CA's Wikiman on the AI

[Watchman]

"Put a thief to catch a thief."

[Puzz3D]

"It’s very easy to make the computer beat a human 100% of the time in a game of this complexity. It would not be very sophisticated but it would be a walkover.

Making a game think like a human is currently impossible given the constraints of the household PC and the development time we have. If you all had clusters of uber-PCs and we had another 6 years to develop this game you would be impressed with our work." - Wickman


He must be talking about putting cheats into the game for the AI. It sounds like they aren't even trying to make an AI that can handle the sophistication of the game's features. This is consistent with the AI that you see in RTW/BI. The AI can't handle the possibilites on the strategic map, and it's not handling the possibilites in the tactical battles either. The sophistication of the game has outstripped the AI.

[IceTorque]

"It’s very easy to make the computer beat a human 100% of the time in a game of this complexity. It would not be very sophisticated but it would be a walkover.

Making a game think like a human is currently impossible given the constraints of the household PC and the development time we have. If you all had clusters of uber-PCs and we had another 6 years to develop this game you would be impressed with our work." - Wickman


He must be talking about putting cheats into the game for the AI. It sounds like they aren't even trying to make an AI that can handle the sophistication of the game's features. This is consistent with the AI that you see in RTW/BI. The AI can't handle the possibilites on the strategic map, and it's not handling the possibilites in the tactical battles either. The sophistication of the game has outstripped the AI.

The AI can already hold it's own on the strat and battle maps, when it is not hindered by a lack of funds, inferior troops and/or huge cities taking up most of the battle map. It certainly does'nt need any cheats, just a level playing field is all.

[Martok]

The AI can already hold it's own on the strat and battle maps, when it is not hindered by a lack of funds, inferior troops and/or huge cities taking up most of the battle map. It certainly does'nt need any cheats, just a level playing field is all.

So in other words, the AI can win just fine on only an open playing field, yes? That sounds like what you're saying, anyway. Sorry, IceTorque, but that's too many qualifiers.

It is because the AI is poor that it suffers from a shortage of funds and inferior troops. It is *because* the AI is poor that it can't win city battles (defending or attacking). You have the cause and the symptoms switched around here.

[IceTorque]

So in other words, the AI can win just fine on only an open playing field, yes? That sounds like what you're saying, anyway. Sorry, IceTorque, but that's too many qualifiers.

It is because the AI is poor that it suffers from a shortage of funds and inferior troops. It is *because* the AI is poor that it can't win city battles (defending or attacking). You have the cause and the symptoms switched around here.

To me the term 'level playing field' does not mean "open playing field". Perhaps the term 'a fair fight' is more clear. If the problem was with the AI and not what is imo minor flaws in game design and maybe the game code. Then why are many of these flaws so easily minimized or eliminated by simple tweaks to the games txt files ?

[Pontifex Rex]

I think the "don't expect miracles" part is the most important thing for players to remember. No PC on the market today is capable of handling the programing that would be required for a TW version of "Big Blue". I think this AI will be improved but once we learn how the AI "thinks" we will *learn* how to defeat it again. The AI, on the other hand, has very little, if any, ability to learn from observing our game play.

That is our advantage over the machine, we can learn and adapt it cannot. One day the computers may be able to think and learn, its just not today.

Cheers.

[Papewaio]

So in other words, the AI can win just fine on only an open playing field, yes? That sounds like what you're saying, anyway. Sorry, IceTorque, but that's too many qualifiers.

It is because the AI is poor that it suffers from a shortage of funds and inferior troops. It is *because* the AI is poor that it can't win city battles (defending or attacking). You have the cause and the symptoms switched around here.

That shows that the strategic AI is not up to scratch and doesn't know how to look after its economy.

I tell all new players that they must conquor their economies first, their enemies second.

The AI needs to do the same by good build choices and correct troop choices to match.

No, you are misinterpreting the meaning. If you read the post in context, I think it's clear that he is just responding to someone else's comment that it's impossible to make a challenging computer opponent.

What he's saying essentially is "we could easily give you a challenging computer opponent (say, by by giving outrageous stats to AI units), but it wouldn't be much fun to play. So we are working on the AI to give you a sophisticated opponent that is BOTH challenging and fun to play. Just don't expect miracles."

It would not be the AI that wins if the stats were reworked to make it very challenging. That is just loading the probabilities.

I have to agree with Puzz3D-sama... the game has too many variables for the AI to get a handle on and hence make good decisions. Shogun had fog, rain, snow... and it was always dicey when playing against monks in the fog... you never quite knew where they were going to appear. The AI could handle itself fairly well. I don't think the other AI's have kept pace with the graphics. So eye candy has outstipped intelligence by quite a bit... we have traded candy floss for challenge.

====

BTW I have never played MP. And when it comes to getting a decent AI, I would choose the experienced SP player over the experienced MP player... but if it was my decision I would get both in the deal and have the MP player playtest the AI.

[screwtype]

I have to agree with Puzz3D-sama... the game has too many variables for the AI to get a handle on and hence make good decisions. Shogun had fog, rain, snow... and it was always dicey when playing against monks in the fog... you never quite knew where they were going to appear. The AI could handle itself fairly well. I don't think the other AI's have kept pace with the graphics. So eye candy has outstipped intelligence by quite a bit... we have traded candy floss for challenge.

Undoubtedly true. However, I am optimistic that having spent four years developing the basic engine for RTW, CA are now in a position to spend more time tweaking the AI up to an acceptable level.

As I said before I'm not expecting miracles, I'd be happy to have a battle AI that was as smart or hopefully a bit smarter than what we got in STW. The strategic AI also obviously needs a lot of work, because the new more detailed 3D map gives the AI a lot more choices.

But heck, surely it couldn't be *that* hard to develop a decent strategic AI. I reckon I could design a pretty decent one myself if someone paid me an acceptable wage and gave me twelve months to finish it.

[Papewaio]

Past behaviour is such a good indicator of future behaviour that it is the preferred method of job interviews...

So what colour candy floss do you want?

[Dead Moroz]

He must be talking about putting cheats into the game for the AI. It sounds like they aren't even trying to make an AI that can handle the sophistication of the game's features. This is consistent with the AI that you see in RTW/BI. The AI can't handle the possibilites on the strategic map, and it's not handling the possibilites in the tactical battles either. The sophistication of the game has outstripped the AI.

Man, you spoiled my impression of this news. Anathema on you!

[screwtype]

He must be talking about putting cheats into the game for the AI. It sounds like they aren't even trying to make an AI that can handle the sophistication of the game's features. This is consistent with the AI that you see in RTW/BI. The AI can't handle the possibilites on the strategic map, and it's not handling the possibilites in the tactical battles either. The sophistication of the game has outstripped the AI.

No, you are misinterpreting the meaning. If you read the post in context, I think it's clear that he is just responding to someone else's comment that it's impossible to make a challenging computer opponent.

What he's saying essentially is "we could easily give you a challenging computer opponent (say, by by giving outrageous stats to AI units), but it wouldn't be much fun to play. So we are working on the AI to give you a sophisticated opponent that is BOTH challenging and fun to play. Just don't expect miracles."

[Mount Suribachi]

"It’s very easy to make the computer beat a human 100% of the time in a game of this complexity. It would not be very sophisticated but it would be a walkover.

Making a game think like a human is currently impossible given the constraints of the household PC and the development time we have. If you all had clusters of uber-PCs and we had another 6 years to develop this game you would be impressed with our work." - Wickman


He must be talking about putting cheats into the game for the AI. It sounds like they aren't even trying to make an AI that can handle the sophistication of the game's features. This is consistent with the AI that you see in RTW/BI. The AI can't handle the possibilites on the strategic map, and it's not handling the possibilites in the tactical battles either. The sophistication of the game has outstripped the AI.

Sorry Puzz, but you are letting your downer on RTWs AI blinker yourself here.

Screwtype has pretty much explained what is actually meant.

Personally, out of all the MTW2 info that has come out, this is the best news. Its an issue within the community and CA is doing something about it.

[Zatoichi]

I'll go along with Mount Suribachi - of all the 'official' announcements made so far, the extra effort being put into AI is the most welcome. Of course, the proof of the pudding is in the eating, but announcements like these will always cheer me up.

Mmmmm... pudding...

[Ghost of Rom]

If CA merely addresses the well known glaring aspects of the AI, I will be happy. Take a look at what some of our modders accomplish, the programmers should be able to do so much more.

[Puzz3D]

Sorry Puzz, but you are letting your downer on RTWs AI blinker yourself here.

Screwtype has pretty much explained what is actually meant.

I read the whole thread. Wickman doesn't give any specifics. It's a feelgood PR post designed to reassure potential buyers and counteract the impression left by RTW. CA is always improving the AI according to them. If that's so, why is the battle AI inferior in several specific ways to STW's AI? If Wickman spent lots of time talking to the programmers about AI, then where are the specifics that he surely learned about?


IceTorque,

There is no way the RTW battle AI can beat me at even odds on a level playing field. The AI in MTW2 is going to require bonuses to offer a strong challenge. It's the way all games currently do it. That isn't even open to question. They have to do it that way.

I use a different approach when I play SP because I think the combat bonuses mess up the AI, for instance, by making it more likely that archers will charge into melee combat. I play at normal difficulty and just don't play my best so as to give the AI a better chance. I actually find the RTW v1.5 SP campaign to be relaxing to play.

[ajaxfetish]

Even without specifics, this is some of the most encouraging news I've heard. The main thing I want to know is just that they care about the AI as much as we do and that they're doing their best to make it better. It doesn't make for the flashiest news, unfortunately, and has been almost ignored in the previews/interviews/etc. Here where there's nothing more important, it's at least nice to have it addressed.

Thank you, Wikiman, and do what you can to give us a true challenge.

Ajax

[Servius]

IMO, if a modder can improve the AI, the company screwed up. No one should know an AI better than its creator, so the AI should be at its best in the vanilla version. For example, if an AI constantly bankrupts itself, and a modder comes in and teaches the AI how to better manage its economy, the company obviously didn't do a very good job, because an unpaid layman designed a better mousetrap than a paid pro.

[Watchman]

On the other hand, modders don't have release-date deadlines breathing down their necks...

[HalfThere]

IMO, if a modder can improve the AI, the company screwed up. No one should know an AI better than its creator, so the AI should be at its best in the vanilla version. For example, if an AI constantly bankrupts itself, and a modder comes in and teaches the AI how to better manage its economy, the company obviously didn't do a very good job, because an unpaid layman designed a better mousetrap than a paid pro.

Modders have more time and testing at their disposal. Plus, they really haven't made any better AIs. Darthmod is, I suppose, better in that the AI doesn't start in one big line, but it's a marginal improvement at best.

If the company gives the tools for complete AI modification, then there SHOULD be some modders out there who eventually make something that excels in at least a few aspects.

[BDC]

The obvious answer would be to have either a multiplayer map so you play humans or revert to a more simple map ala MTW, which the AI coped fine with.