AI building vast numbers of Baneblades, Obelisks

[Smudge]

I played a game as Guard against 4 AI all at strong level with game pace the fastest possible. It got to about turn 110 and I got hit by one or two Necron Obelisks and after concentrating all fire from 8 howitzers and about 5 Leman Russ I managed to kill one and drive the other off. Next turn the AI Guard player sent in three Baneblades. At that point I decided I was going to lose so I opened up the debug window so I could see the rest of the map.

The Guard AI had built 24 Baneblades and about 20 other units, mostly Leman Russ. He was in a three way scrap with the AI Necrons who had 18 Obelisks + stuff and Marine AI who had at least 20 Stormraven units.

I checked the Guard main city and it seemed to have the capacity to build one Baneblade per turn so it did. The way each of the AI had gotten to this vast production capacity seems to have been by spamming loyalty buildings - Guard had 12, Marines had 8 and the Necrons had 15!

Anybody tried that tactic themselves?

[Ail]

Yes... The AI basically inherited the build-orders that I have come up with and taught them to play aswell.

Because the default game-mode is FFA, those buildorders are very lategame focused and usually have a relatively bad earlygame, that can be punished by all sorts of all-ins.

For example the AI will try to keep their pop-growth maximized all of the time. Which early-on means you have to pay the upkeep in both loyalty and food/ore/requisitions for population that will stay useless until you can start to catch up with buildings after constructing a second construction-building. That is a huge economical sacrifice which basically trades away any early-game strength for being stronger later on.

When you play a more balanced build, that gives you a stronger early-/mid-game at the expense of running into the population-cap or at least severely slowing down your population-growth but don't punish their greedy builds, they will have a huge economical advantage in the lategame.

In faster-speed-settings this should be even more severe as there's less time to punish them.

The "loyalty-spam" is simply a matter of math. You can calculate at which point a loyalty-building is better than other buildings. And since computers are good at calculating they do just that. I don't calculate like that but I know that I should start pumping them out at around 12 normal buildings and add a few buildings here and there so that at about 30 buildings in total I shall be at the loyalty cap of +50.

I wonder if the AI should be a little more adaptive with it's build-orders depending on the settings of the game. In a 1v1 on a tiny map you most likely don't want to be at 19/24(9) pop in the hopes of a strong lategame that will never come.

Also note that the approach of the AI works much better on higher difficulty-levels. The reason is that pop-growth doesn't scale with loyalty but both production and income do. So they have a much easier time to actually put their population to use.

[dexgattaca]

Thank you for the feedback Ail!

It would be nice to see multiple strategies. Such as early, mid and late game pushes from the AI.

[Ail]

I have made a small change in my local version to be less population-growth-focused early on, as it really led to some weird builds with more weaknesses than is justified by the small advantage it would gain later on.

I'm also testing another change that hopefully leads to more build-diversity. But all-ins is not really something that I think I want to teach to the AI. However, you may or may not know that ctrl+t or something like that brings up the endgame-scoreboard while the game is still running. And you can get some helpful information from it. The AI might want to look at it aswell and draw some conclusions from it.

[Total Oh No]

The game has some issues in late game if you ask me because units are extremely spammable. I don't think you should play on the fastest game setting because that will just get you into the ridiculous unit spam levels earlier than normal. And yeah loyalty buildings are very important because at 50 loyalty they double your entire resource and unit production in the city.

By the way, even if you think you've already lost, it might still be possible to win because the AI might be great at building units, but it's not that great at using them yet. Here's a screenshot from the scores screen of a game that I totally thought I lost all the way at the beginning already, but I continued to play and eventually managed to destroy more and more of the enemy units until I overpowered him:
https://cdn.discordapp.com/attachments/467032336303652875/509854300059729930/Gladius051.jpg
Of course that was at a normal game speed game though. One on the fastest might be harder to come back from because the enemy will be able to build multiple units per turn. (Even at normal speed they were able to build like 3 units every 2 turns.)

Originally posted by Ail:

The AI might want to look at it aswell and draw some conclusions from it.

Hm, I dunno if you'd really want the AI to look at the ctrl-T scores screen because it's kiiinda a cheat (you can't do it in multiplayer games). But I'm not exactly someone who's against cheating AIs though. As long as they're a good challenge, I don't care that much.

[Smudge]

I was using spamming in the descriptive sense rather than perjorative. I just played another game as Guard and kept my loyalty scores high and was able to get up to Baneblade-production tech very quickly. then had a great, really fun, tight game against swarms of Orc Squiggoth things and an enemy Guard player. Killed the Orcs eventually and then it came down to grinding down the Guard Baneblades faster than they could build them. The AI didn't build aircraft and I found my aircraft being able to nip in and take a bite out of a baneblade, nip out again and go and get repaired was probably the game changer. That and the 10 or so late game Howitzer tanks I backed up my Baneblade wall with.

Still having some of the best fun I have had with a turn-based (my preference) game!