New AI not as challenging

[MasterSncKnight]

I think this is an issue on how their limit were set. I noticed that their largest army cap is different depending on the difficulty set (I don’t remember the numbers rn tho). So after reading the post, I think the reason they work better on lower pop cap is because their “agressive state” checks for the population % compared to max pop while the pop limit is hardcoded (as in, if hard = 50 then it’s 50 regardless of pop limit). This would make sense, since if it tries to atk once the pop reached 40% (assuming they have the same amount of vills as army) then if the max pop limit is 200 they’d try to atk once they reached 80 pop, but if their diffic cap is 100 they won’t try to atk if the max pop is set to 300.

The one thing I notice tho, is that at first sight it might look at easy fix: just change the difficulty cap to check for the % of max pop instead, but the problem is how’s the AI coded to macro? if the AI uses the same macro code in all difficulties it means that both moderate and extreme would have a big army really quickly at higher max pop. I actually do feel like they’re a bit worse macro-wise at lower difficulties because of differences of feudal times, but this is something to have in mind when rebalancing it.

On the other hand

Something I noticed that improved a lot in this AI is doing and dealing with rushes. On HD I could easily steamroll the AI with some archer rush, but in DE I think even at moderate they’re much better at defending from rushes, in fact it was a huge annoyance for me to be able to keep the pressure, they were continuously baiting my army to go under TC fire and it actually ended up going castle before me, so I ended up having to rush some spears since they started sending knights to atk my skirmishers. Trying to rush the AI and keeping pressure imo was a lot of fun, things only got easier once I managed to get some rams ready to destroy their TCs. And all of this makes sence with the pop limit thing, when you try to rush you’re going agressive with a smaller army, which means the AI will have a much better chance to fight back, have even fights and maybe even push back.

[Xilmi]

I want do give some feedback on the AI.

AI picks GRUSH-Strategy on Maps where that’s not really suitable, like coastal for example. It usually transitions out of it but by that time it already has a lot of harbours.

On the other hand when the game-mode is not 1v1, it seems to default to strategies that don’t seem suitable for the map-type. So it went for a lot of land-units on an Island-map instead of trying to assert dominance for the water (as it has tried on maps with lakes or coastal before). So they couldn’t even make ships to threaten anyone because they had all their pop in land-units.

Also it seems to have too much of a bias to attempt to go for FC without any units in Feudal-age under certain conditions where this is not a good idea. AI should always reconsider and swap into a Feudal-type of strat that and get some defensive-military buildings, when it scouts the opponent to have military-buildings already.

Worst part, however, is how the AI reacts to “stand-ground”-archers near their TC. It gets stuck in a loop between garrisioning and going back to work. That behavior would be great if the archer wasn’t on “stand ground”, as that would lure them into the range.

So I think they should consider the command of the archer. If the archer is on hold position, the AI should probably charge them with villagers or let the villagers drop to low before they run away in order to have them gather more stuff that can be used to quickly get out defensive-units.

[Darkness01101]

@FirmSaturn67352

Something else that would greatly increase variety is an “AI strategy” setting when starting a game. This would allow you to choose broad strategies such as “Rush”, “Boom” etc, and perhaps even more specific such as “Knight rush”, “Feudal rush” etc, and of course “Random”.

And have this be a customizable setting for each individual AI player, rather than globally for all of them. And, of course, an overall “Random” choice to choose for all AI at once so you don’t know which color has what strategy. I’d probably always choose this global Random setting for the match, to be honest.

@FirmSaturn67352

Lastly, I always missed the ability to choose a different AI difficulty on different players, so you can add both a Hard AI and a Moderate AI in the same game, making it easier to balance the difficulty, especially in team games. However, I’m guessing it’s a lot of work to implement, so I would prioritize the two things above first, although this would of course give us much added flexibility as well.

In AoE2:HD, my way around it is to assign “Original AI” to an AI or two, which makes them a little easier, and “AI” to another one which makes a tougher opponent, relatively speaking. And then combining these selections with the overall match difficulty of “Easy/Normal/Hard/Hardest”, as well as different team combinations (i.e., 2 easy AI on one team, 1 hard one on another), and you can get some variable skill levels and matches going.

For AoE2:DE, I think they retained the “Original AI” and “HD AI”, but now also have a “DE AI,” if I’m not mistaken(?). So I think there are more options now than HD to configure your matches with, if so.

That said, my advice seems a bit like a workaround to the problem. Having individual AI personalities and difficulties in a vast pool for the algorithm to pull from would be great; whether it be pulled from in a random fashion, a player-selectable fashion, or both.

[Chesqin]

I just want to say thank you for the continued feedback, we’re still working on the AI and player feedback is a great help!

-Promiskuitiv

[ArshiaAghaei]

@Darkness01101

“DE AI,”

Just called AI. for HD and CD names are AI (HD) and AI(CD)

[WoeIsToWho]

Was there an update to the AI in the past three days? Taking two extreme AI’s at once suddenly got very hard, seemingly overnight. I’m liking the challenge and this is not a complaint, but I didn’t notice any patches or updates so I’m wondering if I’m imagining that spike or if something was changed very recently. Thanks.

Also, and this is a completely unrelated note. Could the BBC basic aggro rules edited to prioritize buildings first (as trebs do) so that they don’t lock onto random military units in bases that the non-siege you’ve brought are there to deal with and waste shots? Especially bad in the moments after a TC or a castle’s garrison is released and they all lock onto squabbling villagers with nary a chance to do anything meaningful with the next volley, let alone connect.

BBC’s can be microed to handle onagers, trebs, and the like, and that’s largely the only time anyone wants them fighting anything but the buildings when buildings are in range. Please, please, please make it so.

[breeminator]

My thoughts on the DE AI, as quite a bad player:

I find limiting the number of villagers to be a poor way of making the AI worse on easier levels. I think it would be much more realistic to slow down the rate of performing actions rather than limiting the number of villagers, as I feel this is a big part of what makes lower level humans worse than higher level humans. I’ve watched low ELO games on YouTube, say ~600 ELO, and nobody ever limits their villager numbers in this way. So this makes Easiest and Standard very easy to beat, as they just can’t do much in the long term with so few villagers.

At the same time as being easy to beat due to the villager caps, those two levels play pretty well up to that point. If I play on Standard, my score doesn’t go up all that much faster than the AI’s, until it reaches its villager cap, then I pull away, of course (like I said, I’m bad at this game). So it feels very artificial to play against, as the standard of play up to hitting the villager cap is quite high, but then it becomes terrible in a completely unrealistic way, as no human that is capable of that standard of play would stop making villagers at the AI’s limit.

It also means that going from Standard to Moderate is a huge increase in difficulty because the villager number limit goes from a cripplingly low level to a level that can support a reasonably fast castle and attack with siege etc. I can sometimes beat Moderate, it depends on the map and civs, but on Arena, say, it’s hard for me to match the time at which I get hit by an army of rams and melee units, and from then on it’s just a continuous onslaught. Did I mention that I’m bad at this game?

So I feel it would be more realistic to have all levels have no villager limit, but instead to limit the rate of performing actions, as this is what poor human players do. We make mistakes like getting housed, or forgetting to build a building needed to age up, etc, and we struggle to switch quickly around the map. The AI can micro multiple units like a beast, even on low levels. Ideally, it wouldn’t just be a limit of one action every x seconds, but rather it would limit actions in a way that resembles how a human is limited. I’d have thought it shouldn’t be too hard to extract this information from real online games, it would be possible to calculate average action rates at different ELOs, and also how that varies with factors, e.g. low ELO players might be slower to act in different places on the map, whereas higher ELO players might be better at using control groups to switch, and be less affected by separation between units. Some actions can be performed in rapid succession, e.g. even poor human players can rapidly queue up multiple units to be produced at the same building.