How difficult would it be to improve the play of the bots? It’s so frustrating when a player abandons a game, the bot takes over and makes completely counter productive moves and/or returns troops that are transferred to him.

Making or improving computer-controlled players a very complex problem.

It is technically possible for a mod to allow you to control the moves of a computer-controlled player

Just had a game that was completely winnable but the bot kept advancing in France and cut me off from the opponent.

I had even started on that country bonus first.

There already are much better AI's available (e.g. Wunderwaffe), but the default AI is Prod 2.0 with randomness.

Besides that there is a recent community interest in making some ML models for Wz: https://www.warzone.com/Forum/820207-bot-api-exist

But like Dan said, making any good AI for a environment as complex as WZ is a challenge.

Can’t you just use an AI to make a better AI?

lolz

No, AI-generated code (and generally any "information" they produce) is unreliable.

Another issue is that AI is actually surprisingly good when compared to the average player. WZ Members can check it for themselves by playing full auto-pilot in random QM games.

Making bots even stronger could lead to situations where teams get stronger by having their players booted.

@Edit

Although it depends on the template. AI is fine on vanilla templates but terrible on any non-standard settings: LD, MA, Commanders etc. You say the bot was advancing in France, so you were probably playing some variation of Europe LD.

Edited 4/24/2025 20:10:32

yea i agree with rento, somebody is gonna do it and then the players will get steamrolled in most templates

Another issue is that AI is actually surprisingly good when compared to the average player. WZ Members can check it for themselves by playing full auto-pilot in random QM games.

Did someone ever use autopilot for the ladders before rating system change? How did it perform? Current rating system (for non-seasonal ladders) makes it impossible to know how good players actually are.

Did someone ever use autopilot for the ladders before rating system change? How did it perform? Current rating system (for non-seasonal ladders) makes it impossible to know how good players actually are.

Yes, this was actually done many years ago (on the old / original ladder, 1v1) with someone using autopilot for all games as an experiment. If I remember well, it performed below average, but not horrible.

However, it strongly depends on the template and AI. There was an external Warzone AI competition many years ago as well, with Wunderwaffe and Cowzow among the AI's that performed quite well (I don't remember the specifics and winner, but surely someone still has a link).
Both AI's perform pretty well in specific settings, especially Strat 1v1, but may have errors in special scenarios.

Warzone AI is the more broadly useable, but not that great either. In the AI competition, it had a middling performance, being consistently defeated by better bots like Wunderwaffe and Cowzow. However, this is seen as a feature as mentioned earlier, because you usually don't want a team to benefit from having a teammate booted.

If you get frustrated with this, just compare this to real war: it's not just the strength of your armies that matter, it's also their willingness to fight and follow orders. Getting booted basically means your troops run rampant without central command or coordination with their allies. An excellent player that gets booted easily just isn't that good.

Edited 4/25/2025 08:06:59

The AI challenge was a partnership between Warlight and The AI Games back in 2014. Fun fact: the real time ladder was introduced in this same month. The site has gone offline a few years ago (http://theaigames.com/). Norman competed and coded Wunderwaffe. From what I understood from him back in the day, the challenge was not only coding the AI, but also the pretty extreme limitation of not being allowed to check the history in your code of the game it was playing, as it would ask too many server resources. As this has never changed (as far as I can tell), the real challenge is to write an AI that has to work out a plan based on the current gamestate while not having the previous gamestates/history.

An excellent player that gets booted easily just isn't that good.

I am taking this personally, Math Wolf :p

Edited 4/25/2025 19:50:53