Catch 22 - the AI

[sajer]

From reading more than a few threads - and from my own playing experience it seems that (as in all games) the AI sucks.

You can adjust the difficulty settings all you want, but it seems all that does is put more troops in the field. It does absolutely nothing for strategy thinking. I am not blaming Matrix games, Gary Grigsby, playtesters...etc..etc.. That's just the way it is. Computer AI technology can only take us so far, gents.

I have seen this in almost 99% of the games I purchase.

If you play the Germans, you crush the Russkies until the mud sets in. Then you can most likely survive the winter. Even if he pushes you back through sheer weight of numbers. I have yet to see an intelligently planned offensive from the AI. A smash through the line and then an exploitation (or pincer move).

If you play the Russkies than you can stop the Germans well SHORT of thier Dec. '41 historical stop.

I guess that is not the fault of any creator of this game.

I mean we are are not playing "Deep Blue" guys. Even if we were, we would probably only lose in chess and jeapardy. It really says alot about how intricate the human mind is - able to process vast emounts of information and plan well ahead to move troops for a breakthrough in a well scouted areas of the frontline.

This brings us to the other piece of the puzzle.

From reading hundreds of threads it seems that "most" Human vs Human games do not last past the winter of '42. This is they know that the Russkies will eventually smash thier asses back to Hitler's Bunker - no matter what they do. (If I am wrong on this, please correct me).

The game designers tweak the system in order for you to kill less planes on turn 1, or to correct OOB mistakes etc..etc.. but they can't make the AI more intelligent.

Therein lies the true Catch-22. You can't play the computer- becaue it is to damn dumb! You can't play another human player, because he can see the inevitable coming and would rather resign than to continue on for months in a lost cause.

Please correct me if I'm wrong - or let me see some examples.....

[Commanderski]

I think you should review some of the previous threads regarding the AI. There is a fairly recent one in the War Room. The general opinion is that the AI is one of the best for this type of game. There are also numerous suggestions from the developers and Beta testers on how to adjust the AI settings.

Also in Human vs Human games if the players stick it out they can go into late '44 and into '45.

[sajer]

I think you should review some of the previous threads regarding the AI. There is a fairly recent one in the War Room. The general opinion is that the AI is one of the best for this type of game.

You might be right. But of course that doesn't happen to change the fact that in my game vs. the computer (I am now in Oct. '44) Not once has he advanced more than 10 miles in any offensive action.

There are also numerous suggestions from the developers and Beta testers on how to adjust the AI settings.

Well if there is I missed it. But if this were true wouldn't it be in the preferences screen? Or agin if this were true, why wouldn't it be set at the maximum setting already. If you could tell me how, or point out how this would be possible, please let me know.

Also in Human vs Human games if the players stick it out they can go into late '44 and into '45.

I'm sure there are cases where that is true. But I've never heard of a game going to the bitter end. Where the artillery was bombarding Berlin. But I'm sure that most cases games end in'42 or '43.

[karonagames]

If you play the Russkies than you can stop the Germans well SHORT of thier Dec. '41 historical stop.

If you play on normal you will - challenging and hard are a much different story (check the AARs - there was an early one with the SU playing impossible AI). The AI settings do not increase manpower levels: if SU morale is set to 110+ they get bonus units to help maintain a solid front line but they are filled out from the normal manpower pools.

But you are right changing the settings does not make the AI smarter - the units just fight harder.

[Fänrik Stål]

ORIGINAL: sajer


From reading more than a few threads - and from my own playing experience it seems that (as in all games) the AI sucks.

Yes. But there is sucking and there is sucking. Compared to other games I know of, this is Einstein.

A smash through the line and then an exploitation (or pincer move).

I've seen that. Several times in the same game as a matter of fact. But this was not on normal level, and not with the benefit of experience. The AI knows how to do it, but the game allows countermeasures that are unrealistic from a historical perspective (checkerboard or carpet defense). The human learns, the AI doesn't.

If you play the Russkies than you can stop the Germans well SHORT of thier Dec. '41 historical stop.

Yes, on normal level there is no problem doing that.

[alfonso]

The main problem, in my opinion, with WITE AI, in contrast to chess playing programs, is not about rule complexity or some intrinsic feature of the game. While during the last decades thousands of programmers have struggled to make the most effective chess evaluation algorithms and chess playing software, as far as I know the number of persons working in WITE AI can be as low as ONE. This has to be a bottleneck.

If there were the possibility to develop alternative WITE AI engines as there is with users-developed scenarios, and different AIs could play head-to-head as well as versus humans, Darwinian selection alone would create a boost in AI performance. The computers would tend to be rather “gamey”, though, and therefore the programmers should be alert to keep this gameyness in check.

Besides, it seems to me (with zero knowledge on this issue), that by the way the AI is programmed in war games, the do not benefit of “thinking” for longer time periods (or clock CPU periods). In contrast, the same chess programs that were of middle strength running in a Pentium II 10 years ago are unbeatable beasts in present day computers, with 3 minutes for each movement.

[sajer]

Well.. my goal was to start some posts on this and see what your responses would be. I guess I got that.

I did not mean to imply (and you can look at my original post again) that the designers had anything to do with it.

It is the simple fact that our ordinary garden variety PC's/CPU's cannot handle the strategies that are needed to make a challenging opponent. We live in an age where the processing power is just NOT THAT INTELLIGENT.

Yes, they may make a small pincer move or offensive with a big buildup of forces - but we are always 12 steps ahead of them and we adjust.

For years, maybe decades.. I have dreamt of a great computer simulation of the War in the East.

This game is as close to a great game as can be. It has great detail/maps/design and correct OOB's. I also like the intricies of the game - like the transferring of units, command points etc... etc..

I would not have been playing this game on a daily basis for 5 months if I did not love all those things.

What I'm simply saying is that I have spent 1942, '43, '44 bracing myself for (at least) a little slap in the face and I haven't got one.

I am sure it is not the programmers,playtesters etc.. fault. It is deep down - it is the computer vs the human - no match.

[alfonso]

If a computer can land a 747, beat the World Champion at chess, and help to make an image-based diagnosis, in principle it could also be taught to beat me at WITE. But the thousands of working hours needed to do that are not available.

[Joel Billings]

The bottom line given the reality of the situation with AIs is that the attacking AI especially needs an advantage to get moving (this goes for both the Germans and the Soviets). The German AI needs more of an advantage because a good German player can do more with encirclements than the AI that must be more about brute force. Given the dynamics of the campaign, there must come a tipping point where the Germans are no longer the attacker, and the initiative has shifted to the Soviets (if the Soviet has a chance to win). If you play on Normal as the German, even if you aren't a strong German player and you lose the initiative at some point, you'll probably do good enough that the Soviet AI, without help, will not be able to push you back to Berlin (or even Minsk). When you play at Challenging, once you lose the initiative the Soviet AI may be able to keep things interesting for you.

As several have posted, the AI does not learn, so once you as a human player has learned (whether through repeated play and/or reading the forums), you need to give it help. The basic rule of thumb hasn't changed in my opinion in 30 years. When you start out you can play at normal. After a game or two, once you know the basics of the game system, you have to go up one level of difficulty and this should provide a challenge for a few more games (switching scenarios allows you to stay at this level longer as you're dealing with some new/unknown elements). After a few more games, good players looking for a challenge should go up two levels of difficulty. At this level, we still expect a good player to win, but it could be a challenge, or may require a few repeated plays. Some scenarios and sides are more difficult than others (for both the AI and a human player) so there is some variability, but it's a good rule of thumb. Unlike games in the 80s, we give you several ways to customize your level of difficulty if you want. The AI code is much more complicated now than it was in the 80s because there is much more to the games now than there used to be. By using the play levels you can get hundreds or literally thousands of hours of fun against the AI. I consider that good bang for the buck.

We've been happy to see many posts in the past year by players detailing how they've liked the AI. Gary spent a lot of time on it, and continues to try to improve it where he sees reasonable opportunities to do so (by that I mean reasonable time investment, good probability of success and good chance he won't screw something up). We think it stacks up very favorably to other wargames. I've played many games against the AI and as long as I crank up the difficulty I can still enjoy playing against it. At the same time, I encourage players to try PBEM, if even in the smaller scenarios (I've only had time to PBEM in some of the new smaller scenarios we're working on, but I've really enjoyed these games). If you've got the time, it adds a whole new dimension to the game.

[Rosseau]

Myself not being the sharpest tool in the shed can be a blessing in AI games. Also, I don't analyze and instead play more on instinct - which is probably pretty historical. The commanders in these battles had far less information than we have in the game. They were also tired, hungry and under artillery fire while making their decisions. The opposite is the board game junky that will figure every odd. Of course, using my knee-jerk play style in PBEM will get me slaughtered.

[Flaviusx]

ORIGINAL: alfonso

If a computer can land a 747, beat the World Champion at chess, and help to make an image-based diagnosis, in principle it could also be taught to beat me at WITE.

I'm not so sure about this assumption. All those other things may actually be simpler to do.

[Klydon]

Landing a 747 by computer is not difficult at all. The technology has been around to do that for a long time. (Same thing for landing planes on carriers, which is how I know about it). That a chess program can beat a Grand Master at chess is somewhat interesting, but considerable time and expense has been spent in that area. Chess, while complicated, pales in comparision in terms of varibles compared to a game like WiTE. Chess only has to consider 32 playing pieces and a board that has 64 possible locations.

Part of the trick of a game AI is not only the time to invest in making it better doesn't really exist (not time effective) but the other issue is there is not time between turns for an extremely complicated AI.

Consider the AI is as good as it is AND it performs that well given the time constraints its under. One of the big complaints for game players is how long the AI turn takes (not necessarily the case for WITE; just in general), so most game designers are sensitive to this and design accordingly. Even if they redesigned the AI, what is an acceptable trade off time for a "smarter" AI? 30 minute AI turns?

[Aurelian]

ORIGINAL: alfonso

If a computer can land a 747, beat the World Champion at chess, and help to make an image-based diagnosis, in principle it could also be taught to beat me at WITE.

No one has yet programmed a computer to play and win at Go on a full size board. The largest game of Go being completely solved has been played on a 5×5 board.

[Apollo11]

Hi all,

ORIGINAL: Flaviusx

ORIGINAL: alfonso

If a computer can land a 747, beat the World Champion at chess, and help to make an image-based diagnosis, in principle it could also be taught to beat me at WITE.

I'm not so sure about this assumption. All those other things may actually be simpler to do.

Yep... they are... those tasks are more simple although they seem complicated at the first glance... the WitE has thousands of HEXes and thousands units... it is extremely complicated to program AI...


Leo "Apollo11"

[sajer]

I think Joel Billings nailed it.

It is the fun you have playing the game that makes it worthwhile. I myself have played this game everyday since I bought it in (July I think). I am handicapped - so it is EZ to sit around and play. It also gives me a lot of "face" time with the computer. It has been a joy, but I have managed to figure out how to kick the AI's butt (every time).

I always loved Gary Grigsby games. I think that he is the best coder in simulation games....period.

As far as the discussion, which talked about computers and thier capacity...Well consider this.

In July 1969 Apollo 11 amde the moon landing with help from thier computer. It is now a fact that advanced cell phones NOW have more computing power than the cabinet sized computer that was aboard Appollo missions. It is amazing how far we have really come. But, alas we haven't come that far.

I think when the time comes that they have advanced to the stage of "thinking" computers - games will be awesome. Also it might even be scary. "Skynet" anyone? LOL

As far as this game goes I think that playing it non-stop fot 6 months has been a good investment. Money very well spent.

I think I might go on and play the 42-45 scenario and try to organize the German Army into something that can stop the Soviet juggernaut.
Crank up the difficulty...play that for awhile - and then I think...I am of to WiTP land. I LOVE detail.

Anyone here play that game? I played the original. What is the one they play now? - War in the Pacific (Admiral's Edition)?

I think that would be a good game to play - since the Japs were on the defensive since June '42 (Midway) anyway. I think it would be fun coordinating and conducting operations in the largest theatre of war on Earth - the Pacific Ocean.

Anyone here clue me in on that one?? Thanks

[alfonso]

ORIGINAL: Apollo11

Hi all,

ORIGINAL: Flaviusx

ORIGINAL: alfonso

If a computer can land a 747, beat the World Champion at chess, and help to make an image-based diagnosis, in principle it could also be taught to beat me at WITE.

I'm not so sure about this assumption. All those other things may actually be simpler to do.

Yep... they are... those tasks are more simple although they seem complicated at the first glance... the WitE has thousands of HEXes and thousands units... it is extremely complicated to program AI...


Leo "Apollo11"

I doubt that any carbon-based intelligent form considers the game as a set of thousands of hexes and thousands of units. Precisely the recognition of the limited human ability to deal with many units simultaneously led to the hierarchical structuration of military forces into Army Groups, Armies, Corps, Divisions, Regiments,…

I have with me a chess book written in the 60’s, “Modern Chess Strategy”, by Ludek Pachman. In the final chapter there is a short mention about the future of computer chess. It says: “This (abstract nature of chess strategy) is beyond the ability of the most perfect machine: is an activity reserved to the human brain, and it will always remain so”…Basically, this is an underestimation of what can be achieved by a dedicated groups of programmers working for a long time with steadily more powerful machines. Even GO playing software, arguably much more difficult to program than chess, although loses against professional players, defeats the amateur and casual player (even in 19x19 boards). There is no need to “solve the game” to defeat an amateur. Chess programs do not “solve the game” to massacre chess grandmasters.

One of the most “human” activities is pattern recognition, especially face recognition. It is thought that our brain cortex is somewhat hardwired to that task. But after decades of Government-funded and private research, the facial recognition software has improved a lot, and can even outperform humans. There is no reason to believe that thousands of persons working decades could not program a very skilled WITE AI. Already it seems miraculously able in relationship to the effort invested.

But perhaps I am wrong and the ability to play WITE is the non-plus-ultra of human achievements.

[Shadow666]

I think the AI for this game is good myself, the problem I see is that most players play the game with god like powers and do stuff that the Germans and Soviets of the time did not have the option of doing. If you want a more historic and hard game try holding everything as the soviets in 1941 even if that means having large armies destroyed and do not stop advancing in 1941 even when winter comes.

I think a lot of people forgets that a lot of the big losses on each side was because of Hitlers and stalin's crazy minds and not looking at what their Generals wanted to do. You just have to look at Stalingrad to know this, the German Generals could have saved the troops their with no problems if it was not for Hitler. So just play each side like you have Hitler or Stalin ready to kill you if you retreat and you will have a more historic and harder game

[Straticus]

I understand the need to give the AI help vs a human player, after all we are talking about a rather complex a detailed level of gaming considering all the forces and factors involved. There is no way to make an AI more clever, only to incease its ability to resist or recover more quickly. I really do not care for a-hsitorical settings, preferring to enjoy the game as a replication of history. But lets face it, the human player especially in the full campaign, has the value of hindsight, multiple game plays, and tremendous freedom to efficiently allocate resources from the top down to their full advantage.

I am of the opinion the smaller scenarios are in fact more challenging than the full campaign for a number of reasons. The time sensitive objectives and the unreasonable contraints placed upon his historical counterparts make for some good challenges. The most balanced scenarios seem to be the ones you start when the combatants are approaching parity as in 42-43.

[juret]

my thoughts on the AI

When i first started playing WITE i played normal difficulty. I didnt use buildups and had bad knowledge of game. When i played axis i crushed AI before 43 in 41 campaign. When playing soviet it was even easier. Just used checkerboard defence of inf brigades and in 41. By 42 i was taking hexes in poland with an monster amry of guard cav corps (only made those nothing else )

Game was boring by then and i tryed other scenarios. The 41 campaign is fun but it gets so boring making opening move as axis or make all retreat moves as soviets. So i started playing pbem games but they did goto slow for me.

For now i play 43 scenario as axis on challanging. I have to say the red hordes are strong and keeps attacking all over. i have very hard counter the AI.
We had huge HUGE HUGE tank battles around kharkov attacks counter attacks and static slugfests with tank units. When AI understood i moved basicly everything of my mobile units in center to check em he withraw. half of the forces are now attacking real hard in the south and hes rolling up my defences in the north. I still have my units depleted and badly hurt in the middle. I think this save is the most challanging and fun i ever played in strategy games during my 20 years of gaming. The scenario might not be balanced at start but ai acts realy smart.

[Redmarkus5]

I sense that a part of the problem with the AI is the objective setting. The AI seems to ignore city hexes and other key historical objectives on the defense. Perhaps this is why it doesn't attack aggressively as well?

Obviously, the human opponent can anticipate the AI's goals if cities and production are its focus, but guessing the enemy's goals is only one part of the challenge - Zhukov guessed that Moscow was the Axis main objective as early as June '41 and the STAVKA aligned its forces accordingly. As the player, you will still need to position your forces and supply them to face whatever AI plan is adopted.

By making the AI drive harder for certain goals and simultaneously toning down the magical Soviet ability to move masses of troops all the way across Russia in one week, a model could be developed where the AI makes a random choice between 3-4 strategic plans on turn one and then drives ahead to achieve the selected goals until Winter '41 (i.e. Leningrad AND Moscow, or Moscow only, or Leningrad AND Moscow AND Rostov, etc.). Then the Axis AI makes a line for the winter while the Soviet AI chooses 1 or 2 key cities as its goal for the counter offensive. In the spring it chooses another 1, 2 or 3 major objectives from those remaining and focuses on them, based on the balance of forces on the map.

Rough and ready, but better than what we have now. I am not convinced that the WiTE AI is the pinnacle of what can be achieved. A bit more abstraction would allow for a lot more flexibility, giving the impression of intelligence, or at least some level of free will.