Is this a Stupid AI Trick or what?

[SteveP]

I don’t think this one’s been discussed before, though in a way that’s kind of surprising and really makes me wonder what I am seeing. But here goes anyway:

The theme of this alleged Stupid AI Trick inquiry has to do with the AI failing to open fire as soon as it should. This is not, by the way, the old rant about some tank reversing back into a ravine, without firing, when it sees a Panther or something like that. This is about situations in which the AI has no apparent reason to withhold fire but does so anyway. I’ve done a whole bunch of tests and keep coming up with the same result, in both CMAK and CMBB.

Here is a description of one of these tests (this is using a modified version of JasonC’s Russian Training Scenario 110, if anyone wants to try this for himself – my apologies for using a CMBB example in a CMAK thread, but I do have a CMAK test scenario as well, in case that matters to anyone):

Situation: a single regular German HMG in a trench near the flag, in the middle of generally open terrain, facing two houses about 200m away. Hidden behind the houses is a platoon of green Russian infantry. Two or more of the Russian squads advance into the open toward the trench (with about a half minute command delay). Other parameters: EFOW or Full FOW, it doesn’t matter; no experience bonus for the AI which is controlling the HMG. I am playing this as if the scenario is beginning at this point.

Now, if this situation is played with the AI controlling the Russians (which may mean that the number of squads advancing and where they advance to could be different from what I described), and me controlling the HMG (which I am giving no orders to),the HMG opens up instantly on at least one of the squads within a couple of meters of that squad leaving cover. Similarly, this is what happens if playing this H2H (i.e., Hot Seat Solitaire). In short, in either arrangement, only the TacAI is controlling the actions of the HMG, and the TacAI is happy to fire at the Russians as soon as they stick their noses out in the open. And I think this is what we would all expect to happen routinely – i.e., barring some unusual reason for the HMG to have a short covered arc or something.

So what happens when the AI is controlling the HMG? It turns out that when I run this test (i.e., two squads advancing into the open) multiple times, the HMG will open up on them during that first movement only about 30% of the time. Of course, at that point, they are out in the open and the HMG will start firing at them as soon as the second turn starts. But I’m only concerned with what happened in that first turn – which, most of the time, is nothing!

Let me emphasize that I have now tried this type of experiment with a wide variety of scenarios and QBs, with differing unit configurations, terrain, etc., and I keep seeing this same phenomenon. The only explanation I’ve been able to come up with that fits all of these data points is that the AI is automatically giving its units Hide orders at the beginning of a game (when on defense), and that about 30% of the time the TacAI will overrule those orders in a situation like I described above. If this is the case, however, I think it is not good AI design.

I have also seen some other odd but related characteristics, but won’t elaborate in this first post because I’m guessing we have to get through the “SteveP must be smoking something” phase . Either that, or someone already has a reasonable explanation (I hope) for what I am seeing.

[JonS]

SteveP is smoking something again.

No he isn't

Yes he is

No he isn't

Yes he is

No he isn't

Yes he is

No he isn't

Yes he is

No he isn't

Yes he is

No he isn't

Yes he is

No he isn't

Yes he is

No he isn't

Yes he is

No he isn't

Oh, you're right. He isn't.

Right, now that's out of the way, let's move on ...

[Kingfish]

Maybe its the HMG team that is smoking, in which case they may be tripping from watching all those dudes with rifles coming after me, man!!

WFFFFffffffffffffff..........Mmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm.........

.....Cough Cough Cough

Cough

Cough

Cough

Whooooaaaa.......

[Corvidae]

Is there clear LOS? Is the mg jammed?

Did you try the stuff they are growing in BC?

(sorry)

[SteveP]

Interestingly, after posting the above, I happened to open a new (for me) scenario that offered a chance to experiment with this in a new way. It was one of those scenarios in which the battlefield is fairly segmented, so that at least initially you can have three relatively isolated battles going on. Also, because the scenario was designed to be played with the AI as attacker, the designer had placed the attacker units, in cover, very close to the defense (a good technique for making the AI a better attacker).

Anyway, to run my test, I took the attacking side. I set things up so that one platoon would advance into the open in its segment of the battlefield during the first turn. Then a second platoon would advance into the open in its segment during the second turn. And so on with a third platoon in the third turn.

What happened was pretty much what I expected. The first platoon took no fire in the first turn, even though it was out in the open a short time after starting its movement. However, it got plastered by a bunch of MGs at the beginning of the second turn. At the same time, the second platoon started out in the second turn -- without a shot being fired at it until the start of the third turn. Same thing with the third platoon. I tried this whole drill a second time and got a slightly different result -- but, as I noted above, there seems to be about a 30% probability of the defender shooting in that first turn, so with three platoons moving, I'd expected something different.

So, there you have it friends. I know this is happening, at least on my computer. Frankly, I don't like it which is why I spent so much time trying to figure it out. But maybe one of our tactical grogs will suggest that this is actually sensible behavior on the part of the AI and that we all should be doing this when we play defense.

[Cuirassier]

I don't have CMAK, but I will post anyway. I have noticed this on the Russian Training Scenario 200 series. The AI is silent on turn one, but on turn two, the MG's lay into you.

[SteveP]

Well, guys, this may not be as big a puzzle as I thought. I happened to look in the manual, related to another subject, and came across the section that talks about using the Hide command. Lo and behold, it says among other things that "skillful use of hiding, especially as the defender, will allow you to remain invisible to your opponent until you feel the time is right to engage in direct combat." There are other comments in this section along the same lines.

I think the AI is, in fact, giving Hide orders to all its units when on defense. So, the question is: does that make sense. They certainly don't need to be hidden to be invisible, in FFOW or EFOW. But having them in hide mode certainly handicaps the performance of the AI's troops. And, heaven knows, the AI doesn't need any more handicaps. Maybe we should start a "Free the AI" campaign.

Anyone else want to weight in on this one?

[SteveP]

Originally posted by Cuirassier:

I don't have CMAK, but I will post anyway. I have noticed this on the Russian Training Scenario 200 series. The AI is silent on turn one, but on turn two, the MG's lay into you.

Yeah, Scenario 200 is a good example of why I think this is a handicap. If you correct the map side ownership problem in the Editor and play this using the default setup for the Russians, this is what should be happening in the first fifteen seconds of play: the sniper fires at one of the T34s, the MG opens up on the visible Russian infantry and the PAK gets off a first shot at one of the T34s. By the end of that first turn, there should be at least one dead T34, and probably two, and some messed up infantry. However, most of the time, nothing happens in that first turn. So the Russian gets a free pass.

[securityguard]

Originally posted by SteveP:

Anyone else want to weight in on this one?

I always assumed it's the hide command. You get similiar results with a defender who doesn't use arcs - he'll open up immediately at the start of the next turn.

[SteveP]

Let's keep in mind that if this is what is going on, it's the result of a game play decision. I think we all understand and accept that many of the AI's limitations and behaviors are the result of the difficulty of programming an AI. This does not fall into that category. I also don't think it falls into the category of modeling decisions, like how readily a green infantry unit pins when it is hit with fire. We can argue something like that, but the game had to draw a line somewhere and there it is. This is just a simple either-or decision: to mandate that the AI hide the defending troops or not. At the moment, my view is that they made the wrong decision, because I think that it handicaps the performance of the AI rather than helping it.

[Moronic Max]

Let's keep in mind that if this is what is going on, it's the result of a game play decision.

Maybe. Or maybe it's an oversight; something left in the code from BO, before things like covered arcs were around, when if you didn't issue a hide or ambush order, your guys would shoot at everything they saw. It's possible nobody thought to remove the relevent code (or thought they had removed it).

Just a guess.

[SpitfireXI]

Why has is taken this long for one of us to find this?

[Hensworth]

Have you thought of surrendering immediately on the first turn and checking whether the AI MG is in hide mode when you look at the map ?

[SteveP]

Originally posted by Sgt_Kelly:

Have you thought of surrendering immediately on the first turn and checking whether the AI MG is in hide mode when you look at the map ?

Terrific suggestion! Never thought of it. Ran a test using a CMAK scenario I made that works similarly to the one I described above. The MG was NOT hiding. To sort of test the test, I used the Editor to make the MG hide at the start. When I surrendered, it showed the MG was hiding. So your test suggestion seems to be valid.

So, now I'm back to where I started. The AI is doing something strange, annoying and IMHO bad for its performance, but I haven't a clue what it is or if there is any way to prevent it.

As for why this wasn't noticed before, it may be that people simply chalk it up to the randomness that is built into the game model. However, I am convinced that that is not the explanation. Something in the StratAI or the OpsAI is preempting the TacAI from firing. And if it's not a universal Hide command than I don't know what it is (and I don't think it's covered arcs, because I don't believe the AI uses them and anyway I ran some tests specific to check for that). So there you go, folks!

[SteveP]

Oops ... spoke too soon. The AI IS giving its troops a universal hide command. It's just trickier to confirm it using Sgt. Kelly's suggestion.

If you surrender at the very beginning of the first turn, the Ops AI has not had a chance to give any orders yet. However, if you play through the first turn, quit and launch the same turn using autosave, and THEN surrender at the very beginning, you can see that the MG is hiding. My guess is that whatever commands the StratAI and OpsAI gave originally are preserved by the autosave process.

So, assuming I haven't misinterpreted any of this, I feel that at least I know what the stupid AI trick is.

Now, here's where the whole thing gets kinda funny. I edited my CMAK version of this test so that the trench and MG are relatively close to the green US troops. I also changed the MG to crack (which means he's going to obey the hide order more rigorously). I found that I could frequently use the Fast command to run my US squads right into the trench before the MG opened fire. How's that for a stupid AI trick??

[SteveP]

I have possibly a fuller answer to the "why hasn't this been seen before?"

Imagine a platoon of enemy regulars in the woods 100m across an open field from you. You have a couple of regular platoons that you send advancing toward them (not exactly JasonC approved infantry tactics, but bear with me ) Now none of the enemy has fired yet, or been fired upon, so far (that's essential to this stupid AI trick).

In that situation, the odds are fairly good that one of the enemy regulars will get spooked enough to disobey the hide order. He starts firing. Your guys start firing back, which wakes up the other enemy squads, and you have a fire fight. It feels like what should be happening. However, the reality is that your guys probably got farther across that open area than they would have, had the enemy not been given hide orders. Now someone may argue that this a trivial problem, but consider this: if I am going to advance across that open area in this situation, my preference is that those squads be crack or elite, because they are almost certainly not going to fire until I am right on top of them. If they are green, they will most likely get a fire fight going while I am still some distance away. To me, that's just perverse.

But more generally, I think it is perverse for the OpsAI to be preempting the TacAI. The AI's decision-making gets less and and less reliable the higher you go on the decision ladder (that's not a complaint, just a fact). IMHO, the TacAI is a pretty good decision-making program -- better in fact than some players seem to realize when they preempt it themselves with lots of covered arcs and the like (moreover even BFC has said that players are well-advised to place substantial reliance on the TacAI). But at the beginning of the game of the AI is basically squelching the TacAI and turning the first line of decision-making over to the OpsAI.

Anyway, I really am going to try and stop ranting. Any minute now. I promise ...