How weak is the A.I.?

[thedudeabidez]

This weak.

Campaign.jpg

[thedudeabidez]

Moscow Campaign, vs. Von Bock. Axis entirely wiped from the map with a few turns left to go, actually.

Don''t mean to be ragging on Shenandoah since the this is an EXCELLENT game against a human player, but the A.I. clearly needs some work.

Two suggestions I can make:

1.) The A.I. doesn't seem to understand supply; it frequently made advances into an area to attack even though they will surely be cut off on the next Russian move. The entire game consisted of reckless German attacks, cutting them off, isolating then eliminating.

2.) The A.I. passes as the Axis even when it still has some strong armored units and infantry way to the rear which really need to be moving forward.

[thedudeabidez]

That map above gives a new meaning to "Red Square."

[Jarling]

I think this is a very historically correct situation.

For june 1941 or january 1944, that is.

[GrimmySwe]

I must agree with the OP. An extremely good game but I sense that the AI currently is on the easy side. I consider myself a well experienced player with a lot of play time in different tactical and strategic games. So yes, that might count for me. But currently I don't see any options on cranking up the opposition. I play with Fast AI turned off (but that might just be visuals)

I do though fully understand the challenge with creating a good AI opponent and I only hope that we can help in pointing out things that might be possible to improve.

The first thing that I noticed was that the Expected battle outcome seems to be wrong 99% of the times. Yes, its a wargame, there should be a lot of variety but somehow my personal sense was that the numbers were always off by at least two hits. I though need to play more to really get some statistics backing me up, but as I said, this was my first impression.
So with that in mind, if AI depends on those numbers it might get some skewed utilities.

Second was the understanding of supply already stated, a lot of times Guderian just drove units trying to circle without actually bothering to secure his own supply. Well, for the first 12 turns they wont be totally cut off due to air supply, but its a very simple tactic to open a hole and lure in a couple of panzer into the hole.

As the third note I need to qoute your post regarding 7 ways to improve your game. There one of the items is focus, but I lack a little of that from the AI. In the complete campaign I felt it was an all out assault along the entire front. When I play I usually only concentrate on one part of the front, using leftover time to move other units.

Last I again want to thank for a game well crafted. My overall experience is positive and bug free so far. And I should try to play against humans but Ai games are easy to setup and can be played when I want to.

[Ralpher]

The AI seems to have a weaker understanding of supply than BotB. I can easily cut off the attacker's supply and the attacker is not keen to reconnect it when it should be. Maybe it was a byproduct of the larger map in DoM? Or there are so many defenders it is easy for them to cut off supply?

[Coreysr]

I agree with the others posts regarding ai weaknesses. The most glaring flaw I've seen though, next time I'll try to save some screen shots, is this; I had moscow partially surround with 1 unit in each square, moscow having three units. It was the middle of the turn and the soviet moved all 3 units to attack 1 axis square completely abandoning moscow. Next turn I obviously took moscow and the game. As others have said, I think you guys have made an outstanding multiplayer game. But the ai is so bad I don't recommend playing it as it really only leads to bad habits and you never get a feel of what is required to win the game.

[s_nkarp]

I agree with the others posts regarding ai weaknesses. The most glaring flaw I've seen though, next time I'll try to save some screen shots, is this; I had moscow partially surround with 1 unit in each square, moscow having three units. It was the middle of the turn and the soviet moved all 3 units to attack 1 axis square completely abandoning moscow. Next turn I obviously took moscow and the game. As others have said, I think you guys have made an outstanding multiplayer game. But the ai is so bad I don't recommend playing it as it really only leads to bad habits and you never get a feel of what is required to win the game.

I agree that multiplayer is a better experience, though I still enjoy playing against the AI, working to improve past personal performance. It's particularly challenging to see how quickly the Soviets can dominate in a campaign. Not everyone's cup of tea, but fun for me.

As for abandoning Moscow completely, that shouldn't happen in the latest release (1.0.2). If you received Moscow via promo code, there's a chance that you got an earlier version -- but you can update on the App Store. If you ever see the AI move the Moscow militia out of Moscow that is actually a bug -- please report it (via the "Feedback" button in the info menu -- and we'll get it fixed.

AI in Moscow is a harder problem than the AI in Bulge: more spaces, more units, more turns, more long-range implications of moves. We are working on a fundamentally new algorithm that will help a lot -- unfortunately I can't give a date for it as technical issues remain outstanding.

Some of the approaches that we're considering as interim fixes:
1) Let the AI cheat, in well defined ways (e.g. +10% on attack, chance of units getting a second activation, relaxed supply, altering time passage)
2) Explicitly changing victory conditions for AI games.
3) Allowing players to tell the AI to "try again".

At this point my sense is it's better to direct our effort to fixing it right than on interim hacks - but opinions much appreciated.

[thedudeabidez]

Those are all not really ideal fixes; what you need to do is program it to understand supply.

[s_Titus_Tears]

Hello All,

I needed to add to all this.
I was disappointed to discover a victory after 12 moves for myself yesterday (Axis) due to a silly Allied Russian decision to leave Moscow lightly guarded.
On attacking with three Panzer divisions to move around the right of Moscow I pushed three Russian Infantry units out producing a gap between
myself and another group of German panzers further left. The units guarding Moscow city then moved into the free gap between leaving only 1 infantry units.
My three panzer tanks around on the right hand side of Moscow where able to move into Moscow and force out the 1 infantry unit left, easily.
I was really enjoying it all up until the way to easy finish.

Regards…

[s_nkarp]

Those are all not really ideal fixes; what you need to do is program it to understand supply.

That is my sense as well - disposable interim hacks really aren't worth much distraction from getting it right.

[s_Titus_Tears]

Hello All,

I was able to defeat the Allies and take Moscow completely with three panzer divisions moved into Moscow on turn five — taken it completely by the very first move in turn seven.
Enjoying the game and trying new things in it, I am also still trying to get to grips with things and how it works.
Still not happy with how the Russian troops can move much further distances even though I control most of the rail network, makes no sense to me.

Regards,

[ambext]

Hello guys,

I would like to thank Shenandoah for making great games. I played a lot at Bulge (with all extensions), am an impatient Kickstarter backer of Gettysburg, getting even more impatient about El Alamein, and do enjoy a lot Moscow.
Nonetheless something worries me a little bit, and it’s the level of the AI. I lost my first game on Operation Typhoon. I had a draw at the second one. Till then I mostly played the full campaign scenario, and I never lost a single game, even with the lousiest of focus.

Dear Shenandoah staff, I am sure that you receive a very diverse feedback on it, with some saying it is too hard and others saying it is too easy. I also read you blog entries about the game, and particularly the paragraphs dedicated to the AI.

My understanding is that you now have built a framework to make AI behaviour more flexible. This is great news. It would be even greater news to learn that you will implement a “level of difficulty” setting in your AI. The current level of performance could serve as the middle point, if you receive mixed feedback. However some blatant shortcomings in the current model would still have to be fixed [edit : nkarp mentioned some of them in its recent post – I won’t repeat them here]

Regarding testing you apparently do a lot of it using AI vs AI configurations. Why not use more beta testers, specifically for the AI? I am sure you would find some volunteers out there (I would gladly be one of them). With the ability to rewind a game and the added ability to transfer you the recordings, beta testers could point out to specific “dumb” moves in games. You could then study them and improve the AI thanks to all that feedback.

Hope this message helps; it is why I wrote it for in the first place. I am happy to discuss the matter if needed.

Best regards from a French player living in Australia.

[daviddunham]

Still not happy with how the Russian troops can move much further distances even though I control most of the rail network, makes no sense to me.

If you controlled the rail network, they couldn't do it — railroad movement is only through spaces you own. (Which always annoys me as a Soviet player, when I can't use rail move.)

And while I am not an expert on the historical battle, I have to assume that this was a significant factor during WW2.

[Miguel_TSS]

Regarding testing you apparently do a lot of it using AI vs AI configurations. Why not use more beta testers, specifically for the AI? I am sure you would find some volunteers out there (I would gladly be one of them). With the ability to rewind a game and the added ability to transfer you the recordings, beta testers could point out to specific “dumb” moves in games. You could then study them and improve the AI thanks to all that feedback.

Actually, AI vs AI games are only the first and most basic quality bar. But because our end goal is a General that plays against people far more time is spent vs Human testers, both internal and external. Even post launch, I use your feedback to focus fixes, and discover behaviors even our testers didn't catch. If there is interest, I could do a blog post on the steps to creating an AI General?

As a minor aside, I think most of you are reporting on the Campaign not feeling challenging or interesting enough. But how are the Generals in the other Scenarios playing with you?

[Pat]

Still not happy with how the Russian troops can move much further distances even though I control most of the rail network, makes no sense to me.

If you controlled the rail network, they couldn't do it — railroad movement is only through spaces you own. (Which always annoys me as a Soviet player, when I can't use rail move.)

And while I am not an expert on the historical battle, I have to assume that this was a significant factor during WW2.

There's also the issue of different gauges .. the Russians used a different gauge track and rolling stock than the Germans. So until the Germans got there hands on a lot of Russian trains and wagons, they had to lay their own track as they went and connect them to German railheads far from the front line.