SMAC 444 (AI experiment)

[Lord Avalon]

It'll be in the GOG install.

[gwillybj]

I'd never heard of MAX but found it on gog.com, thanks for the suggestion. I also have Civ-Evo which I really like, has an atmosphere of sorts, gotta put time into these games now D:

You'll need DOSBox if your OS is Win98 or newer.

[MegaTurtleRex]

No, I am unable to make "really good AI", to be honest. "Less stupid things" is what I do, as good AI decisions as possible on the little space I find. MoM is a great game, one of the best in TB genre, but very old. The AI used to be his mega weakness, now it's far better, almost unplayable on the highest difficulty.

If you search a very good AI in turn based game, I have been told several times that Mechanized Assault Exploration (M.A.X.) can really turn you inside out. C-evo is a civ-like game, which has a very good AI (but otherwise it's conceptually a bit dry) - an AI professor told me this.

I'd never heard of MAX but found it on gog.com, thanks for the suggestion. I also have Civ-Evo which I really like, has an atmosphere of sorts, gotta put time into these games now D:

[kyrub]

I've not played multiplayer, but I beat Transcend without cheese (only a few crawlers/no clear AI exploits/etc) basically every game, and am glad you're working on improving the AI. I have only played a few games with the previous patch (SMAX only)

SMAX patch has almost NO AI improvement. Just the terraforming basics. Nothing from all the other, important stuff. So you have seen nothing.

1. Not attacking with hurry
I find the AI units never (in my experience) attack with the hurry penalty. This means that, for example, you can move your army to two squares from an AI base defended by a few Impact infantry without armor and a plasma defender, wait a turn, then move in (accepting a -33% hurry on non-elite infantry) and kill far more than they would've lost by moving out to attack you with their weapons.

Interesting. Not sure I can solve it, but I can try. AI unit movement fonction is incredibly complex.

2. Tendency to stack very badly, especially at chokepoints
Harder to fix, but this and the damage to all in stack mechanics has allowed me to defend much larger attack forces with only a few rovers. -- This issue also makes AI probe teams less threatening, since they tend to all move out of a base at once, and often are all stacked. Easy pickings.

Hard to change, I guess.

3. Lack of co-ordination of air and ground/sea forces

There could be a few small improvements here, which at the end could give us a visible improvement. No air focus on formers or empty transports - it seems ridiculous but these things can really help if used in large quantities.

Using clean reactors and rehomeing to a punishment sphere/specialist base to avoid FM drones would be amazing if you could teach them it, but perhaps too complex.

That is on thhe simple side.

[ete]

I've not played multiplayer, but I beat Transcend without cheese (only a few crawlers/no clear AI exploits/etc) basically every game, and am glad you're working on improving the AI. I have only played a few games with the previous patch (SMAX only, it's got more stuff and everything from SMAC is there anyway, plus I had issues getting it SMAC to run with the patch), and have noticed a bit more challenge, but the AI still can't stand up after the earlygame despite building fairly impressive empires. Here are few key areas which really bring them down along with some ideas about improving them.

Combat

1. Not attacking with hurry
I find the AI uses units extremely inefficiently in many situations, mostly because they are too cautious to attack and never (in my experience) attack with the hurry penalty. This means that, for example, you can move your army to two squares from an AI base defended by a few Impact infantry without armor and a plasma defender, wait a turn, then move in (accepting a -33% hurry on non-elite infantry) and kill far more than they would've lost by moving out to attack you with their weapons.

I'm not sure whether you've looked into the code which controls the decision to attack or not, but something like: If target_unit_cost > own_unit_cost and probability_of_winning_battle > 50% then Attack despite hurry could help greatly. A more advanced version which takes into account the stacking of the targeted unit, perhaps just a multiplier for the number of units stacked there?

2. Tendency to stack very badly, especially at chokepoints
Harder to fix, but this and the damage to all in stack mechanics has allowed me to defend much larger attack forces with only a few rovers. When there are foe units within range, the AI should not have huge stacks of 6-8 units. If possible, perhaps adding in a check for nearby enemy units, and encouraging units to move to spaces with 2 or less allied units so long as they're not able to kill the foe units would improve this?

This issue also makes AI probe teams less threatening, since they tend to all move out of a base at once, and often are all stacked. Easy pickings.

3. Lack of co-ordination of air and ground/sea forces
Probably far too complex/would have negative spin offs, but I find that foe air power often comes in and kills some random units (police defenders, formers), then dies to air superiority counters. If the AI could start using drop pods effectively along with that kind of harassment, they'd be a lot more of a nuisance. And similarly, if they had more of a tendency to focus air power on bases which they had a land unit capable of capturing that'd be excellent.

This is the least specific, but I think possibly a large part of the reason that the AI ceases to provide any real threat after airpower in my games, they don't get how to use that kind of army very well.


I see you've already made a change to the main production gripe I have (far too many transports) and are looking into/working on the main base management issue (handling FM badly and having infinite pacifism drones), when there's a SMAX version I'll do some playtesting and see what else stands out. Using clean reactors and rehomeing to a punishment sphere/specialist base to avoid FM drones would be amazing if you could teach them it, but perhaps too complex.

[kyrub]

SMAC_444 version l

New version. a test build.

http://alphacentauri2.info/index.php?action=downloads;sa=view;down=122

This is a test build of the patch. Be prepared to see some unwanted AI actions, improving requires a lot of patience and a good perception of AI activity throughout the game (any thorough feedback is appreciated, please react in the appropriate thread in the modding forum).

I have applied a few AI Social engineering choices, mainly the support / police mismatch in the original game.
Also the morale setting is now less valued by the AI.

I included a few important and obvious corrections to previous patch, we should see less probes again, less sea transports etc.



a uterran4 (terraforming basic)
b tech_beelines
c AI hurrying
d close base and fungus love from 70
e unit_abilities_8
f more ships
g formers chance
h crawlers
i recycle tanks priority
j speeders inclination

-------------------------
k1 stockpile energy bug
k2 swap base bug
k3 children creche bug
k4 AI hurrybug 3
k5 AI seaformers
k6 colony pods
k7 reversal of near AI cities

----------------------------------------
l8 less probes by
l9 if police =< -4 less ships
l10 less naval tranports
l11 AI socio version 3

[kyrub]

I tend to see the opposite in strategy forums, from Total War to Warband everyone complains that A.I. is weak. The fixes for all these games seem to make A.I. do less stupid things but never play clever or sensibly. Then again I have never played that Master of Magic game you mentioned so I would be curious to try it. I've always assumed it was impossible to make really good A.I. with current technology but I have never really understood how it actually works.

No, I am unable to make "really good AI", to be honest. "Less stupid things" is what I do, as good AI decisions as possible on the little space I find. MoM is a great game, one of the best in TB genre, but very old. The AI used to be his mega weakness, now it's far better, almost unplayable on the highest difficulty.

If you search a very good AI in turn based game, I have been told several times that Mechanized Assault Exploration (M.A.X.) can really turn you inside out. C-evo is a civ-like game, which has a very good AI (but otherwise it's conceptually a bit dry) - an AI professor told me this.

[MegaTurtleRex]

I find it odd that people seem to care less about the AI enhancement. Everybody is gung ho about new modding variants and "tons of" new content, usually. When it comes to: "Does the game represent bigger challenge for me?", silence.

I have had even negative reactions to Master of Magic patch, that clearly raises the game level of AI, some people dislike the extra challenge and they are reluctant to turn down the game difficulty level ("come on, I have always beaten it on impossible"). We laugh at bad AIs, but in fact, we need them to feel like the big winners at the end.
Really strange.

I tend to see the opposite in strategy forums, from Total War to Warband everyone complains that A.I. is weak. The fixes for all these games seem to make A.I. do less stupid things but never play clever or sensibly. Then again I have never played that Master of Magic game you mentioned so I would be curious to try it. I've always assumed it was impossible to make really good A.I. with current technology but I have never really understood how it actually works.

[kyrub]

I'm thinking that each city might have one or two crawlers, not dozens.

The AI actually has an internal cap on crawlers, max 1 per city. This would be doable and it's a very simple and good idea, IMO. But I am not specialized in modding. Go to ac2 forum and ask Yitzi, he could help you.

[TarMinyatur]

Crawlers

Great. No more crawlers at this point.

...My general strategy = lots of nudges to the AI leading to a visible overall improvement.

HELP list (for those who wish to contribute)
----------
a) Advance terraforming choice algorithm
b) faction social engineering, short list with ideas for each faction (early, midgame, late game, War options)

The game is more challenging without crawlers. But I do like the idea of convoying resources...

Could each crawler require a specialist citizen ("convoy engineer")?

If there's a great tile outside of my city, I think I should be able to harvest it if I make appropriate investments in equipment and labor.

I'm thinking that each city might have one or two crawlers, not dozens.

This might be really tough to code...just an idea. I'd love to see Yang come into my territory with an escorted crawler and begin harvesting "my" farms for the good of his collective.

[Kirov]

I find it odd that people seem to care less about the AI enhancement. Everybody is gung ho about new modding variants and "tons of" new content, usually. When it comes to: "Does the game represent bigger challenge for me?", silence.

I hear you because for me it's the contrary. I stopped playing SMAC somewhere in 2007, when everybody move to Civ4 and I was soo bored with AI. You think for hours about how to give the AI something that pretends to be a sliver of chance just for a couple of turns. One city challenge, self-imposed limitations, pre-made scenarios, switch side challenges... Nothing really works, AI is always bound to lose. Well, of course you can't really win the switch side challenge, but when you play it and you switch for the second time like in 2200 to Hive or Miriam and they have 1 (!) base, the other colony pod is stuck somewhere helplessly in fungus, no Centauri Ecology is researched, the HQ support is stuck by several laser squad units... It makes me cry.

[kyrub]

I find it odd that people seem to care less about the AI enhancement. Everybody is gung ho about new modding variants and "tons of" new content, usually. When it comes to: "Does the game represent bigger challenge for me?", silence.

I have had even negative reactions to Master of Magic patch, that clearly raises the game level of AI, some people dislike the extra challenge and they are reluctant to turn down the game difficulty level ("come on, I have always beaten it on impossible"). We laugh at bad AIs, but in fact, we need them to feel like the big winners at the end.
Really strange.

[MegaTurtleRex]

I suppose though, a bit of nerve stapling might be ok

[Kirov]

Kind of unfair to expect just one person to work on a game if there's not much interest. On that note though, someone has added your patch kyrub to theirs over here alphacentauri2.info/index.php?action=downloads;sa=view;down=121

Their forums seem active over the game too alphacentauri2.info/index.php?board=13.0

It would be unfair, which is why I specified that there is pretty big interest (well, as big as you can get for SMAC in 2013) over ac2, along with help from other players and modders.

[MegaTurtleRex]

Kind of unfair to expect just one person to work on a game if there's not much interest. On that note though, someone has added your patch kyrub to theirs over here alphacentauri2.info/index.php?action=downloads;sa=view;down=121

Their forums seem active over the game too alphacentauri2.info/index.php?board=13.0