AI difficulty levels

[mimo]

This thread is to start a discussion on the different AI levels to be tuned before A17. So general AI discussions should still go on the other thread http://www.wildfiregames.com/forum/index.php?showtopic=18425

To summarize, we have presently 5 AI levels, with the settings:

- sandbox: AI has a gathering rate factor of 0.5, does not trade/barter and does not attack
- easy: AI has a gathering rate factor of 0.66,does not trade/barter and does not rush
- medium: AI has a gathering rate factor of 1
- hard: AI has a gathering rate factor of 1.33
- very hard: AI has a gathering rate factor of 1.66

but as I've only used medium (either to develop or play), I'd like to know what people think of these different levels, how difficult they are and if we should change some of these settings ?

[wraitii]

I've lost twice to medium. I think you can make medium hard, possibly skip current "hard" and keep very hard as it is. Possibly give them the HP buff we talked about. Sandbox and easy should be fine, but I haven't played them.

[meap]

I played a couple of times with Petra Bot in very hard mode but with a pop limit of 150 (to prevent perf problems). I find this mode very challenging (but not impossible), far more than with Aegis (of course).

The problem is that I can't finish the game as it begins to lag way too much when I start attacking the AI (each turn it freezes for ~10sec and let me play for about 1sec), but this is another problem...

So mode "very hard" can stay like this IMHO.

[WhiteTreePaladin]

I generally play with all the standard options (basic starting resources, 300 pop) on very hard. One AI opponent, random civs, and usually a random two or three player map. I can generally beat the very hard difficultly if the map favors defensive play (i.e. narrow channels, walls, etc.) or if the map has sparser resources because the AI doesn't know how to manage low resources well. On open maps with lots of resources the AI is very tough indeed and I pretty much always lose. I would prefer to keep the top difficulty as difficult as possible though.

What I would like to see regarding difficulty is better battle tactics and strategy. The AI's resource gathering ability is already fantastic although it does struggle when there few resources. As far as levels, I would prefer to keep the five levels. I guess the main thing I would like to keep is a really hard level that doesn't cheat. I think the hardest level could allow cheating, but wouldn't want to have to play a cheating AI just to have access to better tactics.

A death match mode where the AI mostly forgets about resource gathering would be fun and allow fine tuning the battle tactics / strategies.

Edited July 4, 2014 by WhiteTreePaladin

[leper]

I mostly play against at least one very hard AI, sometimes against a medium AI because I either forgot to change it or because I want a quick game, and I think having more than one setting where the AI "cheats" (resource bonus) is better than just having one of those cases.

I'd oppose an HP buff for the AI.

[WhiteTreePaladin]

Generally the AI is already very good at collecting resources so I don't think a resource cheat will make it too much harder. It is poor at tactics, so the HP increase would help there. Personally though, I really don't like the idea of an HP increase and would really prefer it to use proper tactics and strategy. That's very difficult to code though.

[mimo]

Thanks for all the answers:

I fully agree that improving tactics is the way to go, but it's a long term development (this thread is rather to tune the levels for next alpha) and it would help to have more people working on it.

Otherwise, for the short term, I conclude that the hard levels are more or less ok, possibly improving the AI attack management if possible before the release.

But I'd like to know also how beginners feel about the easy level ? is it fine, or should it be made a bit easier (e.g. smaller attack size, ...)

meap, when you play with max pop = 150, have you already faced a "huge attack" from the AI ? huge is more than 100 units attacking you ? I realise now that this size should be made dependent on the max population.

[meap]

meap, when you play with max pop = 150, have you already faced a "huge attack" from the AI ? huge is more than 100 units attacking you ? I realise now that this size should be made dependent on the max population.

Difficult to answer. Pretty hard to count the number of units attacking me ;-) I would say they were certainly more than 50, maybe 100 but not more.

BTW, I played recently and got endless javascript warnings saying "path toBeContinued" after a couple of attacks from AI, maybe after I finished building the walls protecting my territory. The game then started to lag a lot.

(Should probably move this to another topic though).

[agentx]

I'm also trying to make this ladder useful to beginner. I like no attacks for sandbox and very hard as unleash everything. May I suggest no technology for easy? So:

 - sandbox:   no attacks, no tech - easy:      no tech, simple economy, no rush - medium:    advanced economy (, no barter?) - hard:      advanced combat (, no heros?) - very hard: no limit

Still lacking a useful criteria for hard...

Btw, am I right the AI settings dialog could be a mod and the selected settings just go straight into bot constructor? Spending extra options for AIs without compilation?

[WhiteTreePaladin]

I would say medium, hard, and very hard should use all features (barter, heroes, etc). It would be dull to have to up the difficulty just to see enemy traders or heroes. I wouldn't want the core experience to change as the difficulty rises. I would prefer the AI to expand territory faster, gather resources quicker, and use much more advanced battle tactics and strategy as the difficulty goes from medium to very hard.

[auron2401]

I think we should have difficulty set up something like This.

Military | Sandbox Easy Medium Hard

Economy

_______

Sandbox

Easy

Medium

Hard

Criteria

You would just tick whatever you want your AI to do....

EG sandbox economy / hard millitary , or medium/medium.

Military

• Sandbox: Will defend, but not attack
• Easy: Will Defend and attack
• Medium: Will Raid.
• Hard: Will use champion units and healers in combat.

Economy

• Sandbox: Will gather all resources inside of territory, (and a little bit out of). No farming.
• Easy: Will Farm, and push outwards to gather. Will only expand when out of resources
• Medium: Will use rudimentary technology an expand Whenever the opportunity presents itself
• Hard Will use advanced Technologies

Suffice to say it will of course get better at doing all the previous tasks in each tier.

Edited July 14, 2014 by auron2401

[wraitii]

My personal preference for AI behavior would also have two criteria, but those would be "difficulty" and "personality". Difficulty would basically make the AI use less varied, less efficient strategies. Personality would range from "efficient" to "plays for fun/excentric". "Efficient" AIs would try to win the game as quickly and as ruthlessly as they can. "Play for fun" AIs would just play, attacking once they've amassed a formidable army or things like that, use a wide variety of strategy (some not necessarily really efficient). Basically they'd be in control, so if you attacked them they'd retaliate and stuff, but wouldn't necessarily try to beat you as quickly as possible.

The "efficient" AI would simulate a "real" MP game between players that want to win, the "excentric" one more of an MP game between two laid back players.

But that requires a much more efficient AI than what we now have. For now, I agree with WhiteTreePaladin.

[auron2401]

Fair point.

[raymond]

Some modifications:

- sandbox: AI has a gathering rate factor of 0.33, does not trade/barter and does not attack, no healer
- easy: AI has a gathering rate factor of 0.66, does not trade/barter and does not rush, no healer
- medium: AI has a gathering rate factor of 1, does trade/barter, does not build fortress/defence tower/walls, no healer
- hard: AI has a gathering rate factor of 1.33, does trade/barter, build fortress, does not build defence tower and walls, healer
- very hard: AI has a gathering rate factor of 1.66, does trade/barter, build fortress/defence tower/walls, healer

Edited July 15, 2014 by raymond

[agentx]

I would say medium, hard, and very hard should use all features (barter, heroes, etc). It would be dull to have to up the difficulty just to see enemy traders or heroes. I wouldn't want the core experience to change as the difficulty rises. I would prefer the AI to expand territory faster, gather resources quicker, and use much more advanced battle tactics and strategy as the difficulty goes from medium to very hard.

It depends on the user model. I assumed users advance by adding features to their repertoire, you seem to assume users start with the full set of features and advance by improving. The reality is probably in between and different for each user. But I think it is easier to reason about difficulties if basically it is about switching features on/off and not moving along a scale.

[mimo]

BTW, I played recently and got endless javascript warnings saying "path toBeContinued" after a couple of attacks from AI, maybe after I finished building the walls protecting my territory. The game then started to lag a lot.

(Should probably move this to another topic though).

The best is to put somewhere the commands.txt file and to note the revision version you've used. You can either put these infos here in the Bug reports section of the forums, or create a new ticket on trac.

[WhiteTreePaladin]

It depends on the user model. I assumed users advance by adding features to their repertoire, you seem to assume users start with the full set of features and advance by improving. The reality is probably in between and different for each user. But I think it is easier to reason about difficulties if basically it is about switching features on/off and not moving along a scale.

Well, there is easy for that. I don't disagree, but I think traders, heroes, etc. shouldn't be reserved for the highest difficulties.

My personal preference for AI behavior would also have two criteria, but those would be "difficulty" and "personality". Difficulty would basically make the AI use less varied, less efficient strategies. Personality would range from "efficient" to "plays for fun/excentric". "Efficient" AIs would try to win the game as quickly and as ruthlessly as they can. "Play for fun" AIs would just play, attacking once they've amassed a formidable army or things like that, use a wide variety of strategy (some not necessarily really efficient). Basically they'd be in control, so if you attacked them they'd retaliate and stuff, but wouldn't necessarily try to beat you as quickly as possible.

The "efficient" AI would simulate a "real" MP game between players that want to win, the "excentric" one more of an MP game between two laid back players.

But that requires a much more efficient AI than what we now have. For now, I agree with WhiteTreePaladin.

I agree completely. A "laid back" AI that focues more on defense (with walls) and masses of units that a good AI would not horde all at once, would be awesome to play.

Edited July 16, 2014 by WhiteTreePaladin

[agentx]

> but I think traders, heroes, etc. shouldn't be reserved for the highest difficulties.

Good point. There is a another catch: Assume multiple bots, how do you compare them at same difficulty, if the criteria behind are more or less subjectively chosen by the dev?

[meap]

I think traders, heroes, etc. shouldn't be reserved for the highest difficulties.

Agree with that too.

BTW, one thing I would do at hard level (compared to medium) is to remove the 'grace period' at the beginning of the game. This would allow the AI to attack before walls and towers are built and making it less predictible. This seems to me pretty easy to implement.

I would allow the AI to 'cheat' only at the very hard level.

Some modifications:

- sandbox: AI has a gathering rate factor of 0.33, does not trade/barter and does not attack, no healer

- easy: AI has a gathering rate factor of 0.66, does not trade/barter and does not rush, no healer

- medium: AI has a gathering rate factor of 1, does trade/barter, does not build fortress/defence tower/walls, no healer

- hard: AI has a gathering rate factor of 1.33, does trade/barter, build fortress, does not build defence tower and walls, healer

- very hard: AI has a gathering rate factor of 1.66, does trade/barter, build fortress/defence tower/walls, healer

Mmm, if the AI can't build fortress (and thus can't build units effective against walls), that means you are almost invincible unless you play at the highest difficulties.

And playing at hard level without the AI building defense towers and walls would be far too easy...

[WhiteTreePaladin]

If we allow the AI to cheat, I would like that to be it's own difficulty. In other words, if "very hard" cheats, than "hard" should be the same, just without the cheating. The cheating advantages should include Line of Sight visibility in addition to resources. Right now, I think all AI's have full LOS, but that will change eventually.

Edited July 19, 2014 by WhiteTreePaladin

[meap]

If we allow the AI to cheat, I would like that to be it's own difficulty. In other words, if "very hard" cheats, than "hard" should be the same, just without the cheating.

I completely agree with that, i.e. very hard = hard + cheating. This allows the player to explicit agree with giving the AI an "unfair" advantage.

[agentx]

I've been looking for opposite pairs possibly describing a bot personality like:

• Fortification - Spreading/Claiming
• Defensive - Attacking
• Exploitation - Barter

Which gives a nice 3D space a bot can choose of. However the pairs are not exclusive, e,g, exploitation includes claiming. Also it depends on the map, which brings me to cheating: Humans memorize maps. LOS is most interesting on the first play, from there it is getting more and more relative. After the fifth replay you know form where the attacks are coming and build towers likewise.

Currently bots can't remember maps. They could however include pre-compiled information about maps/scenarios. If any of the devs might fancy an "online brain" for bots I'm all ears, all needed is a kinda XMLHttpRequest object supporting GET/POST/HEAD. Maps is only one thing, even more interesting is a database of players and their strategies. Who wants to play the same game twice?

[Stan`]

Well we could use user feedback for that. But would that mean you need a internet connection to play (I'm thinking of lan parties)

[raymond]

In A17:

I played against easy bot (Petra bot) with my friend against two other players: lost. Against one player: win, but it was hard.

The AI build civic centers to the enemy and build defence towers on the border of the own and the enemy territory. It is very aggressive and should be only on hard or very hard.

The easy AI are also possible to walk around the defence towers, not the direct path from their village. It is possible that easy AI and medium only attack in the direct path, so it is better to defend. Normal, hard and harder AI should able to use indirect paths, so walk around defence towers or fortresses? Only normal, hard and harder AI should build other civic centers.

It is possible to lower difficulty in general?

Edited October 27, 2014 by raymond

[mimo]

Adding a very-easy level which would have the same economic handicap as the sandbox, but with the ability to attack, would be a possibility for A18.

But, have you tried this sandbox level on A17: it should help improving your economic management and attack strategies as you don't have to care for an attack from the AI.