A&A AI

[Subotai]

A&A AI.

A philosophical or techichal question, or both?

Is it even possible to create a competent AI in A&A?
I think so.
Some of the differences between chess and A&A is that for computers chess is purely calculations.
An A&A AI algorithm cannot be based on brute force only.
Another difference is that it takes only short time to learn a human how to play chess. The rules etc. is
quite simple. To be good at anything is never easy, but to reach a "decent" level in A&A is much more difficult than in chess.
Whereas in chess the AI must "know" to value the king and the queen higher than the pawns and bishop etc.,
an algorithm in A&A cannot be based on the same judgement values. It might be a good move to sacrifice
both fighters, bombers and naval units, in some cases.
In chess there are many 0-1 moves, it's obvious what to do or don't, and especially for humans who can't think much more than 2-3 moves ahead, to some extent.
There are big debates among the best human A&A players on what is the best tactics, for which power, and
in which situations. When I ask experienced players in the triplea lobby what is the best opening moves, they say it depends on what the opponent does....

There's an interesting thread in the triplea dev forum on AI in triplea and AI in general.
I think it boils down to time and money. How many programmers must spend how many years to make a good, decent, or hard AI in an A&A computer game? Or how many highly trained monkeys

At least until no one has ever tried hard enough, then I still think it's fully possible. But I know
for sure it won't happen in the triplea version, it's too much work to implement "good" AI into
a freeware application of A&A, because the A&A game itself is very complicated.
A&A is all about numbers, mathematics and economy. And computers are generally better than humans in this
aspects.
In 1996 (?) it was the first time ever that a chess computer beat the world champion.
When will we see an A&A AI that beats the best human A&A players?

[Nuclear]

The reason why AI is not possible, because it does not think.  AI runs off functions.  It has no understanding of how a game works, it is just instructed what to do next based off a function.

Further A&A deals with probability/odds.  No computer is ever truely random, not even a human can be truely random.  Do not believe me, then take a look at how a computer expert would win Keno a bunch of times because he could figure out the next numbers that the computer was going to pull off.  As long as someone is able to figure that out, it is easy to use that information to your advantage.

I remember the first computer beating the human in chess.  To my knowledge the guy came back and killed the computer.  All he had to do was make moves that the computer could not handle.  The computer functioned in terms by giving values to units.  Thus the guy used that info to his advantage.

You are asking for a computer program that some how deals with a pure dice game.  There is no way of doing such.  Hate to bust the bubble but A&A is all dice.  Just roll and there you go.

[Subotai]

... but to reach a "decent" level in A&A is much more difficult than in chess.

I strongly disagree.

Why?

[ncscswitch]

Nor you can be any good at it.

I get it!

You are TRYING to get banned from another gaming site so that you never have to prove your vitriolic boasting!

[tekkyy]

The chess AI working really we all come to know of is simply brute force.
This is might not even be possible with Axis and Allies.

But we can still work via hardcoded strategies.

[Cmdr Jennifer]

Honestly, I don't think a great AI can be made with today's knowledge and technology.  In the future, I think one will be made.  However, we could improve TripleA's AI by forcing it to limit transports to 8, max and by telling it to make attacks at 70% odds instead of overwhelming odds like it prefers.

I said improve, because I don't think it'll ever be good.  Anyway, in my mind, the TripleA AI is good to teach rookies the basic rules of the game, after that, you need humans to teach you the intricacies.

[Subotai]

I guess I should initially post the AI issue in the revised forum, could get more inputs on this philosophical question.

I don't think we will see "decent AI" in triplea, even if we say 3 years from now. I don't think it can be done in triplea.
As for java and coding, I don't know much. TripleA is a simple free game which lets you play A&A against others.

Technically it's different question. I used to think that it is impossible, but who knows what's possibly in the future?

Jennifer, what if the team behind Deep Blue spend $100.000.000 during a 5-10 year period.
The pure goal of this project is to make A&A AI as good as possible.
I really think that it is possible to create AI than can give a match for many players, and especially people who are
new to A&A revised.

As for AI can't "think", a plane can't fly?

[tekkyy]

If pay a research team they might find us some maths and discover several fail-safe strategies.
Like those discovered by the public after decades of gameplay of classic.
Then no worries about compuatational power.

The AI is currently not that good even for teaching new players.
Can someone think of a naval AI?
Buy one destroyer for every enemy ship?

[Subotai]

If pay a research team they might find us some maths and discover several fail-safe strategies.
Like those discovered by the public after decades of gameplay of classic.
Then no worries about compuatational power.

The AI is currently not that good even for teaching new players.
Can someone think of a naval AI?
Buy one destroyer for every enemy ship?

I guess you have read the AI thread in the triplea dev forum.
There are several detailed suggestions and observations on the peculiar AI.
For a start, to set a goal on decent/good/strong AI, AI has to look/know the whole map.
According to one of the coders, that is not possible within the current engine....   
You cannot have decent strats without looking at the whole map!
You can do sealion G2 
The current AI only conducts SBR combat moves longer than one TT.
There are maybe 10-20 different details that can be improved if someone does a lot of work.
Worst case, you have to make a much bigger and better triplea 
I don't expect this to happen because of obvious reasons. I do hope that triplea will be 100% 4th.ed. and 100% LHTR 
in the near future.
You don't have to pay researchers, that's wasted money. You have to pay programmers and techies to do a lot
of coding and stuff. 

[tekkyy]

Yeah I have read that thread. I use to read TripleA forum. I actually made one small contribution to tripleA. In the last or the second last release there is my alias. So cool.

No it wasn't AI. It was just allowing one sea zone to be connected to more than one canal. This is essential for mappers who wanted black sea canal.

But until we improve the engine...support for AI, house rules or even just "standard" games like LHTR, AAP, AAE, etc will not be straight forward.

I guess we can't even do my "Buy one destroyer for every enemy ship" if the engine doesn't even let you see the world.