[Anonymous Player]

Has anyone ever attempted to use deep-learning AI technology to create an AI for the game Age of Empires II (AOE2)?

I’m sure many people have seen Google’s AlphaGo program, which used reinforcement learning to beat professional Go players.

This AI was able to come up with strategies that no human player had ever seen before and ended up working out better than expected.

Similarly, DeepMind created an AI for StarCraft 2 that was better than a group of professional players.

It would be amazing to see a similar AI applied to AOE2, where it would be able to play against a professional and come up with strategies that no one has ever seen before.

Is anyone up for the challenge?

[January_6_2021]

Check the aoe2 “siege engineers ” discord.

That’s where you’ll find people working on this

[cloudstrife559]

DeepMind has hundreds of people working on these systems for *years* before they’re any good.

You’d need to find someone with those kinds of resources that is interested in this game.

It’s not something a single dude on Reddit is going to make happen, at least not to the same level of success.

[Hjoerleif]

The ai which won the ai tournament was pretty insane and would ~~probably already match pros~~ (correction: plays at over 1200 at best).

I think it was called Rehoboam or something, although it wasn’t a self-learned AI through any machine learning algorithms or anything like that that I know of

[Parrotparser7]

Would this work for a game with imperfect information?

[TheCulture1707]

Yeah the deeplearning AI’s are different from all the current AOE2 AI; the current AI even Rehoboam just follow scripts (e.g.

if enemy has >10 knights research pikeman, but obv much more complicated).

These deeplearning / neural ones use complicated deep learning techniques, I don’t really know how they work but I’ve read they “train” by playing billions of games against themselves (& using training data) – so when they are playing a game they can check back on the history of games played, and get a probability of what the best likely move would be at that moment.

Like in SC2 it’ll see a lot of zerglings being made so it will build disruptors.

It’ll see e.g. 4 zerglings going to the expansion, but it knows with high probability this is a feint so it’ll move 3/4ths of its force to the main base.

It has weaknesses though where it can get stuck doing a strat that is doomed to fail but it is very powerful combined with an AI’s godlike potential micro.

I think AOE2 would be harder than SC2 partly because there are more civs, but mostly because of the random maps.

I think the SC2 AI could only play on the map it had trained on, it would need all new training to play a different map.

So the AOE2 AI would need weeks of training for just 1 generated map.

Maybe they could make one that works on generated maps, it might not be as godlike as the SC2 AI in that case but it might still be good.

Anyway I’d love to see one in AOE2.

Hell or even the original Starcraft.

I always thought SC2 and some of the other 3D RTS games, they just had worse tactics, worse micro, things tended to just bog down into blobs of units clumping around in 3D RTS’s.

So in most AI games of SC2 you ended up with it just spamming splash damage units trying to get the highest DPS.

But in 2D games you might see a lot more novel strategies. (like in SC2, the AI built mass disruptors which are a fiddly protoss unit not many pros used)

[Author]

Posts