The search for AI

[Avenoct]

While I think we've brought this subject up plenty of times, I need help, so I'm bringing it up again.

Specifically, I'm wondering if there are any new frontiers to get Smac to a better level of AI performance. For those new to these thoughts, I'll try to recount where we are so far below. The major stumbling block I'm facing now is the fact that the AI doesnt' maximize it's research effectively. I'm sure I'll forget a few points, or miss some unread threads somewhere, so feel free to add in to what I've missed.

1. Smac AI not challenging enough, or for various reasons of curiosity, the search begins.

2. In a nutshell, changing the faction predispositions toward passive and giving them a focus on everything but 'Conquer' in the governor queue forces the AI to build infrastructure, expand like crazy, and attempt to get as much tech/SP's as possible. (this is accomplished by setting the factions personality to [-1, 0, 1, 1, 1]).

a. This has the effect of eliminating the AI's annoying tendency to build TONS of military units in the early-mid game in exchange for better infrastructure. The AI, given a chance (ie, not involved in a war) will eventually cause minerals ecodamage, respond with parks and treefarms, and, when faced with war later on, be one tough SOB with the vastly improved minearls.

b. The reason I say 'not involved in a war' is that, unfortunately, if the AI is provoked, it will turn from 50-100 percent of it's build queues into military production, which isn't good military strategy!

3. Further, facilities and techs can be reorganized to give the AI factions acess to needed facilites, especially to respond to and to create, mineral ecodamage. The long and the short? Give access to some kind of factory, and at least Tree Farms in the early to mid-game.

4. Amplifying the AI strengths, forbidding expression of it's weaknesses.

a. I'm not sure how much has been done along these lines, but that's where I am in my development. Specifically, think about things the AI might build which really won't help it? I pushed back (further in the tech tree) biology labs for this reason, and it helps. Sure, the AI might make worms at the base it happens to have a biolab in, but the +2 research isn't worth it, and neither is the build cost/time.

b. What forms of attack/research/tactics/strategy is the AI good with? Sadly, not too many. The AI can't handle probe teams very well, so making these either late in the game or ridiculously expensive might be worthwhile. The AI sometimes likes missiles (yes, you can 'encourage' that in the scenario editor), and actually uses them fairly effectively, storing up piles of missiles to wallop a base in conjunction (sometimes) with a land based attack.

c. Ai is great at cheating the build mineral reqs (on Transcend) and growth nutrient rows. How can that be exploited? I've tried using more rainy maps, or drier ones. Both help in different ways.

5. Limiting Player strengths and amplifying their weaknesses. Well, a lot of work has been done here.

a. Choppers...limit them (tested and found mildly effective)

b. Crawlers...limit or eliminate (quite effective, no fun!)

c. Economy...humans can tweak the SE settings much better than the AI..so make the SE horrible. See Vel's Incredibly difficult mod, or the upcoming Aldebaran. Also, impose rules that the human player can't change SE settings more than once/X turns, ala some MP PBEM rules.

d. Research: force the game to 'Tech-stagnation' either in the alpha-text, the faction text, or just in the game setup. Works to a degree, maximizing the AI's ability to produce more for less, limiting the players ability to FLY through the tech tree.

So, before this get's too long, let me just say, help!

-Smack

[Googlie]

I'm experimenting in some scenario testing with making one or two AI's already subservient to a third, which forces tech-share among them (at least upwards to the master)

So far I've found making the Hive or drones the master with a Uni and a Consciousness as slaves creates a super AI faction by the mid game. And as you say, making the Hive "aggressive" means that they fritter away that advantage on weapons becoming a small, but strong faction. Making them "peaceful", but "team up against human player" seems to be the best combo yet, with lots of colony pods and terraformers.

In other experiments, I've found with my AI that setting them to "Sea Power" moves them thru Ind Auto quicker ( to Doc Initiative) and seems to get them building crawlers, funnily enough (at least in one test with the PKs).

Am currently testing a PK superfaction with the Uni, Consciousness and Angels as slaves.

Interesting experiments, these.

G.

[Ned]

Smack & Googlie, Two ideas:

First try accelerated start. This gets the AI its fair share of early SPs.

Second, rather than limit chopper range, try boosting base defense to 100% and infantry advantage to 100%. I believe this will significantly retard the usefulness of airpower against bases. This has been a significant weakness in the game since the introduction of Stealth Fighters.

Slaving two factions to a lead faction is fairly equivalent to giving the Datalinks to the lead faction. In my most recent game, Yang got the Datalinks and was, because of it, the lead faction late into the game.

This game, btw, was played with all the settings you listed, SMACK. I was playing DOMAI. I never got an SP until late mid-game. I believe it was the Nural Amplifier, IIRC. Until that time, the SPs were fairly distributed among the AI factions.

This game was also very interesting in that the AI was constantly sneak attacking each other and me after mid-game. It was amusing to watch pacifist faction behave in this manner.

The game was also interesting in that the faction with the highest research rate, AKI, was consistently behind in tech. I believe this was because she did not trade, but I may be wrong.

BTW, what happens if you slave both Alien factions to a Human faction?

Ned

[Black Sunrise]

I like bio labs better than network nodes. Cheaper to build, same research gain, and if you don't need to cash in on a artifact, it's a better deal. So making it harder for the AI to build such a useful item doesn't make sense from where I'm sitting.

Is there a way to change the cost to swap SE settings? If so, multiply by 10x. Now humans =cant= swap SE settings very often, it just costs too darned much.
Indra

[Fitz]

Same research gain? What are you talking about. A biology lab gives a fixed +2 to labs. A network node gives a +50% bonus. That makes them the same only if your base energy into Labs is 4 points or less.

[Avenoct]

Thanks all for the suggestions so far,

Ned, I've a question about the game where you didn't get SP's till the mid-game..was it accellerated and with tech stagnation? Also, it helps to increase the AI's interest in the SP's with the facilities-list AI influencers. You can have them all try to get the HGP and VW, for example.

Black Sunrise: Ah, don't think that's possible..never seen the SE switch cost in any of the files...too bad. It should be limited to one switch every 10 years, and cost a bundle.

Ah, scenarios. There are so many more possibilites in the scenario format! I just wish that the AI strategies could be set by the player or modified in a text file. 'Shamelessly betray human player', 'pact with X AI faction...make permanent'. Would be quite fun.

I'm leaning toward making two sets of factions, one that is fun, but tough to play, and one for the AI, which will be overpowered. The player then has to 'reload' the player faction version they want to play, but it's not too big a deal.

By the way, have you ever experienced what I'm talking about with warfare and all-passive AI factions? That they will give up their nice-infrastructure builds for War-making if they are provoked? It's a real splinter in the all-pacifist AI strategy..one I'm having trouble getting over. Also, the factions will make treaties and pacts with a weaker human player more often, which can be exploited (see Blake's post in Aldebaran on Creation threads r/e that).

(Shameles Plug: If any of you working on improving the AI would care to beta-test and make suggestions for Aldebaran, drop me a line at Rmontalbon@aol.com)

Thanks again,

Smack

[Ned]

Smack, Yes and Yes. Accelerated start and tech stagnation. Domai starts way behind in tech and has no opportunity for a very long time to even start building an SP.

Ned

[Avenoct]

Well, it's not good medical practice, but it's what to do in triage: Treat the Symptom.

1. The use of AI superfactions vs. 'normal' factions used by the player is quite effective. Give all AI versions say +2 to Industry, Support, and Econ, or Research, and they do pretty well. This is basically the same as adding a level above Transcend, where similar bonuses are conferred on the AI. I've also taken to giving the super-factions 'Social, +++Talent', but that's a more obvious cheat to the player who happens along in to view an AI city.

2. Scenarios: I'm not sure why we don't have a bigger library of them. They are pretty easy to design, and, while also not improving the AI, can put the player in pretty tough situations that feel 'legitimate', because it's a scenario, not meant to be a fair shake.

3. Accellerated starts and/or tech stagnation. Nice bulwarks to the treatment of AI symptoms of early senility.

-Smack

[Black Sunrise]

hmmm.
Really simple step that makes the AI much better, IMO: scatter forests over planet.
Since the AI doesn't terriform much/effectively, that one step seems to work really well.
Since that's a mod to the map, get a friend to build you 20 or 30 maps , randomly generated, and stick 1-2 squares of forest on each major landmass.
Then you just play today on map 1, tommorrow on map 2, day after on map 3. For you, its a random map, cause you never saw it, and only after you play through all 20 maps five or six times will you start recognizing them, and need your friend to make some new ones.

Indra

PS: I would -love- access to a library of custom maps. I often play scenarios just to see the map.
Any idea where I can find such a resource?

[Avenoct]

Well, you can try out Googlie's or Toka's recent scenarios, also check:

Chironian Guild

for Dilithium Dads' and other maps. It is surprising that there isn't a larger pile of maps sitting around. Maybe you should start a website Black Sunrise?

-Smack

[Black Sunrise]

Maybe you should start a website Black Sunrise?

I have one, if we want to host maps on it.
Indra