Page 1 of 2

Disappointing AI(?)

Posted: Mon May 18, 2009 4:55 am
by Tost
Sean O'Connor's Slay has a really great AI. There are 4 difficulty levels. It's hard to beat it at the third and takes a lot of tries and sometimes even luck (a certain AI attacking not you but the neighbor just at the right time) on the hardest setting. Since the rules of the game are very simple, it seems to be relatively easy to program a good AI.
Yet, in Conquest!, the AI in my experience is extremely dumb. Even on the highest difficulty level and without actually thinking much, i.e. placing the units on a whim, leaving undefended huge areas of my territories, I beat the AI easily.

So, what do you think? Am I just having luck or is the AI really that disappointing?

I know, there is always multiplayer, but this is one of the very few strategy games where it's feasible to have a (non-cheating) AI and have a challenging game due to the easy rules. So unless I am mistaken or the AI gets a huge overhaul, I will probably go back to Slay. Which is a shame since I like the graphics, advanced options and general feel of Conquest!

Posted: Mon May 18, 2009 7:33 am
by IainMcNeil
I think it may be related to the map generation more than the AI. I thinkw e need to have a look at how hexes are allocated to different players in the initial setup. I think too often the player gets large blocks of land and too often there are many hexes of unowned land.

Posted: Mon May 18, 2009 10:59 am
by Jekky
If you want a harder game, one way you can get it is to change the turn order - during the game setup, click on the coloured circle which indicates your player colour, and use the + and - buttons to change your position in the turn order - the further down the order you put yourself, the more AI players will move before you get to respond.

Obviously this isn't a perfect solution - but we'll keep looking into the AI to see what tweaks and improvements we can make to it (slay was tweaked over a period of years to get the AI as good as it is).

Posted: Mon May 18, 2009 12:57 pm
by honvedseg
I found that the game got a lot tougher when you don't get that first turn advantage from the start. The AI gets a lot of the easy hexes that can make or break the game from the start.

The AI is occasionally pretty stupid, but at other times pulls a few nasty ones on you. In my opinion, the overall challenge on highest difficulty is not "overwhelming", but sufficient to make it interesting; you can come back from a bad start, but it takes effort. Once you've got a lead, though, it's a foregone conclusion.

Making the AI opponents begin to slightly "favor" attacks on the largest faction, increasing as the size margin grows, might have been one way of keeping the difficulty factor high throughout the game.

Posted: Mon May 18, 2009 5:27 pm
by corey420
the ai builds way too many castles. i think that is a significant problem for the ai that always causes problems with lack of 3lvl units.

Posted: Mon May 18, 2009 8:37 pm
by marlborough
so far i have played small maps with level 3 AI and i like it! maybe you should mix (maps & level AI & if you want to play first ) to have the difficulty you want.

Posted: Tue May 19, 2009 1:58 pm
by Tost
I don't think the order in which the players start should play much of a role. Take Slay, again, and play any map on the second difficulty level. You start first and yet usually after the first turn each AI has more territory than you.

I tried changing the order and going last doesn't make Conquest any harder, apart from the first few turns. Neither does changing map sizes. And I'm no strategic mastermind.

I'm not sure what the Conquest AI does wrong, though. Maybe it's not aggressive enough in expanding its territory? I notice that sometimes one AI manages to actually conquer its neighbours but by that time I usually own 70% of the map.

Played some more Slay for comparison and it's like a whole different game. I hope you get the AI up to par because I'd rather play Conquest than Slay. :)


Quick unrelated question: I can chose a widescreen resolution of 1920x1200 but I still get black bars left and right. Is that intended?

Posted: Tue May 19, 2009 2:36 pm
by Jekky
Yes, the bars are intentional - the game only really supports 4:3 resolutions, but some widescreen monitors can't handle displaying them, and they simply stretch the image which doesn't look very good. To get around this, the game allows you to set a widescreen resolution, and them applies the black bars.

If your monitor has built in scaling options (or if your graphics driver supports it) then there it won't make any difference at all if you play in normal or widescreen resolutions.

As for the AI, we will of course keep working on it and tweaking it.

Posted: Tue May 19, 2009 4:12 pm
by corey420
the skirmish ai is better than the campaign ai because of the starting positions. i haven't played everyone yet but most are very easy on the highest level. i agree with Tost that the ai is not aggressive enough. on the other hand i have see the ai build a bunch of units it cant afford only to lose them all next turn. the ai will take time but if chess programs are now almost unbeatable then we can get a conquest ai of equal challenge.most top chess programs use the brute force method of calculating all moves then evaluating (the important part)the position. with limited pieces and moves in conquest that might be the best way to go.im no programmer so im just throwing out ideas.

Posted: Thu May 21, 2009 4:43 am
by Seleukos
The Campaign battles have been pretty uneven so far. I've lost a few to the AI and had to fight them again (sometimes multiple times :oops: ) to determine the winning strategy. Nice ! But most have been one-sided walkovers. It seems to depend a lot more upon the map and the victory conditions than the strength of the AI. When people post things like not being able to win at Tewkesbury due to a critical bridge missing from the map, it makes me wonder how much time Illustrious had to test the scenarios.

Is the AI to provide a challenging single player game, or just serve as a sparring partner to teach you the ropes for multiplayer ? I can accept the latter in an Xbox Arcade title where you can pick up some easy achievement points in the solo game and enjoy quick beer & pretzels style mulitplayer games on Live (which Conquest's attractively simple style and small maps would be well suited for). From a PC game, I expect better AI.

I certainly haven't written off Conquest. The AI could be better, but it's not totally brain dead.

Posted: Thu May 21, 2009 2:53 pm
by Cerbykins
Unfortunately during development there was a lot of call to have the difficulty toned down for the campaign because it might put too many more casual players off the game - there was one notable case where someone during the beta testing took a beating from Wakefield and I had to prove internally it was even beatable at all (it was, I'd already tested it myself... :) ). However, the prevailing feeling was that it was time to tone the campaign AI down. On top of that, the campaign is more of an addition to the core skirmish-based game, so the AI can never truly compete at it as well as it should (unfortunately a problem a LOT of games with single-player campaigns have) but by tinkering with the maps we can create unusual situations you'd never see in a standard game, and add something a little new to the game.

Tewkesbury is sadly one of a host of problems that are very similar. You wouldn't think that the scenarios had been checked and re-checked, and that there were many beta releases, all (iirc) with the full campaign set in to try and avoid exactly this. As it happens very late on a bizarre bug was found in scenario-based maps where the AI thought crossings existed where they didn't (even if not printed on the map, and all the printed ones were correct). The only symptom was that the AI player seemed to 'jump' or 'double-move' on occasion, and it didn't originally look like a fault in the map. In short, all the crossings had to be wiped out and re-applied to every single map, and somehow Tewkesbury got wiped but not re-applied. It came so late in the beta and development cycle, it's only being picked up now, as are many of the problems fixed in the recent patch.

Skirmish-wise, we're expecting some maps and setups will be harder to win than others, and that some will be sickeningly easy even with top AI. While we can't get the AI absolutely top-notch (not too many do) we can offer you ways to make the game a bit tougher.

To finish, the net result of having had to write then re-write a lot of the game's systems recently is that previously tested systems suddenly develop new bugs at the last minute, not seen in several betas. However, since the rebuilding is usually to wipe out more serious bugs (like AI that can jump tiles), it's better to look bad and fix a simpler bug in the short run than leave a more sinster bug lurking.

Posted: Fri May 22, 2009 12:26 pm
by corey420
i cant understand the reasoning behind toning to ai down. this is a strategy game. people who buy these kinds of games want to give their brain a workout. if u want this game to appeal to people who dont play strategy games then have the lowest setting of the ai do nothing. if you had an ai that was too challenging for your beta testers,and it doesnt cheat, that was just crazy not to include it or dumb it down.

Posted: Fri May 22, 2009 12:54 pm
by Jekky
the AI itself has never been "toned down" - all of the changes which were made to the difficulty of the game were done to the scenarios only, and done by modifying the maps and/or the objectives.

We recently spent an entire day doing nothing but getting the AI to output detailed information about what it was doing, and analysing that information to target the weak spots in the AI - in the end we discovered a number of small bugs which were responsible for some of the stranger decisions that the AI was making with regards to what tiles it chooses to attack (simply put, it had a bad habit of not joining up some of it's smaller territories despite having a clear path to do so).

With these tweaks (and possibly others depending on how many other problems there are and how much time we have) the AI in patch 1.3 should be an improvement over it's current state.

Posted: Fri May 22, 2009 7:45 pm
by marlborough
Great news! Keep up the good work you are doing :)

Posted: Fri May 22, 2009 11:35 pm
by Seleukos
Thanks, Jekky. Keep working on it. :)

Getting the AI right is difficult. It took years to get chess programs to the point where they provided a strong challenge, and that on an unchanging 64 squares. I remember players of People's Tactics wondering why the AI just sat around doing nothing in certain scenarios. Turns out the AI was building so many units it then choked on its own lack of suppies. There's a lot to take into consideration.

Posted: Wed May 27, 2009 1:46 pm
by corey420
your right about it taking awhile to get chess programs up to par, they still cant play openings on their own. chess has 6 different pieces with 6 different types of moves. conquest has nine pieces but all have the exact same moves. making a strong ai is difficult but nothing like chess. there a millions of tactics in chess. so far the only viable strategy in conquest is to own as much territory as possible as fast as possible. so even if the board had 1000 hexes, the ai should be doing the same as if there where 64 squares. i still think the ai values castles way too high. they are important but not to the exclusion of 3lvl units. i also am still seeing the ai over build units it cant support. i mean, im not cutting off half his land or anything like that. it just build a bunch of units takes some land then bye bye. i dont mean to sound like a jerk, i love this game, thats why im so interested in getting a serious ai to battle against. this game has the potential to be a classic and with a solid ai i think it will.

Posted: Wed May 27, 2009 3:45 pm
by Jekky
Much of the problem is about the time it takes to calculate the possible moves it can make - unlike in chess the "board" is not square, is not a fixed size and has tiles which cannot be taken (water and mountains), and the units are less restricted in how they can move - as a result there are a lot of calculations for each possible move that the AI can make, it has to work out what moves are physically possible, then if they are strategically sound, and finally the best/cheapest way to make the moves it actually wants to take.

The other problem we've had is that when trying to get the UI finished off and hunt down last minute bugs as well as deal with all of the admin associated with releasing, the AI can't get full attention - the game mechanics have changed a fair amount from the earliest iterations of conquest through to release and the AI hasn't always been caught up with the other changes.

Now that the game is out however, we can devote more time to tackling the AI properly, without having to be concerned that gameplay might change in some way - as a result we've also largely managed to overcome the first problem and get the AI to make it's moves a lot faster - of course since there is an artificial delay in the AI moves to allow human players a chance to follow what is going on, this will not be a noticeable difference for the most part. What it does mean is that the door is open for us to allow the AI to take more factors into account when deciding on it's moves - something which previously we were concerned might make the AI too slow.

Posted: Thu May 28, 2009 3:48 pm
by corey420
it sounds like the most important part is what the ai finds as a sound move. how is the ai determining that? does the ai factor in what the other player has and what they are doing? or is it just what his own situation is?

Posted: Thu May 28, 2009 5:25 pm
by Jekky
The AI takes a variety of factors into account when deciding on how good a move would be to take.

The AI will consider the value of a tile, how well defended it is, how much damage it will do to the enemy if it takes the move (in terms of the units that will die, both directly and as a result of not being able to pay for them) and the strategic importance of the tile (such things as how easily it will be able to defend the tile, if the tile will help it to connect separate territories together or if it will split an enemy territory apart).

At the moment it doesn't take anything other than the territory being targeted into account - the overall situation of the other player, or their general military strength for example - mostly this is because each territory is self-contained for the most part. It also lacks the ability to long term plan - it will in some cases make moves with a view to the next couple of turns, but most of the moves are decided on the current situation rather than the future one.

Posted: Wed Jun 03, 2009 12:20 pm
by corey420
okay, i have the new patch(v1.3). played a couple of campaign maps where you need to kill the marked unit to win. i replayed it a couple of times just to make sure it wasnt a one time event. the ai does not protect his marked unit at all. i won on the second move each time cause the ai brought the marked unit to front line. i know its probably hard to code that in but if thats going to be a campaign goal then its a must that the ai protect that unit. the unit just needs a buffer between itself and its front line. i also think the version should be displayed somewhere. i havent seen it on any screen so i just assume the game was patched okay.