Development Update - 9th November 2012

[Chris]

GJ may have had his tongue in his cheek when he made the comments, but it is possible to write an AI that can always perform the most mathematically perfect way to achieve an objective like killing the most vulnerable Xenonaut unit in fire range, and it can do that better than any human could.

I'm sure it could be exploited, but that's why it'll be interesting to see how well people do against the AI on the hardest settings and what exploits they end up using.

[xcorps]

GJ may have had his tongue in his cheek when he made the comments, but it is possible to write an AI that can always perform the most mathematically perfect way to achieve an objective like killing the most vulnerable Xenonaut unit in fire range, and it can do that better than any human could.

I'm sure it could be exploited, but that's why it'll be interesting to see how well people do against the AI on the hardest settings and what exploits they end up using.

That goes without saying. The problem isn't how to calculate that plan, it's how to implement it. The firefight is just a culmination of manuevering. It's the manuevering that makes an AI good. How do you deploy the aliens to achieve that? If the player is performing decent recon, it will be rather easy to defeat a fire sac by flanking or enveloping. How would the AI react to that? Will the AI be able to respond to recon manuevers by using Z axis (second floors and such). Will the AI be able to redeploy his frontage to deny a flank? Will the AI be able to respond to a base of fire/flanking manuever by withdrawing to secondary or tertiary positions or will it just shoot it out with a base of fire while the manuever element flanks it? Will it over compensate for a flank and allow the base of fire to achieve suppression?

[zzz1010]

the most mathematically perfect way to achieve an objective like killing the most vulnerable Xenonaut unit in fire range, and it can do that better than any human could.

Not drastically better, implying it not cheating, it not have full information about units and can have only average decisions about who is more vulnerable,(we attack caesan instead andron - he attack unit in basic armor instead in wolf (and this is far from always better solution, it can be wasting fire possibilities for trying to kill some cannon fodder rookies instead some more significant and dangerous guys))

If talk about chance to hit - we see same mathematical (and very not precise too) numbers for decisions(If he not have more infomation then we here it can be also weakness, because, at least in current builds, I can shoot with my sniper, even though math say there zero chance to hit, just because I think he can, and hit ).

He can more accurate predict(but still averages) where unit (only which he sees) can move and what tiles he will see durning turn for more precise hit and runs. I don't want to say he can't play very well, but just calculations like this don't give drasctical advantage and not make him superhuman.

Edited November 10, 2012 by zzz1010

[radister]

I'm sure it could be exploited, but that's why it'll be interesting to see how well people do against the AI on the hardest settings and what exploits they end up using.

Pick a buidling in the middle of the map as your base, destory all the building in the area around and then wait...

Wait to see what kinda of plan this AI can think of now

However thinking about it if i was the AI i would just:

make one of my guys go berserk and kill half of us and in that time were dealing with him rush us from outside.

If it can do that then we won't stand much of a chance.

(and that would make me happy for some reason)

[HWP]

GJ may have had his tongue in his cheek when he made the comments, but it is possible to write an AI that can always perform the most mathematically perfect way to achieve an objective like killing the most vulnerable Xenonaut unit in fire range, and it can do that better than any human could.

Well, to take an example, AFAIK best backgammon AI implementations are still losing to skilled backgammon players. It's a pretty similar game to TBT, especially similar to 2012's XCOM, you roll the dice and move your units.

Unlike poker, it's not a game of incomplete information. There's no human element. It's also far, far simpler than proper turn-based tactics. You would think computers should have a ball, but that's not the case, it's still not beaten.

So maybe AI will get its kill more often - but will it avoid exposure to player fire? Will it balance risks and rewards just right?

Some testing with balanced sides (identical units even) could show more, but I'm pretty sure an experienced human player will still be better, even if you give the AI all you can short of cheating. This is not to put down AI programming, rather to say that the human mind is still quite competitive.

[GoodGuyEddy]

There used to be a popular notion that XCOM EU is unbeatable on Impossible.

By bloody who? By people who never played a strategy before? As soon as you capture your first sectoid and see "plasma pistol recovered" you are already tiped off that you can use alien tech ,and as soon as you research the plasma pistol tech it's a game ending revalation. I had a squad decked out in light plasma rifles before i got near any high end laser tech, and there was no one that could touch me untill the berserker came about, and even by that time i had leveled up my squad considerably. I beat the game on "impossible" once i got a hold of the games mechanics. Hell considering the watered down nature of the ground combat and the "tunnel" nature of some of the maps you can easily predict every move the enemy can possibly make. The only casualties i suffered were due to bugs and bullshit. (yes i am looking at you assholes that forget to take the lid off of their rocket launchers! )

So i say GJ should go all for it! If all else fails grenade rain reighns.

Edited November 11, 2012 by GoodGuyEddy

[chacineiro]

I say, bring it on, I´ve been waiting for like 15 years to see any decent AI that can put up a great challenge without cheating.

[Max_Caine]

Those reverse psychology courses were worth every penny, Chris.

[Rutger]

Challenging AI sounds great, don't hold yourself. You can always make it optional.

[Gauddlike]

Challenging AI is a much better start point when balancing your difficulty levels than trying to work up from a cakewalk.

[Wisq]

There used to be a popular notion that XCOM EU is unbeatable on Impossible.

The devs didn't think so. They thought that very few people would beat it. But in order to have an achievement for beating it on impossible, they had to beat it on impossible, otherwise they weren't allowed according to the rules.

They did, but they considered it very hard, and as I recall, only one tester managed to do it. I don't think they ever did it on ironman — but then, there's no achievement for Impossible Ironman.

[HWP]

There were other post, but really, early Oct it was said a number of times.

http://www.goldhawkinteractive.com/forums/showthread.php/622-Firaxis-Games-XCOM-Enemy-Unknown-announced!?p=41446&viewfull=1#post41446

I don't think they ever did it on ironman — but then, there's no achievement for Impossible Ironman.

Well, now this is becoming routine fast.

AFAIK Second Wave all options on Impossible Ironman has yet to be beaten, but that's a matter of quite little time.

Diminishing Returns is the most important option, it makes sat costs increase 1.45 each, so for 16th the price should be about 36,000 (haven't tested it really). This can be circumvented by multi-building, e.g. if you build sats 10-14 in one go (and ignore 15-16) it's only 3,800 each.

[Gr3y]

I've yet to see a strategy game AI be "too" good.

So if you can create that, I say bring on the challenge.

[bernlin2000]

My first post here, and I thought this update was an appropriate place to start. This project is floating all around the interwebz, no doubt partially as a result of the fervor surrounding Firaxis' recent XCOM release (which I have bought and played through quite thoroughly the last few days). At first, when I read the "project status" I had the impression things were going quite slow (perhaps the project had even stopped), but after reading the most recent news updates it seemed pretty clear to me that you guys are working pretty hard on this game. So much so that I was willing to throw 20 dollars into the ether in the hopes we're going to get the kind of game Firaxis just released, but much deeper are far more strategic/tactic-"y". From what I've seen in the forums so far (not much), I think this will be a given. I wish you guys good luck, and I can't wait to start toying around with what you guys have done. As someone who never played the original (although I have recently bought it off Steam, and probably will some time here...), I probably am coming from a different perspective than others, likely one that might emphasize graphics more. Frankly, as long as the game lands somewhere between X-COM and XCOM (hopefully at least at the 50% mark) I'd be a very satisfied pre-orderer.

Continued good fortune on wrapping up Alpha and moving on to greener (and less buggy) pastures.

[bernlin2000]

The only thing I'd say about your update is: there's no such thing as a too-smart AI. You guys know what they did for XCOM right? Normal difficulty features, in fact, a partially "shackled" AI, while Easy is even more-so (it seems, mainly, that some of the enemies don't use some of their more powerful abilities either ever or not as often, like rocketing instantly hovering from one part of the map, to behind your soldiers). You have to go up to "Classic" mode just to get the full AI experience, and by then they are penalizing soldiers (less hp) and strategy elements (and I think having both at the same time makes it a bit too hard...there should have been 5 difficulty settings in that game, an extra between Normal and Classic).

I'm really looking forward to doing scenarios that will be an important part of development testing: this might be the first game I've ever played where I had that kind of possibly game-changing input.

[bernlin2000]

Not drastically better, implying it not cheating, it not have full information about units and can have only average decisions about who is more vulnerable,(we attack caesan instead andron - he attack unit in basic armor instead in wolf (and this is far from always better solution, it can be wasting fire possibilities for trying to kill some cannon fodder rookies instead some more significant and dangerous guys))

If talk about chance to hit - we see same mathematical (and very not precise too) numbers for decisions(If he not have more infomation then we here it can be also weakness, because, at least in current builds, I can shoot with my sniper, even though math say there zero chance to hit, just because I think he can, and hit ).

He can more accurate predict(but still averages) where unit (only which he sees) can move and what tiles he will see durning turn for more precise hit and runs. I don't want to say he can't play very well, but just calculations like this don't give drasctical advantage and not make him superhuman.

I think you run the risk there of making the AI too smart for its own good: i.e. "book smart" but not "street smart". For me, the calculation of who to attack is pretty simple: if they have the lowest health (and there isn't another unit who might be better suited for taking the final kill shot, not sure how that sort of thought process could be implemented into an AI) or I have the highest chance of making the shot without missing, then that's who I'm going to shoot. I don't think a ton of factors should go into things like that, and of course visibility should factor into all this stuff: I got the vibe that you might be suggesting they have information on all my soldiers. They should only have info on those soldier they or their alien pals can see normally (i.e. the same visibility my soldiers have).

[zzz1010]

2bernlin2000

I didn't suggest anything at all. Chris just said what AI is able to do precise calculations much better then any human(and it's true) and this can be it's advantage to which I replied that there is no too much room where such calculation can give great advantage(at least I didn't see).

For example:

I have the highest chance of making the shot without missing, then that's who I'm going to shoot.

For Ai is used same rules of hit and same algorithm of calculation to hit probability percentage, if we see above our cursor that our chance to hit is 75% - for AI it will be same result, there nothing to calculate to become much better then human. AI can more precise find spot where it will have best chance(from possible for this positions) to hit this target, but in current system there no too much spots where chance is drasticaly change and not obvious(for human player) or it will be about +5% and RNG most time obliterate such precision. It still will do it little better, and maybe somehow it's little help, but it's not make it supehuman or unbeatable

Edited November 16, 2012 by zzz1010

[TrashMan]

I'd add a bit of randomization in the alien AI..

Just to make it a little more chaotic and unpredictive.

[mercy]

I just realized, there is a functional Map Editor in the game. If it will be enabled at release will add a very strong point to the fun-factor and hopefully generate a lot of sales. Which UFO game has a publicly available, easy to use and functional level editor?

The fact that people can whip up a preliminary design in an hour and refine it later speaks for the dedication of the Xenonauts Team. Another golden star added to the Game Rating!!