BTS: Changes to Difficulty and AI (Highlights)

[player1]

Originally posted by Flash
7EA,
Theoretically maybe the AI doesn't cheat. But on the first turn and every time you reload your game the AI is allowed to have an extra turn to rush buy things as long as funds are available. I'm not sure if this is good or bad.

I stoped playing GalCiv2 with patch 1.3.
Are you saying this bug is still in?!?

Very sloppy, if true...


EDIT:
Whatever those above said...

[joncnunn]

That would be normal, even on the offical scenerios it's the same AI optimized for the normal game; not the scenerios.

Originally posted by 7EA
I have been playing the 'Final Frontier' mod and I find myself building lots of carriers for my squadrons because of the ability of squadrons to kill (not just damage) enemy units. It doesn't seem that the AI "knows" how important the squadrons (and Anti-ship missiles) are. I haven't seen the AI build any squadrons yet and it doesn't seem to use its anti-ship missiles. I've been playing for more than 200 turns (on emperor).
Needless to say the AI gets its butt kicked by my lightly defended carrier fleets (and a couple of destroyers).

So I guess the AI hasn't been optimised for the various mods.

Are my observations correct or was it just a coincidence in that particular game?

[Wiglaf]

That is amazingly sloppy then.

[joncnunn]

I wouldn't go that far, I think we're only 5 to 10 years away from being able to in a game with the complexity livel of the Civ 4 series having a no bonsus / no disadvantage AI be at the comparable level as a human.

Oh, and with a single cheat not included, the AI would already be there. (If the AI could perfectly predict the next random number in the sequence and make war accordingly, it's attacks would be virtually unbeatable.) This of course is the mirror image to the human cheat of turning off "preserve random seed " and consantly reloading bad attack outcomes.

Originally posted by snoopy369

Humans are always smarter than the AI, so the game would be pathetically easy I don't care how well programmed the AI could be, it's never going to adapt to changing circumstances as well as a human. Not in the foreseeable future, anyway.

[joncnunn]

Closest to even, yes; subantly more positive than negative though.

(Ultra cheap unit upgrades, better odds against barbs.)

Originally posted by snoopy369
Indeed, the 'noble' level is pretty much no AI bonuses (it has some bonuses and some negatives, but it's roughly even).

[joncnunn]

Re: Auto-Workers and Auto-Searchers

Those features were already in Blake's "Better AI" mod for warlords.

It sounds to me like that warrior was assigned to fog busting; once the border expands twince, there is no longer fog forming too close to the cities.

Yes, moderately damaged scouting units are generally better off continuing to explore as long as there's nearby unexplored territory. It's only when they get heavily damaged that they need to rest.
This is because I would glady trade a scout reaching a goody hui before the opponents in exchange to losing it. Most of the postive results to a single goody hut poping more than make up for the loss of a scout or two.

Originally posted by Chopper
Blake, AI was too good for me before, now its even too good for all the family.

I have a question re auto worker and scouts.

It seems we have a schizo-situ where until second culture border pop on a Creative civ, warrior will alternate between two squares (parrallel with border) outside borders for up to 14 turns, or that was until the border pop happened. Are they too afraid to go too far from culture?

Also, I have posted under the CyberShy thread, I have had mine/cottage swapsies with workers like the old farm/cottage swapsie. Something to look into for the auto peeps.

Also scouts and warriors love to search with less than full health whereas in Warlords and Vanilla they would heal before moving if you selected search after the battle. Is this a new feature?

[Blake]

Ummmm when a computer can compete with a human really depends on how smart the human is.

Many players play at Warlord level, granted they probably usually win at warlord level which means it's below their skill level but it means that the AI is basically playing at human level.

Making an AI which can challenge an expert human without cheating is a whole 'nother ballgame, this is because expert humans constantly improve their understanding of the game and the opponents.

In CIV the AI is deliberately handicapped by having a personality, in some cases this is a handicap against simple efficiency (ie preferred terrain improvements and technologies), in other cases the predictability is a weakness, you can almost guarantee that Gandhi wont invade you, you can almost guarantee that Monty will attack someone.

This deliberate handicapping is one reason why Noble isn't really that unfair - the AI is forbidden from adapting to the maximum possible extent, the human is under no such constraint.

I did weaken the personalities somewhat in BTS, in particular when it directly impacted efficiency (ie preferred terrain improvements) but the AI personalities still make it easy to play the AI's.

Also even turning on randomized personalities isn't a perfect solution, for one it's possible with some deduction to figure out what personality an AI has - for example if the AI wont open borders it's tokugawa, if another AI will open borders with that AI then it has one of the warmonger personalities... when you observe how quickly relationship bonuses (like shared religion grow) you can quickly deduce which personality an AI is using.
And asides from that, there is the possibility of a personality and civ being badly mismatched, although civs are pretty flexible so this isn't a huge deal.

[tsubame]

Perhaps in future, there should be three sliders for the AI

1) Personality- How much should the AI pre-scripted personalities affect things

2) Cheat level- how much resource bonus the AI gets

3) Ruthlessness- how well the AI thinks...

Have a general slider as well for those who don't want that detail.

[JimMac]

What I want from a CiV AI is challenge and 'fun' ; and in the 2 epic BtS games I have played so far I have got that.

However, if I had to pick between a ruthless A.I that always picked the optimum strategy and tactics, and one which wasn't as efficient since it was behaving in line with its personality, I'd go with the personality A.I.

Pretty good job Blake.

[Blake]

For sliders, I'd probably go with about 3 settings:

Opportunistic - At 0% the AI makes decisions nearly entirely based on personality, at 100% they only consider benefits.
Pacifist<->Militaristic - At 0% the AI puts little weight on military, at 100% they maintain the largest possible military which wont crush their economy.
Predictable<->Erratic - At 0% the AI is highly predictable, at 100% it makes major decisions almost at random.

[tsubame]

I like that system, if it allowed the current difficulty level for bonuses/penalties, and it had a system for AI plays under same rules as human.

[mjs0]

Rather than removing the personalities perhaps stronger and more varied traits that are in synergy with the personalities would be a nice option. So, the personalities for each civ/leader are actually logical within the context of the game rather than just an attempt to force a historical attitude on the leader.
Perhaps this would encourage humans to play like the AI

[LordGek]

Originally posted by Blake
For sliders, I'd probably go with about 3 settings:

Opportunistic - At 0% the AI makes decisions nearly entirely based on personality, at 100% they only consider benefits.
Pacifist<->Militaristic - At 0% the AI puts little weight on military, at 100% they maintain the largest possible military which wont crush their economy.
Predictable<->Erratic - At 0% the AI is highly predictable, at 100% it makes major decisions almost at random.

So idle speculation or something you're looking into? I for one think sliders like this would be AWESOME (especially if there were a "all random" option...although I guess it should be wieghted a bit toward the center of each slider).

[tsubame]

I like this feature- how I would use it.

80% Opportunistic. Allow for personality occasionally, otherwise the AI's would all act alike

10% Randomness- to keep the AI from being too predictable

Random Pacifism for each AI

0 Resource Bonus/Penalty- I want the AI to play under the same rules I do. That's always a sore sticky point with me.

[JimMac]

Originally posted by tsubame
I like this feature- how I would use it.

80% Opportunistic. Allow for personality occasionally, otherwise the AI's would all act alike

10% Randomness- to keep the AI from being too predictable

Random Pacifism for each AI

0 Resource Bonus/Penalty- I want the AI to play under the same rules I do. That's always a sore sticky point with me.

The advantage of such sliders would the AI could meet different preferences of players. For example, my preferences are very different to Tsubame; I would go 20% Opportunistic, as I really enjoy the appearance of the different A.I. personalities.

And 'shock' I really don't care that much what bonus the A.I gets! Providing I get a fun and challenging game, it doesnt bother me about what goes on 'under the hood.'