardins
Who is "we" and what are you basing your knowledge on?
The screenshot, the experience of sitting behind the guy
when he did it, or your own ineptitude at the game?
Derk Solko
> You, Maverick, have made your "true name" fairly well-known and don't
> hesitate to provide evidence of your non-Usenet activities. You're in
> the middle--you have a handle.
>
> "Bagheera", on the other extreme, has not made his real name known on
> Usenet and, to the best of my knowledge, has not provided any evidence
> to back up his claims to AI programming expertise. (Checking his
> website, I deduce that his real name is *probably* Jason Cordes, but
> it took me a fair amount of digging to determine that.)
I'll echo what others have said, "What the hell am I doing contributing
to this thread?" But I feel this is close enough to a real topic (if
perhaps off-topic for rgb), to warrant the time.
This is nothing new, this whole arguement: "Yeah? Well you're wrong
because you're posting under a pseudonym." But it is a form of ad
hominem, and not much better arguement than: "Yeah? Well you're a
poopiehead!"
With the advent of the modern 'net, does using your given name
necessarily lend accountability? I don't think so. By the guidelines
you and David outline, I'm a blithering idiot simply because I post via
Deja. While the blithering part is up for discussion, the real issue
is whether you believe me. And the only way I've found to determine
believeablity is by judging the content of what someone says over a
period of time.
For instance, if I read something written by Greg Schloesser, it's
probably about German gaming and it's written with honesty and
enthusiasm. If I read something by Maverick, it's probably about
wargames and it's his thoughts on the punched/mint status in auction.
If I read something written by desJardin, I know there's a kernel of
truth buried in a bunch of hair-splitting knee-jerk reactions. If I
read something from Kevin, he's probably trying to lend credence to
desJardin's less than user-friendly responses.
But the point is, there is no accountability because everything you
read could be faked, if someone really wanted to. The anonymity
arguement might hold water if someone started posting to rgb out of the
blue, but Jaebear/Bagheera/Jason isn't a new figure on this newsgroup.
I should clarify before I get ripped apart. Regarding Bagheer's AI
arguements, I have no idea about the validity of his AI references. If
I had to make a judgement, I'd say that his postings seem to take a
rather simplistic view (especially compared desJardin's nit-picking).
And I would probably disagree with some of his points for that reason
(most specifically the pruning of the a/b tree and simplistic fuzzy
logic), and NOT the fact that he's posting them via Bagheera.
derk aka derk.
David desJardins
>>> The group of people I play CivNet with have clients (much like mud
>>> clients) that allow them to run "botscripts" that play the game. A
>>> couple of them are good enough to trounce the best Civ player I
>>> happen to know (whose high score is 144% on Emperor level of
>>> original Civ, 160% on Deity of CivII).
>
It's not possible to produce a program that can play such games remotely
as well as you claim. "We" is anyone who is familiar with what has and
has not been accomplished in this field. They know that you can't
produce a program to play this sort of game that can do what you say
("trounce" a competent human opponent) without extensive handicaps or
other advantages.
It's theoretically possible that your "group of people" all have
accomplished something that is well beyond what anyone else has been
able to accomplish and publish (either in an academic setting, where it
would achieve great recognition, or in a commercial realm, where it
would have substantial commercial value). I don't think that many
people are willing to believe that the possibility is more than
theoretical, when you claim the existence of unrealistically strong
computer players which are conveniently unavailable.
If your computer really could play Civilization remotely this well, then
your aspirations for it to play Acquire wouldn't be unreasonable. But
it can't.
David desJardins
David desJardins
> By the guidelines you and David outline, I'm a blithering idiot simply
> because I post via Deja.
That certainly isn't what I said. I give exactly the same credibility
to all posters who put their names behind their postings. Whether a
given poster uses Dejanews, AOL, or any other service, is completely
irrelevant, as I haven't found it to be a useful indicator of anything.
> But the point is, there is no accountability because everything you
> read could be faked, if someone really wanted to.
Could be, could be. Anything could be. The fact is that the
overwhelming majority of signed postings are from exactly who they say
they are. So that's why people give them credibility based on that.
> For instance, if I read something written by Greg Schloesser, it's
> probably about German gaming and it's written with honesty and
> enthusiasm.
Why do you say that? Your analysis would imply that you have no idea
who it's really from, or what to expect. If you really don't think that
the name attached to the posting tells you anything about who actually
wrote it, since it "could be faked", then seeing the name "Greg
Schloesser" at the top of the posting doesn't tell you anything at all
about what to expect.
I observe that the anonymous posters generally don't deserve as much
respect as the posters who put their name behind their postings. It's
not a preconceived bias, it's an observation about many years of
postings on this and other groups. They tend to make many more
assertions that can't be checked. They often disappear and then
reappear under new aliases---why would people do that if they are
willing to be associated with their own writings?
Obviously there's some reason they don't want to be associated with
their writings. Otherwise they would use their names. It's really as
simple as that.
David desJardins
David desJardins
> Does my calling myself "Geenius at Wrok" prevent me from being
> associated with what I say? No (and it wouldn't even if I didn't put
> my real name down at the bottom).
Yes, of course it would. That's exactly why "Bagheera" et al. do it.
To make it more difficult for someone else, who knows their identity, to
associate their words with them.
David desJardins
Neil Carr
automaton run by this guy Ken who lurks on this group but refuses to
post. Every weekend at our gaming sessions Ken goes on and on about
all the wackos who inhabit this group and then gives instructions to
Derk to post deeply seating reactions to what goes on here.
The Maverick
>
> "Bagheera", on the other extreme, has not made his real name known on
> Usenet and, to the best of my knowledge, has not provided any evidence
> to back up his claims to AI programming expertise.
David's point was purely ad hominem (or perhaps more aptly, ad
pseudonym.) There is nothing that makes a looney's ideas more valid
just because they happen to be posted under a "real" name any more than
someone's ideas or experience are invalidated by the fact that they post
under an assumed name (as opposed to anonymous, which is something
different altogether.) Having been raised (as it were) with online BBS
systems where aliases were accepted and appropriate, I view it as
digital snobbery to discount someone on the basis of their choice of
e-mail address...
The Maverick
>
> Obviously there's some reason they don't want to be associated with
> their writings. Otherwise they would use their names. It's really as
> simple as that.
Or, they could have the same name as you and not want to be associated
with your writings? ;-)
In any case, I've made my point and will let the digital snobs roll on
undaunted.
Geenius at Wrok
I AM Geenius at Wrok. I am the only one. I'm also Keith Ammann, but I'm
not the only Keith Ammann, so if I say I'm Geenius at Wrok, I'm actually
giving you a BETTER idea of where my words are coming from than if I say
I'm Keith Ammann. By using a pseud, I've made it EASIER to associate my
words with me, by eliminating the chance that it will be associated with
the other Keith Ammann(s). What matters is whether the words come from a
recognizable source, not whether you can look up that recognizable source
in the phone book.
There was a guy at my high school named John desJardins. Maybe he's your
brother or cousin or something. DesJardins isn't all that rare a
surname, and David isn't all that rare a given name. I go onto InfoSpace
and turn up 35 people with your name, from California to Maine; who knows
how many more there may be in Canada, France and other French-speaking
countries? Now you're going to tell me that the signifier "David
desJardins" is easier to pin down to a single person than "Geenius at
Wrok"? Please! What if my real, legal, given name were "John Q. Smith"?
Would I be more credible posting under that name than if I'd adopted a
handle -- say "Q-Smif" -- that I used online without exception?
Online livin' poses some tricky questions when it comes to the nature of
identity; you can't apply 19th-century standards to 21st-century
communication. All you can do is tie all your balloons to one hook, and
if that hook happens to be "Geenius at Wrok" instead of "Keith Ammann,"
well, so what? If you're planning to write me a check, THEN we can
discuss what name the bank knows me by.
Geenius at Wrok
> Having been raised (as it were) with online BBS
> systems where aliases were accepted and appropriate, I view it as
> digital snobbery to discount someone on the basis of their choice of
> e-mail address...
Unless his alias is something like --->>>g0d|0f|EvErYtH!nG<<<---. Then
you know he's an idiot. :-)
Bagheera
David desJardins <da...@desjardins.org> wrote:
> If your computer really could play Civilization remotely this well,
then
> your aspirations for it to play Acquire wouldn't be unreasonable. But
> it can't.
>
> David desJardins
I've figured out that you, sir, are an idiot who lives in a cave.
The agent exists. If you want a copy of the agent ask CaveMan
on another.net IRC for a copy of the source, he may or may not
give it to you. He wouldn't give it to me.
If I had half his talent, or some of the others in our group
of friends, I probably wouldn't have a hard time writing an
AI for Acquire. Two things are holding me back though.
(1) against good players, I suck at Acquire.
(2) my friends don't share their source with me.
Rgtft
Game, set and match to David. His replies have been
logical and well reasoned, while you've dropped to
name calling.
RGT
Bagheera
That's fine. Since his logic is that he's lived in a cave his
whole life and therefore life must emulate his cave.
He's thrown some descriptions of simple ai algorithms around like
Alpha Beta, and brute force searching.
The situation, quite plainly, is that I have not the time nor
desire to get into a research battle with someone who isn't willing
to consider the possibility that something outside his paradigm exists.
While I understand his point of view, as I wouldn't believe in
transparent steel if I hadn't seen a 2' x 2' x 1" slab of it myself, it
doesn't make sense why he would take such a naive and high-handed
approach to a problem which is intuitively simple for an adequate AI
programmer. My only guess, which I only recently considered due to a
lengthy conversation with my friends on this issue, is that our friend
David is nothing more than a troll.
In which case, I must apologize for taking the troll bait, hook, line
and sinker.
I still maintain that it is possible to generate a competent AI agent to
play Acquire by simply implementing basic strategies of someone who is
an expert Acquire player. I could throw out fancy terms like neural
nets, and Phased Aggregate Transition Matrices, or even Subsumption
Elimination. These would mean almost nothing to the average
rec.game.board reader, and probably mean something to comp.ai.games
readers, but certainly not all.
The fact of the matter is, that just about anything you can think
can be programmed. It's complexity may be great, but it can be done.
There are only a few exceptions, such as the Halting problem, and other
classic Computer Science problems, but these are extreme examples and
can be ignored. Actually, I'm of the personal belief that the halting
problem can be solved, and I'm quite sure I'll win the Nobel Prize if
I ever really figure it out, or some other equally prestigious prize
at least. I, of course, have no such delusions of grandieur of actually
being able to come up with this solution, but everyone has to have a
goal, right?
I may not say everything I say correctly, but there is truth behind
everything I say.
or in the words of a very good friend, "I reserve the right to be
wrong. But any attempts to prove the matter will be considered libelous
and/or slanderous and pursued accordingly."
--
Bagherra <jaebear @ frenzy.com>
http://www.frenzy.com/~jaebear
"A little nonsense now and then
is cherised by the wisest men."
Gerry Quinn
>In article <vohn1ww...@yuban.berkeley.edu>,
> David desJardins <da...@desjardins.org> wrote:
>> If your computer really could play Civilization remotely this well,
>then
>> your aspirations for it to play Acquire wouldn't be unreasonable. But
>> it can't.
>>
>> David desJardins
>
>on another.net IRC for a copy of the source, he may or may not
>give it to you. He wouldn't give it to me.
>
>If I had half his talent, or some of the others in our group
>of friends, I probably wouldn't have a hard time writing an
>AI for Acquire. Two things are holding me back though.
>(1) against good players, I suck at Acquire.
>(2) my friends don't share their source with me.
>
They don't have source that does what you say. Civilisation is a
strategy game, unlike Quake or even Chess (which is a mix of strategy
and tactics). Strategic agents such as you describe don't exist. (For
a start, they would need to have diplomatic and agent-modelling
capabilities far in excess of any current AI.)
- Gerry Quinn
http://bindweed.com
Richard Wein
Bagheera ; the jungle scout wrote in message
<7mrog2$rcp$1...@nnrp1.deja.com>...
>While I understand his point of view, as I wouldn't believe in
>transparent steel if I hadn't seen a 2' x 2' x 1" slab of it myself, it
>doesn't make sense why he would take such a naive and high-handed
>approach to a problem which is intuitively simple for an adequate AI
>programmer. My only guess, which I only recently considered due to a
>lengthy conversation with my friends on this issue, is that our friend
>David is nothing more than a troll.
I've generally found David's posts to be well-reasoned and informative, and
I don't believe for a moment that he is a troll.
The mistake you seem to be making is believing that what is intuitively
simple for a human being must be simple to program. I'm no expert on AI, but
I am perceptive enough to realise that this is not the case.
David listed some of his Acquire strategies, and you dismissed some of them
as too "speculative" to be of use. No doubt they are of no use to a computer
AI, which lacks the intuition to use them, but they *would* be useful to a
human Acquire player.
To put it another way, it isn't possible (in general) to model a computer AI
on human thought processes, because human intuition is very poorly
understood.
Perhaps, when you talk about human thought processes, you're only
considering *conscious* thought processes. It may well be possible to model
those. But a large part of human strategy formulation and decision making is
*unconscious*.
Richard Wein (Tich)
--------------------------------
Please send email to <insertmyni...@primex.co.uk>, replacing the
first part with the word "tich" (antispam device). See my web pages for
multiplayer PBEM games and EFS Nova at http://homepages.primex.co.uk/~tich/
Dennis Ugolini
> In article <7mq6c7$d97$1...@nnrp1.deja.com>, Bagheera, the jungle scout
> >I've figured out that you, sir, are an idiot who lives in a cave.
>
> logical and well reasoned, while you've dropped to
> name calling.
>
> RGT
???
Are you joking? The last four posts before the one above have essentially
looked like this:
"Well, I have a friend with a CivNet program that beats the best human I know."
"No you don't."
"What? Yes I do!"
"No you don't."
I seen more logic and reason in how my puppy attacks tennis balls. This
argument can't die fast enough.
Dennis Ugolini
ugo...@leland.stanford.edu
Klaas
>
> In article <7mq6c7$d97$1...@nnrp1.deja.com>, Bagheera, the jungle scout <bagh...@my-deja.com> wrote:
> >In article <vohn1ww...@yuban.berkeley.edu>,
> > David desJardins <da...@desjardins.org> wrote:
> >> If your computer really could play Civilization remotely this well,
> >then
> >> your aspirations for it to play Acquire wouldn't be unreasonable. But
> >> it can't.
> >>
> >> David desJardins
> >
> >on another.net IRC for a copy of the source, he may or may not
> >give it to you. He wouldn't give it to me.
> >
> >If I had half his talent, or some of the others in our group
> >of friends, I probably wouldn't have a hard time writing an
> >AI for Acquire. Two things are holding me back though.
> >(1) against good players, I suck at Acquire.
> >(2) my friends don't share their source with me.
> >
>
> They don't have source that does what you say. Civilisation is a
> strategy game, unlike Quake or even Chess (which is a mix of strategy
> and tactics). Strategic agents such as you describe don't exist. (For
> a start, they would need to have diplomatic and agent-modelling
> capabilities far in excess of any current AI.)
I find it hard to believe that there are NO tactics in Civ. What about
temporarily putting a unit on a boat to flee an enemy?
-Mike
the Dave
>can be programmed.
You may think it can be programmed, but there are many of things that
the thousands of people who have worked on AI have not been able to
program. Believe it or not, the thus-far negative results from those
thousands of other people carry just a bit more weight than your
optimism and hand-waving.
--
Bay Area Games Day VII is Sat. July 31 : http://www.io.com/~dk2/gamesday
Rgtft
(Dennis Ugolini) writes:
>In article <19990717201137...@ngol04.aol.com>,
>rg...@aol.comREMOVE (Rgtft) writes:
>> <bagh...@my-deja.com> writes:
>> >I've figured out that you, sir, are an idiot who lives in a cave.
>>
>> logical and well reasoned, while you've dropped to
>> name calling.
>>
>> RGT
>
>???
>Are you joking? The last four posts before the one above have essentially
>looked like this:
>
>"Well, I have a friend with a CivNet program that beats the best human I
> know."
>"No you don't."
>"What? Yes I do!"
>"No you don't."
>
>I seen more logic and reason in how my puppy attacks tennis balls. This
>argument can't die fast enough.
No I wasn't joking; but maybe your puppy should join the thread. :-)
RGT
Stephen Tavener
>>In a lot of two player games, I think players do something similar to a
>>pruned alpha-beta search
Kevin Maroney wrote:
>But a human will generally quickly learn how to prune the tree in a
>way that no computer program has yet learned to do.
Fair comment - a poor old computer has to consider many more cases than
an experienced human player would, to play at a similar level. It's a
good job computers are faster than humans these days! I reckon,
however, that there are advantages to searching a wider part of the
tree...
>Even a novice
>chess player won't consider moving her Queen's Rook Pawn when her
>Queen is under attack, but a computer has to be explicitly told to not
>bother down that road.
I realise you used a random example here, but it serves to illustrate
the point that a wide search does have it's advantages. Here's a very
simple situation where the best move *is* to advance the Queen's pawn.
In this case, I'm sure that even a beginner would realise they had a
win. With a little thought, I'm sure that some of the better chess
players on this list could come up with a more believable scenario!
+---+---+---+---+---+---+---+---+
| | | | | | |*K*| |
+---+---+---+---+---+---+---+---+
| P | | | | |*P*|*P*|*P*|
+---+---+---+---+---+---+---+---+
| | | | | | | | |
+---+---+---+---+---+---+---+---+
| | | | | | | | |
+---+---+---+---+---+---+---+---+
| | | | | | | | |
+---+---+---+---+---+---+---+---+
| | | | | | K | | |
+---+---+---+---+---+---+---+---+
| |*P*| | | | | | |
+---+---+---+---+---+---+---+---+
| Q | N | | | | | | |
+---+---+---+---+---+---+---+---+
TTFN,
Stephen
--
Stephen Tavener | There is no such thing
Games bought,sold,traded,played | as "just a cat"
http://www.scat.demon.co.uk/ | - Tanya Huff
Stephen Tavener
>> In a lot of two player games, I think players do something similar to a
>> I play..."
>I guess our difference is in our interpretation of "like".
Quite possibly...
>Alpha-beta
>search uses precise numerical cutoffs: if a value is a fraction higher
>than another value, then a whole subtree will be cutoff, while if it's a
>fraction lower, then a large subtree may be searched.
Maybe. I don't know what the standard is for these things, but I do
know that when I wrote an AI, I actually used probabilities - the
probability of searching a subtree was equal to (score-{worst
score})/({best score}-{worst score}) - the program wasn't, perhaps, the
best player in the world, but I'd still call this an alpha-beta
search... as is a full-depth search...?
>but they can't efficiently discover
>that certain lines of analysis apply to many very similar but not
>identical positions
>And, of course, humans only consider a very tiny fraction of the
>positions that any alpha-beta search would consider
Agreed on both counts. As I say in my response to Kevin Maroney's e-
mail, it's a good job computers are faster than humans these days!
The issue of computers searching one subtree before moving on to the
next one isn't relevant, in my opinion - a program *could* be written
that added each position to the end of a stack, and considered board
positions in a different order, switching between subtrees with each
board, it's just more efficient in terms of memory to search one subtree
fully before moving on.
>> the "human" element is not the basic algorithm, but the way
>> players decide which move their opponent will make.
>See above for a strong disagreement.
Well, let's agree that humans differ from computers in both their
pruning algorithms and their evaluation algorithms. I'm not convinced
that this means humans aren't alpha-beta'ing.
Derk Solko
> > For instance, if I read something written by Greg Schloesser, it's
> > probably about German gaming and it's written with honesty and
> > enthusiasm.
>
> Why do you say that? Your analysis would imply that you have no idea
that
> the name attached to the posting tells you anything about who actually
> wrote it, since it "could be faked", then seeing the name "Greg
> Schloesser" at the top of the posting doesn't tell you anything at all
> about what to expect.
Huh? By my analysis, I'm implying that it's immaterial whether I'm
reading something from a psuedonym'ed poster or a named poster. The
name on a post is merely a label that gives me by which to judge the
contents, regardless of how the name is derived. I've learned much of
my 'netiquette from mudding for several years. On muds, everyone used
another name, and the only way to tell if someone was full of it was to
talk to them over a long-ish period of time and judge the content of
their words as a whole.
My whole entire point to my response was to say that you really can't
judge someone solely on their choice of using a psuedonym. Your
assumption that a person is full of poop simply because they choose to
have a handle is faulty, and I've seen you bring it into arguments
before (Geenius comes to mind first).
> I observe that the anonymous posters generally don't deserve as much
> respect as the posters who put their name behind their postings. It's
> not a preconceived bias, it's an observation about many years of
> postings on this and other groups. They tend to make many more
> assertions that can't be checked. They often disappear and then
> reappear under new aliases---why would people do that if they are
> willing to be associated with their own writings?
I agree with the spirit of what you're saying here. That the fly by
night b...@gotohell.com's aren't worthy of getting your dander up. But
that is simply not the case here.
> Obviously there's some reason they don't want to be associated with
> their writings. Otherwise they would use their names. It's really as
> simple as that.
They are associated with their writings. The only difference is they
chose to use a name other than their given name. I don't see a problem
with that.
derk.
Derk Solko
"Stay away from that curtain, you little dog!!"
Dammit, Neil! Now the truth's out and I'll have to start posting my
real thoughts, which aren't deep enough to fill a thimble. Oh well, at
least now I won't have to be Ken's little finger puppet anymore. Yeah,
I'm free, I'm free!!
On the other hand, I'm should be inclined to disbelieve everything
you've said in this post, simply because you chose such a stupid email
address. Bah, what the heck kinda Star Wars freak uses a psuedonym
like Echoota and writes detailed meta-philosophical dissertations
regarding the distinctions between geek-dom and nerd-dom? Get real!
Kevin J. Maroney
>Unless his alias is something like --->>>g0d|0f|EvErYtH!nG<<<---. Then
>you know he's an idiot. :-)
Well, in fact, at this point I know that someone who uses "Bagheera"
as a handle is an a flamer with no posting value; he's established
that part of his identity quite well.
--
Kevin J. Maroney | Crossover Technologies | kmar...@crossover.com
Games are my entire waking life.
Kevin J. Maroney
>I've figured out that you, sir, are an idiot who lives in a cave.
>The agent exists. If you want a copy of the agent ask CaveMan
>on another.net IRC for a copy of the source, he may or may not
>give it to you. He wouldn't give it to me.
In other words, your refutation of Dave's point is that someone who
might or might not exist has a program which he won't give to you
which might or might not do what you claim.
Baldrick and Percy couldn't argue it better.
Nick Wedd
deja.com> writes
>By my analysis, I'm implying that it's immaterial whether I'm
>reading something from a psuedonym'ed poster or a named poster. The
>name on a post is merely a label that gives me by which to judge the
>contents, regardless of how the name is derived. I've learned much of
>my 'netiquette from mudding for several years. On muds, everyone used
>another name, and the only way to tell if someone was full of it was to
>talk to them over a long-ish period of time and judge the content of
>their words as a whole.
If you apply for a job, your potential employer might scan usenet for
posting by you. Some people are influenced by this when they post.
If your name is "David desJardins", a potential employer is likely to
come to the conclusion that you are intelligent, and do not suffer fools
gladly. If your name is whatever "Bagheera"'s real name is, he may
well find nothing.
When I read a posting from desJardins, I am aware that I am reading the
views of someone who is willing to stand behind those views. When I
read one from Bagheera, I am aware that I am reading the views of
someone who, for whatever reason, is not willing to stand behind those
views. I therefore tend to discount those of Bagheera.
Nick
--
Nick Wedd ni...@maproom.co.uk
Geenius at Wrok
> Bagheera, the jungle scout <bagh...@my-deja.com> wrote:
>
> >I've figured out that you, sir, are an idiot who lives in a cave.
> >The agent exists. If you want a copy of the agent ask CaveMan
> >on another.net IRC for a copy of the source, he may or may not
> >give it to you. He wouldn't give it to me.
>
> In other words, your refutation of Dave's point is that someone who
> might or might not exist has a program which he won't give to you
> which might or might not do what you claim.
And desJardins' counterpoint is that that person doesn't have it. Not
much better.
George Crawshay
Having looked through these diatribes with mounting astonishment, all
I can feel is a sort of awe at the thought that my unassuming little
Acquire program has spawned it all !!!!!
George Crawshay
Andrew Hardin
>Actually Derk doesn't really exist, instead he's an elaborate
>automaton run by this guy Ken who lurks on this group but refuses to
>post. Every weekend at our gaming sessions Ken goes on and on about
>all the wackos who inhabit this group and then gives instructions to
>Derk to post deeply seating reactions to what goes on here.
>
Shhhhhh, Neil! The truth about Derk must never be revealed, for the
group is not yet ready.
- Drew
Gerry Quinn
>> They don't have source that does what you say. Civilisation is a
>> strategy game, unlike Quake or even Chess (which is a mix of strategy
>> and tactics). Strategic agents such as you describe don't exist. (For
>> a start, they would need to have diplomatic and agent-modelling
>> capabilities far in excess of any current AI.)
>
>I find it hard to believe that there are NO tactics in Civ. What about
>temporarily putting a unit on a boat to flee an enemy?
>
Of course there are some, but the strategy aspects dominate. Not to
mention the diplomatic aspects. I could almost believe an AI could play
a 2-player game well, but how on earth could it predict the actions of
humans in a multiplayer game? How could it maneuvre to keep two
opponents fighting? We have no AIs like that.
Edge
A Civ AI would need to know how hard to "lean" on the leader(s), and how
to calculate the space it "deserves" to stake out for itself.
Both of these seem imminently doable with a rules based program. It
wouldn't fit the academic definition of AI, but it could play the game
pretty well.
Certainly, such a program could fill in a spot or two in a multi-player
game and do justice to the position.
I've sketched out some of the algorithms in the past, to deal with split
positions, isolated cities, and the like, although I've never coded it.
You do a lot of graph searching, but that runs fast and is easy to
code. <shrug> Doesn't seem so hard, as long as the coder really knows
the game.
Edge
--
Edward "Edge" Montgomery
remove NOSPAM to reply
Kevin J. Maroney
>On Mon, 19 Jul 1999, Kevin J. Maroney wrote:
>> In other words, your refutation of Dave's point is that someone who
>> might or might not exist has a program which he won't give to you
>> which might or might not do what you claim.
>
>And desJardins' counterpoint is that that person doesn't have it. Not
>much better.
I think David's counterpoint is more like "If such a thing existed, it
would be such a tremendous breakthrough in game AI that its existence
would be widely trumpted throughout the land."
Bagheera is making the exceptional claim, not David.
Bagheera
kmar...@crossover.com wrote:
> Geenius at Wrok <gee...@albany.net> wrote:
>
> >On Mon, 19 Jul 1999, Kevin J. Maroney wrote:
> >> In other words, your refutation of Dave's point is that someone who
> >> might or might not exist has a program which he won't give to you
> >> which might or might not do what you claim.
> >
> >And desJardins' counterpoint is that that person doesn't have it.
Not
> >much better.
>
> I think David's counterpoint is more like "If such a thing existed, it
> would be such a tremendous breakthrough in game AI that its existence
> would be widely trumpted throughout the land."
>
> Bagheera is making the exceptional claim, not David.
Things are not always as loud as you would expect.
The Chinese had the first rockets, but we usually don't
think of their existence until Germany in WWII.
The first movable type printing press was made in ancient
Rome. We attribute it to Gutenberg.
The first coin operated vending machine was a roman invention,
as were the taxi meter, the fast food restaurant, and automatic
sliding doors.
(*note I may be mistaking roman for greek here*).
An item's herald is when that item has a significant impact for
a wide cross-section of society.
Recent inventions: ion propulsion drives, transparent steel,
teleportation devices based on "spooky action", flying cars,
gravity well engines, plasma engines, pulse plasma torches,
these are all mostly considered science fiction, but all exist
RIGHT NOW, they may be too expensive for the casual person to
buy one off the shelf (Air cars are estimated at a cool $3million
each, to drop to $200-300K within 5 years), but they DO EXIST!
Just because my friend's AI isn't heralded as the Holy Grail of
programming, doesn't mean it does not exist. Perhaps it is because
he doesn't want to sell it. Perhaps he just isn't interested in
being pidgeonholed as an AI programmer (which he doesn't...he's a
demo coder, and will be happy to scream it to you if you ask).
*IF* I knew how it worked I would be happy to expound.
And as previously stated in this thread, it is possible that
the casual observer wouldn't consider it "academic AI".
It does exist.
And if we could please get back to the object of this thread
and away from the pointless argument of the existence of
golden bullets, I think it would be better for everyone.
--
Bagherra <jaebear @ frenzy.com>
http://www.frenzy.com/~jaebear
"There's a snake in my boot!"
ddnguyen
Bagheera ; the jungle scout wrote in message
(chop wow chop)
>Just because my friend's AI isn't heralded as the Holy Grail of
>programming, doesn't mean it does not exist. Perhaps it is because
>he doesn't want to sell it. Perhaps he just isn't interested in
>being pidgeonholed as an AI programmer (which he doesn't...he's a
>demo coder, and will be happy to scream it to you if you ask).
>
>*IF* I knew how it worked I would be happy to expound.
>And as previously stated in this thread, it is possible that
>the casual observer wouldn't consider it "academic AI".
>It does exist.
>
>And if we could please get back to the object of this thread
>and away from the pointless argument of the existence of
>golden bullets, I think it would be better for everyone.
>
>--
>Bagherra <jaebear @ frenzy.com>
>http://www.frenzy.com/~jaebear
> "There's a snake in my boot!"
Is there any independent verfication on your friends AI in action? Perhaps
he has a website you could point us too? Im not intresting in flaming
anyone, im asking in sincere curisoty. (btw my spell checker is broken!
please forgive the spelling mistakes, the grammatical mistakes are all mine
though 8^) ),
-ddn
Bagheera
"ddnguyen" <ddng...@mindspring.com> wrote:
<snip my post>
> Is there any independent verfication on your friends AI in action?
Perhaps
> he has a website you could point us too? Im not intresting in flaming
> anyone, im asking in sincere curisoty. (btw my spell checker is
broken!
> please forgive the spelling mistakes, the grammatical mistakes are all
mine
> though 8^) ),
>
> -ddn
Dave's words, not mine:
"Webpages are penis toys for the inept and insecure.
I have a girlfriend for the first, and a teddy bear for the second."
This was how he responded several months ago when I asked when he would
make his own webpage. He is softening to the idea, but right now it's
all pictures of his girlfriend, his hedgehog, and his Delorean.
I'll poke him to see if he'll make it public readable and post his
agent. I don't think he will though. He won't even let people into
his room to see the artwork he draws (well, except for his girlfriend).
Maybe I could train her to be a hacker and swipe the files for me to
post :)
--
Bagherra <jaebear @ frenzy.com>
http://www.frenzy.com/~jaebear
"There's a snake in my boot!"
David Damerell
>programming, doesn't mean it does not exist.
On the other hand, you yourself have not verified that it exists.
--
David/Kirsty Damerell. dame...@chiark.greenend.org.uk
CUWoCS President. http://www.chiark.greenend.org.uk/~damerell/ Hail Eris!
|___| fak...@fowler-schocken.culture.dotat.at is not my email address,|___|
| | | and email sent to it will be assumed to be spam and blocked. | | |
Kevin J. Maroney
>Just because my friend's AI isn't heralded as the Holy Grail of
>programming, doesn't mean it does not exist.
The fact that no one, including you, has seen it in action might make
you hesitate to use it as the key supporting piece of evidence in an
argument in which you are calling your opponent a miserable wretch for
daring to disbelieve you.
But, of course, this is Usenet. Many people would like to pretend that
all discussion on Usenet should be conducted in such a way.
Mickael Pointier
:I'll poke him to see if he'll make it public readable and post his
:agent. I don't think he will though. He won't even let people into
Well, perhaps we can thrust a guy that have a Delorean.
Perhaps he has AI technologie that comes back from the future...
Mike
David desJardins
> Just because my friend's AI isn't heralded as the Holy Grail of
> programming, doesn't mean it does not exist.
Indeed. It might be the case that you have a space alien in your
basement. It might be the case that you have a goose that lays golden
eggs. It might be the case that your anonymous friend has produced a
spectacular breakthrough in artificial intelligence for games, which no
one else knows anything about.
In all three cases, I'll believe it when I see it.
David desJardins
>mag<
David desJardins wrote:
> Indeed. It might be the case that you have a space alien in your
> basement. <snip>
Gee... I thought *I* was the only one with a space alien in my basement
! :P
I've missed some of this thread, so was there any ending to the saga ?
Was there a viewable bit of s/w, paper on the subject etc to show ?
Mike
Bagheera
> Perhaps you are correct. However, in a dynamic environment of a
> real-time strategy game, the "best" site to build a particular
> building (just an example of one problem that such an AI must solve)
> may change with each clock tick.
>
> How would any pre-engineered pattern-matching engine deal with such
> volitility? And do so within a very limited time. Notice I'm not
> saying it can't be done, but that if there were more time, it probably
> could be done better.
Well, the original intent of this thread was to examine AI for turn
based games, not real time strategy games. Specifically, *I* personally
am interested in writing AI for computerized board games. Due to the
nature of most board games, the AI has a considerable amount of CPU time
with which to generate its response. (if you think back to the ATARI
2600, some players were willing to wait 2 hours for the chess cartridge
to make one move!)
Also, the second point (of which this branch of this thread has come
from) is that the AI I used to demonstrate individual "state of the art"
was a script written for CivNet, a turn based strategy game. The AI
takes about 5 minutes to generate some responses (upwards to 15-20
minutes for some endgame moves), but the point was that this thing
exists.
It was Mr. David desJardins who brought the complaint that such an AI
could not exist, and an expert AI for condition one could not be
generated. Especially so from examining input from skilled players of
the game that you are writing the agent for.
Now, I believe there has been enough discussion in this thread to
satisfactorily prove that such equipment and knowledge exists to write
a skilled AI agent to play a turn based game, and to write one such that
good players can be challenged. Also, I believe that my theory that the
reason such an AI does not exist in most commercial games is the lack of
funding, has been more than adequately stated.
I maintain my position, a STRONG AI can be written for simple turn based
games, either by an extensive expert system, or by mapping strategies of
a strong player to an AI agent (via a rule based expert system).
I also maintain that such an AI could (and does exist), but is not
presented publicly due to lack of economic interest for such a thing.
And that, is the state of the art.
--
Bagherra <jaebear @ frenzy.com>
http://www.frenzy.com/~jaebear
"The only valid statements are those
which are empirically verifiable." - some famous philosopher
Steve Okonski
> Now, I believe there has been enough discussion in this thread to
> satisfactorily prove that such equipment and knowledge exists to write
> a skilled AI agent to play a turn based game, and to write one such that
> good players can be challenged. Also, I believe that my theory that the
> reason such an AI does not exist in most commercial games is the lack of
> funding, has been more than adequately stated.
>
(Since I was responsible for starting this thread by discussing Acquire,
I'll jump in again.)
In addition to a lack of funding, another reason the AI is poor in most
games is that humans prefer to beat the machine. If you create a powerful
AI, many people will shun the game.
I've observed this with my RB Player software which contains AI to play AH's
Rail Baron boardgame. The AI is tough, and so far no expert human player
has been able to win much more frequently than expected (i.e. 25% in a
4-player match, 20% in a 5-player match, etc.). Several less experienced
players have complained that the AI cheats (via rigged dice rolls, etc.),
but I wrote the code, and know the AI is totally cheat free. After I reply
with this information, the players generally stop playing the game (perhaps
they think I'm lying?).
Now, I'm seriously considering dumbing down the AI to make it easier to
beat. Presently, RB Player lets you choose from five different built-in AI
types, four of which are excellent players, one of which is clueless. You
can also create additional AI players and configure their characteristics,
but few people seem to want to fuss that much. Therefore, I hope to add a
few built-in intermediate AI players.
-Steve Okonski
Rail Baron Fanatics http://www.insystem.com/rbp/
Patrick Carroll
>games is that humans prefer to beat the machine. If you create a powerful
>AI, many people will shun the game.
Indeed. Speaking for myself, I've found the AI in most computer games I've
played to be sufficiently challenging. I've run across a couple games in which
the AI is a pushover (even for a lazy player like me); and yet I still play
those games sometimes, and find it relaxing and satisfying.
OTOH, I've run across games where the AI is (for me) absolutely unbeatable.
For instance, the Mind Games versions of chess and 5-in-a-row, now that I've
moved them to my Pentium machine. Even on Novice level, the chess program
beats me every time; I don't stand a chance. So I avoid that game like the
plague. Don't know why I haven't erased it from my hard drive yet. It seems
almost masochistic to click on that game's icon.
A program like Chessmaster 5500 can also beat me soundly. But it's much more
adjustable. I can set it to a manageable level, where it becomes enjoyable for
me to play (i.e., where I stand a decent chance of winning, provided I pay
attention to a reasonable and comfortable degree).
>Several less experienced
>players have complained that the AI cheats (via rigged dice rolls, etc.),
>but I wrote the code, and know the AI is totally cheat free.
I can relate to that too. I regularly accuse AIs of cheating when they beat me
too often. My wife is perhaps even worse about doing that. But when she
complains about cheating while I'm looking on, it's pretty clear to me that she
has no good reason to suspect cheating. And when I'm playing, I probably don't
have any sound reason for the accusation either. It just makes me feel better.
If you can't win, the next best thing is to upset the game board. Or shoot
your opponent for cheating, or your partner for being stupid. Anything to
point the finger of blame somewhere else & thus save face.
A couple domino sites I've enjoyed:
General info & links:
http://xs4all.nl/~spaanszt/Domino_Plaza.html
Windows downloads & more:
http://members.aol.com/DominoPage/
Don Woods
> Presently, RB Player lets you choose from five different built-in AI
> types, four of which are excellent players, one of which is clueless. You
> can also create additional AI players and configure their characteristics,
> but few people seem to want to fuss that much.
I haven't played RBP for a while now, but back when I was addicted to it
I tried creating various fiddled AIs. One of the more interesting ones
I came up with was one that placed extremely high weight on upgrading to
the Superchief, so it would forego purchases after its first 1-3 deliveries
in order to save up for the 'chief.
Naturally, if one such AI got matched against several more normal strategies,
it did very badly, since by the time it gets the 'chief AND makes additional
deliveries to build up money again, all the good lines have been snapped up.
But I found that if I played in a 5-6 player game where ALL the AIs were the
save-for-super variety, I often got trounced! I would usually be able to
grab some good lines (I only once failed to get the PA), but I couldn't
usually get a solid enough position (PA+NYC, or PA+ATSF, etc.) before I found
myself facing 4-5 opponents all of whom were making money 50% faster than me.
I thought it said something about how "group-think" could affect the game.
I also wonder whether it means saving for a Superchief is a valid strategy
in a 3-player all-human game.
-- Don.
-------------------------------------------------------------------------------
--
-- Don Woods (d...@navilinks.com) NaviLinks provides real-time linking.
-- http://www.navilinks.com/~don I provide personal opinions.
--
Robert C Ames
> In addition to a lack of funding, another reason the AI is poor in most
> games is that humans prefer to beat the machine. If you create a powerful
> AI, many people will shun the game.
I'm reminded of an old game named 'tic' ... "For those of you frustrated
with regular computer tic-tac-toe, which always beats you, I have created
'tic' ... Gameplay is simple- the first one to get one in a row wins, you
go first..." ;^)= It was a fun, positive reinforcement game, and I'm sure
the AI played flawlessly, I just never saw it. :^)=
> Now, I'm seriously considering dumbing down the AI to make it easier to
> beat. Presently, RB Player lets you choose from five different built-in AI
> types, four of which are excellent players, one of which is clueless. You
> can also create additional AI players and configure their characteristics,
> but few people seem to want to fuss that much. Therefore, I hope to add a
> few built-in intermediate AI players.
Expert AI's are great IMHO, but I'd hate to play them in a game I'm not an
expert at. It all depends on your target audience- the more difficult (or
easy) the AI and the less fun the game, the less people will want to play
it. There are games where I lose all the time but I still play because
they're fun.
A good example of this is "ADOM", a roguelike game that probably has no AI
to speak of. I play, walk to the dungeon, kill some stuff and die. Next
time, I beat that dungeon, walk to the next one and die. Next time, I'll
die in the first dungeon again. I've never "beat" the game, not even come
close, but I know that when I died, it was because of something that I
did wrong, and not because the computer is "that much better than me".
Maybe that should be the definition of a "human-like" game- if you
die (or lose), but can see where you made your mistake, then it is a
sufficiently "human" algorithm. Because it exploited *your* weakness, not
*it's* strength, it could be considered ~human~. And it would also be
nice if you could learn from an AI- "Hey dummy, next time don't send all
your Laser Tanks to my side of the board, because I'll just crush your
undefended base with my Air Strikes"
I'm O.K. with losing repeatedly to an AI (or a human player) just so long
as I can get feedback and tips for improvement. Doom was good for this,
because you'd run through a room and count up how much damage you took.
Thus you quickly learned a way to minimize your damage taken. That sounds
suspiciously like a simple "heuristic" for a GA/NN, right?
Another observation- I guess that's where MS is going with "help-agents"
in their Office Suite- it watches you and gives you tips on how to do
better. It may be wrong sometimes, but it's better than nothing.
Anyway, enough of this meandering rant for me- thanks for reading! :^)=
--
--Robert (ra...@utdallas.edu)
"Words are, of course, the most powerful
drug used by mankind" -- Rudyard Kipling
Yancarlo Ramsey
>
> In comp.ai.games Steve Okonski <inte...@insystem.com> wrote:
> > In addition to a lack of funding, another reason the AI is poor in most
> > games is that humans prefer to beat the machine. If you create a powerful
> > AI, many people will shun the game.
[snip]
> Expert AI's are great IMHO, but I'd hate to play them in a game I'm not an
> expert at. It all depends on your target audience- the more difficult (or
> easy) the AI and the less fun the game, the less people will want to play
> it. There are games where I lose all the time but I still play because
> they're fun.
All the game's I've played that allowed you to set the "difficulty
level" would give the computer-controlled players bonuses (and sometimes
give you penalties) when set to "hard" difficulties. I have never played
a game that specifically allowed you to adjust the level its AI would
play at, rather than giving the AI players an advantage and leaving
their AI alone.
For games where the AI takes a significant amount of time to make its
decisions, such as Civilization (and by the way, TA:K and MYTH are NOT
the 'same type of game' as Civ), the amount of time the AI is given to
make its decisions could be adjusted, rather than other factors.
I have a strong dislike for increasing a game's difficulty artificially,
by letting it 'cheat' (make use of information a human player would not
have) or by giving it unfair advantages (Master of Orion 2, on hard and
impossible difficulties - although I do like the game, and I play it on
AVERAGE despite that being not too hard to defeat).
For a turn-based game, I would suggest that changing the structure of
the game could free up a great deal of time for the AI to be making
decisions. In Civ and Alpha Centauri, all the computer players have to
take their turns BETWEEN one of the players turns and the next. Creating
a game structure where the AI players could do their turn WHILE the
human player is doing its turn would allow the AI to take as much time
as the human player ends up taking to decide on its moves. However, this
would, as I said, require a game structure where each player makes their
moves based on the same overall game situation that all the other
players are basing their moves on for each turn, in parallel - rather
than the structure where player one goes, and player two bases its
decisions on player one's moves, and player three's decisions are based
on what players one and two did, in a sequential manner.
This would require that, instead of a player making its moves and
immediately seeing the results, the player would select its moves, and
then once the other HUMAN players were done selecting their moves, all
the moves are done at once somehow, and the results displayed as the
situation for the beginning of the next turn. So that the AI players
could stop processing as soon as the human player(s) is(are) done, they
could do their processing in stages, and could carry over processing
from one turn to the next on units that have yet to arrive at their
destinations, or are being produced and will NEED a destination when
finished, such as a site for city construction.
The AI would be constantly spending some of its time figuring out good
locations for cities, and weighting them by how good the location is
(not just in terms of available resources, but also strategic placement,
including coastal access, proximity to enemy territory, and defensive
value of terrain). If the AI were to find an extremely good location for
a city that was right next to enemy territory (in other words, within
reach of an opponent, such that you want to grab it before THEY can), or
if its explorations had revealed an abundance of useful unoccupied
territory, it might base its decisions on whether to build new cities,
how quickly it builds new cities, and the locations of those cities on
that previously-calculated information.
Another fraction (not to imply that it would be either a large or small
fraction) would be spent on deciding what strategy it should use in
dealing with each of the opponents it had discovered. Whether it should
be diplomatic, aggressive, sneaky, or whatever, its decision in this
area should be affected by its overall situation, and should affect its
decisions as to how it should modify that situation. I include in
"modification of its situation" everything that a player of the game is
allowed to do - whether it be building armies, gaining technology,
negotiating with (or bullying, or placating) opponents, building new
bases, upgrading its bases, and so on. It should pick high level goals
and work to achieve them. Conquering an opponent (or defending itself
against one), catching up to the technology of an opponent or working to
exceed their technology, or finding ways to expand when the terrain is
not advantageous, might be examples of some higher level goals. To those
ends, it may need to pick lower level goals, like "capture this city",
"get one of the friendly players to trade tech with me", "steal tech
from an enemy", or "build a coastal base to gain access to the ocean"
might be examples of lower level goals - steps in the process of
achieving its higher level goals.
Then it must turn those goals into actions. Another part of its time
must be spent figuring out tactically sound orders for its armies, which
technologies might be of greatest benefit to it, and good locations for
new bases (which I mentioned before).
All of these functions, plus any others that must be done, could take
place in stages. For orders for its armies, for example, it could at
first do a quick near-term check of what would be most advantageous.
Then it would start expanding on this check, and if more long-term
possibilities were better or worse than it had at first 'believed', it
would alter its 'final orders' appropriately. It would continue looking
for reasons to alter those orders until the human player was done with
his turn, and the 'final orders' it had come up with by that time would
be implemented. This reflects the way in which a human player in such a
game, after looking at the area a certain unit was in, might decide to
give certain orders, and then later, upon seeing a different part of the
situation, might go back and change that unit's orders to reflect his
revised opinion of the situation.
The AI would at first pick orders that were "good enough" if that was
all the time it had. Then it would consider the situation a bit more
thoroughly, and modify its orders to reflect its changing opinion of
what the best orders would be. This process could have limits placed on
it, as well, to create "difficulty levels" based on how much the AI can
think, rather than on what advantages it is given.
The AI control routines could be set up as low-priority threads, so that
they would not take precedence over user interface processing (which
should always have the highest priority), or they could be given a
somewhat higher priority if the player desired a higher difficulty
level. Graphics work could be prioritized as well. These priorities
could then be modified (except for user interface!), so that if one did
not care so much about getting super high frame rates or snazzy effects,
those things could be reduced, and the AI could be given a greater
fraction of the CPU power. Or, if the player had a super-fast system, or
had set a low difficulty level, the AI priority could be reduced
instead, either since it needs less time to do the same calculations, or
in the case of a low difficulty level, because it is limited in how much
processing it is allowed to do anyway.
I talk about threads and priority because they *could* be implemented
that way, but it is not necessary to use threads to achieve this
division of available CPU power that I have described. In fact it may be
better to NOT use threads. The main program thread could just divide its
time between drawing graphics frames, checking user input, and doing AI
processing from within the main loop. It would base how much time it
spends in each of those on ratios of the specified priorities, or on
whatever other method desired.
To be more specific:
Record current time; Draw one frame; Record time again.
Do AI processing for a certain amount of time, based on time taken to
draw that frame.
Check for user input.
Repeat.
> A good example of this is "ADOM", a roguelike game that probably has no AI
> to speak of. I play, walk to the dungeon, kill some stuff and die. Next
> time, I beat that dungeon, walk to the next one and die. Next time, I'll
> die in the first dungeon again. I've never "beat" the game, not even come
> close, but I know that when I died, it was because of something that I
> did wrong, and not because the computer is "that much better than me".
>
> Maybe that should be the definition of a "human-like" game- if you
> die (or lose), but can see where you made your mistake, then it is a
> sufficiently "human" algorithm. Because it exploited *your* weakness, not
> *it's* strength, it could be considered ~human~. And it would also be
> nice if you could learn from an AI- "Hey dummy, next time don't send all
> your Laser Tanks to my side of the board, because I'll just crush your
> undefended base with my Air Strikes"
No, that's just a "well-designed" game. It doesn't necessarily oppose
you in the same way as a human. Being able to tell where you went wrong
so that you can correct your error is just an important element of good
game design. If you die (or lose), and can't figure out why, or it
seemed random, then someone didn't do a good enough job designing the
game.
> I'm O.K. with losing repeatedly to an AI (or a human player) just so long
> as I can get feedback and tips for improvement. Doom was good for this,
> because you'd run through a room and count up how much damage you took.
> Thus you quickly learned a way to minimize your damage taken. That sounds
> suspiciously like a simple "heuristic" for a GA/NN, right?
Sounds good to me! Booming voice: "There be one commandment only:
Minimize damage taken."
Heh... thinking about a booming voice from nowhere saying something like
that reminds me of those commercials with the telepathic tibetans
conversing about things like email and Lotus Notes... 'You mean no more
telepathy?' 'Notes is better.' :-) Sorry about that digression...
Ken Agress
wrote:
>>In addition to a lack of funding, another reason the AI is poor in most
>>games is that humans prefer to beat the machine. If you create a powerful
>>AI, many people will shun the game.
>
>Indeed. Speaking for myself, I've found the AI in most computer games I've
>played to be sufficiently challenging. I've run across a couple games in which
>the AI is a pushover (even for a lazy player like me); and yet I still play
>those games sometimes, and find it relaxing and satisfying.
I think this can easily be handled by adjusting the AI for different
levels of play. My biggest complaint with most games (a la the Civ
series) is that the AI is allowed to cheat in order to make up for
it's deficiencies. I've yet to find a strategy game where the AI
stood much of a chance against a good human player.
>
Ken Agress
Bruno Wolff III
> level" would give the computer-controlled players bonuses (and sometimes
> give you penalties) when set to "hard" difficulties. I have never played
> a game that specifically allowed you to adjust the level its AI would
> play at, rather than giving the AI players an advantage and leaving
> their AI alone.
Heroes of Might and Magic III has dumbed down AI for the normal
and easy settings. There are also resource bonuses as well. Except
in some unusual situations, the AI is supposed to be limited similarly
to human players.
> I have a strong dislike for increasing a game's difficulty artificially,
> by letting it 'cheat' (make use of information a human player would not
> have) or by giving it unfair advantages (Master of Orion 2, on hard and
> impossible difficulties - although I do like the game, and I play it on
> AVERAGE despite that being not too hard to defeat).
Well there aren't a lot of ways to increase the difficulty. Pretty much
the AI, either gets better, gets more information or more resources.
I don't think the first helps much as the best AI is usually pretty
poor in strategic games. I don't like giving the AI extra information,
as that changes the way the game is played. That pretty much leaves
resources (including production bonuses) as a way to balance a game.
Tarquelne
:>by letting it 'cheat' (make use of information a human player would not
:>have) or by giving it unfair advantages (Master of Orion 2, on hard and
:>impossible difficulties - although I do like the game, and I play it on
:>AVERAGE despite that being not too hard to defeat).
I think a game's difficulty _should_ be increased "artificially" if a
"straight" AI becomes too hard/resource intensive at higher levels.
The key thing, though, is to make the cheating invisible to the
player. (I think, BTW, the easiest way to do this in most computer
games is to let unseen computer pieces "teleport" to advantagous
positions or allow resource point expendetures to be reto-actively
spent.)
MOO2 is a good example. I would appreciated a more of a challenge,
but "Impossible" difficulty level should have just been called
"Absurd." - the computer's cheats were just too obvious.
Tarquelne
<os...@apk.net>
I know how God can make a rock so big He can't move it.
************************
"Anybody that wants the presidency so much that he'll spend two years
organizing and campaigning for it is not to be trusted with the office."
--David Broder.
nekhbet
>:>I have a strong dislike for increasing a game's difficulty artificially,
>:>by letting it 'cheat' (make use of information a human player would not
>:>have) or by giving it unfair advantages (Master of Orion 2, on hard and
>:>impossible difficulties - although I do like the game, and I play it on
>:>AVERAGE despite that being not too hard to defeat).
>
>I think a game's difficulty _should_ be increased "artificially" if a
>"straight" AI becomes too hard/resource intensive at higher levels.
>The key thing, though, is to make the cheating invisible to the
>player. (I think, BTW, the easiest way to do this in most computer
>games is to let unseen computer pieces "teleport" to advantagous
>positions or allow resource point expendetures to be reto-actively
>spent.)
>
IMO, this last is one of the most annoying cheats that civ/civ2 uses. I don't mind overly much that
the computer players have massive resource bonuses, but it bugs the holy heck out of me when an AI
player builds a Wonder the same turn as the required tech advance is gained, or hits me with a
nuclear missile the same turn that these are discovered. Because of UI constraints, it is totally
impossible for me to do this sort of thing--it should be impossible for the AI as well, until at
least the next turn.
I suppose this falls under the heading of "Only cheat when the player can't tell".
-----------
Jeff S.
Brady Severns
> I suppose this falls under the heading of "Only cheat when the player can't tell".
>
> -----------
> Jeff S.
You wouldn't suffer a cheat at a table, so why do people pay good money to computer game companies to
buy a game whose AI cheats them the same way or, worse, plays by different rules altogether? Shouldn't
a good AI be programmed to play an expert game under the same constraints that real players are forced
to play under? Perhaps a better AI "cheat" that would at least pay lip service to fairness would be for
the AI in a multi-player computer game to form unshakable AI player alliances. That can certainly
happen in FTF play.
Brady
ccr...@pacific.net
at 04:43 PM, Brady Severns <bradys...@earthlink.net> said:
:Shouldn't
:a good AI be programmed to play an expert game under the same constraints
:that real players are forced to play under?
Of course it should, but as far as I know, the only game company which
agrees with us is Stardock.
Best,
-- Chuck Crayne
-----------------------------------------------------------
ccr...@pacific.net
-----------------------------------------------------------
nekhbet
>You wouldn't suffer a cheat at a table, so why do people pay good money to computer game companies to
>buy a game whose AI cheats them the same way or, worse, plays by different rules altogether? Shouldn't
>a good AI be programmed to play an expert game under the same constraints that real players are forced
>to play under? Perhaps a better AI "cheat" that would at least pay lip service to fairness would be for
>the AI in a multi-player computer game to form unshakable AI player alliances. That can certainly
>happen in FTF play.
Because, in a multi-person game, if the game is to be enjoyable for all players, fairness is
required. But in a computer game, there's only one person to be entertained. The point of a
single-player game is to (enjoyably) challenge the player. Since practical constraints on AI make
it unlikely that an AI player can beat a skilled human player, or at least come close enough to
winning to provide a challenge, there has to be *some* way of making things a bit more difficult for
the human. I'd feel far more "cheated" if I payed for a game that I won every time, without even
having to think hard. So, in order to create a challenge, the AI player must have some other
advantage(s). IMO, resource/production bonuses are the "fairest" way of doing this.... the human
player's units are at a numerical disadvantage, but otherwise all rules are the same for both (or
all) sides. Units "teleporting" when out of sight started, I imagine, as a way to save a few cycles
to deal with higher priority stuff. And as long as nothing "teleports" to someplace that would be
impossible to reach otherwise, this is fine. But when the AI player is blatantly exempt from
restrictions that I must deal with, it annoys me. Btw, your suggestion of "unshakeable AI
alliances" would also irritate me. If making/breaking alliances is intended to be part of the game,
then it should apply to the AIs as well as humans. Otherwise, why bother with multiple AI players?
I might as well just have one extremely powerful opponent.....
-----------
Jeff S.
Ken Agress
wrote:
<snip>
>>
>IMO, this last is one of the most annoying cheats that civ/civ2 uses. I don't mind overly much that
>the computer players have massive resource bonuses, but it bugs the holy heck out of me when an AI
>player builds a Wonder the same turn as the required tech advance is gained, or hits me with a
I've never had this happen on the *same* turn the advance was
discovered, but I have had it happen the turn after. And actually,
it's not all that difficult for the human player to pull off either.
Start building a Wonder you're willing to sacrifice at a high
production site, stage some caravans, and wait for the tech advance to
hit. Change to the Wonder you wanted in the first place, hit it with
enough caravans, and you may very well be able to finish it during the
turn you picked up the technology.
>nuclear missile the same turn that these are discovered. Because of UI constraints, it is totally
>impossible for me to do this sort of thing--it should be impossible for the AI as well, until at
>least the next turn.
Not necessarily impossible, just much more difficult.
Ken Agress
Ken Agress
<bradys...@earthlink.net> wrote:
<snip>
>
>You wouldn't suffer a cheat at a table, so why do people pay good money to computer game companies to
>buy a game whose AI cheats them the same way or, worse, plays by different rules altogether? Shouldn't
Because developing a game playing algorithm that handles the vast
array of possibilities well without cheating or special rules is
*very* difficult to do. Take a look at any bridge game (where
cheating isn't an option). It's a game with numerous good rules of
thumb, developed bidding systems and playing strategies, and a wealth
of easy-to-obtain research material, yet most computer players are (at
best) mediocre bidders and players.
I wouldn't suffer a human player to cheat, because I expect that
they're either playing to learn, playing for fun, or playing for the
challenge. In the first case, I should probably adjust *my* play to
make sure they aren't "screwed by the rules" that they don't know and
enjoy the experience. I should also give them tactical and strategic
help, as they aren't likely to understand the game. In the second,
cheating likely wouldn't add to the "fun" factor (unless the game
specifically allowed cheating, which some do). In the third case, the
challenge is to win within the confines of the rules, not to alter the
rules to bias the game.
However, I doubt you'll find AI players that could address that level
of thinking for some time to come. In every wargame I've ever played
where the computer wasn't allowed to cheat, they were normally either
quite predictable (and therefore beatable) or almost entirely random
(and did little to help their position). AI for games like Civ just
can't accommodate the many options available.
>a good AI be programmed to play an expert game under the same constraints that real players are forced
>to play under? Perhaps a better AI "cheat" that would at least pay lip service to fairness would be for
>the AI in a multi-player computer game to form unshakable AI player alliances. That can certainly
>happen in FTF play.
In many games, they do exactly that. Try getting a reasonable
diplomatic conversation with an AI opponent of approximately equal
size while playing Civ II.
Ken Agress
Ziyat
computer opponents to play by the exact same rules that I do.... I expect
the game to be fun. If the game is fun, and challenging, and keeps me
entertained, then who gives two cents whether or not the computer can "see"
every single one of my units, or can "phase" units through each other to
avoid multiple collision resolvance? I don't buy a game, then complain
because the programmers didn't jump through hoops and do one-armed
handstands in order to ensure the game AI was constrained by the same rules
and penalties as the human player. (Taken to absurd lengths, you would have
to make sure the AI suffered, at some point during a lengthy 6 hour session,
the computer equivalent of mouse hand fatigue, stiff back, blurry vision,
and neck cramps.)
But that's just my humble opinion.
Ziyat
Brady Severns
> I don't know about anyone else, but when I buy a game, I don't expect the
> computer opponents to play by the exact same rules that I do.... I expect
> the game to be fun. <snip> (Taken to absurd lengths, you would have
> to make sure the AI suffered, at some point during a lengthy 6 hour session,
> the computer equivalent of mouse hand fatigue, stiff back, blurry vision,
> and neck cramps.)
Hmm. This is the kind of fun I'd rather avoid! :)
Brady
Gerry Quinn
>(Taken to absurd lengths, you would have
>to make sure the AI suffered, at some point during a lengthy 6 hour session,
>the computer equivalent of mouse hand fatigue, stiff back, blurry vision,
>and neck cramps.)
AND it should have to get up for work in three hours!
- Gerry Quinn