1.5 request

[Velensk]

I've seen lists of changes that are planned for 1.5. One thing that I would appreciate if you spent some effort working on during the 1.5 development is the AI.

As a minor note on the topic, most of the AIs sent with the game don't seem to work anymore, or act strangely when put ingame.

[Iris]

I've seen lists of changes that are planned for 1.5. One thing that I would appreciate if you spent some effort working on during the 1.5 development is the AI.

We Are Working On It (tm), and is part of the Google Summer of Code ideas.

As a minor note on the topic, most of the AIs sent with the game don't seem to work anymore, or act strangely when put ingame.

This is beyond me.

[Dave]

We are working on it. However, http://wesnoth.org/wiki/WhyWritingAWesnothAIIsHard.

[Nebiros]

It seems easy to say that some kind of measurement of how exposed you are to attack, and some way to take it into account, would be better than the current mindless aggression, even if those measures are crude; but it would have to be tried in order to say that it actually worked.

Even risking dying on your own attack is *sometimes* a good move - you basically have to do it to use or fight against ulfs, and even aside from that it can be worthwhile in the right situation. Other incautious moves (leaving wounded units exposed, moving non-skirmishers into surroundable positions, etc.) can sometimes be useful in the right situations. But defining the right situation is hard - it's clear that the current AI errs on the side of recklessness, but exactly where is the line? Even experienced players may differ.

The AI also doesn't coordinate its forces and think about overall tempo; even if a few wounded units go back to heal, the rest go on fighting without them, leading to defeat in detail. Fighting at the wrong time of day is another thing that sounds simple to define, but isn't (and which time is wrong depends on your units and theirs, in a game where either side can control user-defined units).

[mameluke]

I'm not an expert.

But I just had the idea of some heat-maps that could be created.
Like a heat-map of:
* enemy strength
* own strength
* resources
* ...any special/tactical heat-maps you could think of

You could overlay these maps and calculate some "best" values by some algorithm for each field.
The complexity seem not so high. It's linear with the amount of fields in the game (I think).

What do you think of this idea?
Is it worth to think about it more in detail?

[mowerpower]

The complexity seem not so high. It's linear with the amount of fields in the game (I think).

What do you think of this idea?
Is it worth to think about it more in detail?

I'm not quite sure what the suggestion is: how are moves determined? I don't see how this sort of information can determine that is the best unit to move to that place. Given this, do we need to recalculate the heat map before we make the next move, to avoid bad unit formations? If so, then we might lose linear time.

In general algorithmically cheap AIs might be interesting for Wesnoth. I'm guessing there is some way of formulating the idea of a class of "Wesnoth problems" that show that better AIs need more computational resources to do a good job.

[Dave]

I'm not an expert.

But I just had the idea of some heat-maps that could be created.
Like a heat-map of:
* enemy strength
* own strength

The AI already does this.

[mameluke]

hm, ic... so I'm not such a genius as is thought

So the biggest problem is, the lack of programmers who want to understand the AI-code?
Or is the code limited due to design-flaws?

Or you think, this "heat-map" stuff has reached already it's potential?
Wouldn't it be possible to "add" more heat-maps that influence the behaviour in a more intelligent way?

Sorry for all those questions, don't want to bug the people...
My brain was just thinking about those things today, while sitting around in the train.

[Dave]

hm, ic... so I'm not such a genius as is thought

So the biggest problem is, the lack of programmers who want to understand the AI-code?
Or is the code limited due to design-flaws?

Or you think, this "heat-map" stuff has reached already it's potential?
Wouldn't it be possible to "add" more heat-maps that influence the behaviour in a more intelligent way?

Umm...the "biggest problem" is that making an AI for Wesnoth is an incredibly complex task. From an AI perspective, Wesnoth is an incredibly complicated game.

David

[Fosprey]

First, trying to do a decenet ai in a game like this is obviusly too hard for a small team.
Big teams with big budgets like blizzard can't make even a decent AI, ALL AI i know in strategy games of incomplete information, are VERY VERY WEAK.
TEAMS AND TEAMS, have tried to make a decent NL Poker bot, and the best ones can be beaten by anyone with the most minimal concept of the game.
Basically Making a good AI is something that requires a lot of work, more than all the wesnoth project together, and is definitly not worth it.
The actuall AI is fine for campaigns and Story-based scenarios, and if you really want a challenge, there is plenty of humans that will put quite a challenge.
I don't working in the AI is worth it, the effort-benefits relationships are horrible.
Still I know that there is always people that love to work with AI, Just for the pleasure of trying, But it's clear that not major advancments will come from the wesnoth team, I think it's more a NASA project than a non-budget small team like wesnoth. And as i said, i don't see the need.

[Wesnothphile]

First, trying to do a decenet ai in a game like this is obviusly too hard for a small team.
Big teams with big budgets like blizzard can't make even a decent AI, ALL AI i know in strategy games of incomplete information, are VERY VERY WEAK.
TEAMS AND TEAMS, have tried to make a decent NL Poker bot, and the best ones can be beaten by anyone with the most minimal concept of the game.
Basically Making a good AI is something that requires a lot of work, more than all the wesnoth project together, and is definitly not worth it.

Pessimism and cynicism don't jive very well with opensource projects. If people believed in "no, we can't" rather than "yes, we can", you'd never have these projects in the first place, since any volunteer run project requires a massive dose of faith. And there are chess programs that can beat humans (even chess grandmasters on occassion, and certainly beat most regular players). There are even strategy games that have good AIs that don't rely on cheating (I've heard good things about GalCivs from Stardock).

So if Big Teams with Big Budgets(microsoft?) can't do something right, opensource volunteers (linus torvald?) shouldn't try?

The actuall AI is fine for campaigns and Story-based scenarios, and if you really want a challenge, there is plenty of humans that will put quite a challenge.

Currently, campaign scenarios sometimes rely on ridiculously short time limits and massive preset resource disparities and recruitment asymmetries between humans and AIs to create artificial difficulties. A better AI would allow campaign creators to rely less on such crutches.

I assure you that if a better AI came along, most players would be overjoyed. Further, even in multiplayer situations, there are cooperative multiplayer scenarios that make use of AI competitors. So any improvements in AI would not be wasted.

And there are users without internet access (say a guy on a laptop on a plane), or who can't dedicate a long block of time to multiplayer gaming (like babysitting screaming children). For these people, playing against computers offline or playing a game divided into 5 or 10 minute chunks saved here and there would be an option. It's easier for a computer AI to accommodate these chunky sessions rather than have a human player be available for these discontinuous random chunks of play time.

I don't working in the AI is worth it, the effort-benefits relationships are horrible.

I believe this misconception about "effort-benefits relationships" has been addressed by the developers before on this forum. This is an opensource project that depends on the involvement of volunteers. It is not a corporate top-down control structure. This means that volunteers contribute according to their skillsets and inclinations. You cannot easily repurpose human resources in an opensource environment. Someone who stops working on AI(or music, or graphics, or whatever else) would not necessarily refocus their efforts on something else you wish to direct them to (either because of different skillsets or different interests -- remember many volunteers are coding here for leisure, not to complete a chore they have no interest in).

Still I know that there is always people that love to work with AI, Just for the pleasure of trying, But it's clear that not major advancments will come from the wesnoth team, I think it's more a NASA project than a non-budget small team like wesnoth. And as i said, i don't see the need.

If there are players wishing for a better AI, then there's certainly a need, though the triaging of priorities and needs remains a developer's perogative. Better AIs would help in storyline scenarios, human vs. AI matches, and even cooperative multiplayer.

Besides, I think you're selling the Wesnoth Team short. They're amazingly talented individuals, and you'd be surprised at what they can do if they put their minds to it

[AI]

http://www.wesnoth.org/wiki/WhyWritingAWesnothAIIsHard

Writing an AI for wesnoth is far harder than writing one for chess, chess AIs have been worked on for ages and until recently required supercomputers to beat grandmasters.

[Fosprey]

I mean, i SAID INCOMPLETE INFORMATION GAMES.
making AI for complete information games like chess is a LOT easier.

I was suggestion people to don't lose time un an utopic endeavor, if they will spend their time in something not wesnoth related instead of AI, GOOD for them, and i suggest them to do it.

And as i say, the AI is ok for single player games, difficulty is not what people look in Campaigns. WEsnoth campaigns as far as i played , even on hard are incredibily easy, almost impossible to lose unless you really have horrible luck. (i played HFTT and TALE OF TWO BROTHERS),Single player games that are VERY EASY like Final fantasy games and Zelda games, are EXTREMELY POPULAR, people enjoyed them, so it clear that difficulty is not the issue with people having fun with campaigns or single-player games, so again, what's the deal?

The only real porpouse with improved AI is to allow people to train as multiplayer players before jumping online, or if they can't play online. And i think it's not a big issue.

PD: About the Pesimistic thing. Wesnoth has worked as a project because developers work under the kiss principle, and one part of it is to not lose time in things that are utopic. Well, good AI is harder than all the wesnoth project togheter, and a lot of ideas that have been rejected just because they are too hard.

[turin]

difficulty is not what people look in Campaigns. WEsnoth campaigns as far as i played , even on hard are incredibily easy, almost impossible to lose unless you really have horrible luck.

Uh, what?

(i played HFTT and TALE OF TWO BROTHERS),

Ah, that would explain it.

TB and HttT are two of the easiest campaigns out there. Try playing, say, Sceptre of Fire on HARD and then tell me that wesnoth campaigns "even on hard are incredibly easy".

[Dave]

First, trying to do a decenet ai in a game like this is obviusly too hard for a small team.

I don't think it is so much related to any of the characteristics of our team as it is to making a good AI for a game as complex as Wesnoth just being incredibly difficult. I think our AI is very competitive with AI's from any commercial game, and we are continuing to try to improve it.

David