House Rules against the AI

[Gandalf Parker]

Little human type pretenders tend to stay home, and of course immobile pretenders stay home. Its just the SC pretenders that tend to go out. And thats what they are meant to do.
The AI doesnt seem to have a concept of pulling back within the domain candles to recover and wait for new backup to arrive. But if the god goes too far and dies then it just does "Call God" and gets it back to start out again (losing some magic) so I guess piling magic on it fits the tactic. Not a GREAT tactic, but a tactic.

[Sombre]

Originally Posted by Fantomen

Are you sure?

The AI often funnels these design points in super uber magic paths that it is too stupid to use, then cheerily sends the supercharged pretender out to die. All the while succumbing under some completely ridiculus scales.

To me it has seemed more effective to design an imprisoned god, preferably immortal, with some death for skele spam and other paths suitable to cast the nations need of globals, plus good income scales and suitable other scales.

Even with horrible economic scales, the income and resource bonus from impossible difficulty completely outweighs the negatives.

The AI gets + 100% to gold, resource and gem income on impossible. That's far more than going from all negative 3 scales to all positive 3 scales can net you.

So to reiterate: Don't start the game with human players with sensible scales then set them AI to try and give the AI an economic advantage. It doesn't make sense. You can give it a research advantage or something more marginal to do with cold blooded etc, like I said, but economically speaking the AI is always best off being set to impossible.

[thejeff]

That may be what SC gods are supposed to do, but the AI can't handle SCs.
They go out with random or no gear, run up fatigue casting marginally useful buffing spells and get killed.

It's fine to say that's intended, but it doesn't work.

[Gandalf Parker]

That happens every time? Or randomly?
I never said it did it well.

[Tollund]

It happens virtually every single time. The AI will take an SC pretender, not give it any equipment, have it cast flying shards once then attack.

[rdonj]

When I've had the AI draw one of the bulls as a pretender, it tends to not completely waste it. Generally, imo, good pretenders for the AI are pathless SC pretenders. The AI will usually get best use out of these, and if it dies it can just be resurrected with no penalty. Of course, it will eventually become useless anyway due to afflictions unless it has nice regeneration and or recuperation.

[Makinus]

My major gripe with the AI is that it´s incapable of keeping up in the research curve... it simply does not research fast enough.... thats the reason of my house rule of 1 researcher per lab... it´s the only way for the AI to keep up...

[Edi]

The AI seems to do fairly well with independents at strength 7 and impossible AIs, especially if it has good scales (happens sometimes). Indies 7 means you need a lot more troops to beat them than indies at 5, and the AI resource and gold bonuses mean that it gets going pretty fast. It's not at all uncommon to get swamped by two or three hordes while you're still trying to get basic things done.

I play with the Better Independents mod and difficult research, so that slows me down a lot more than it does the AI.

[Gandalf Parker]

There are pros and cons to the AI doing things totally randomly.
On one hand, it usually gets it wrong. On the other hand it can surprise us.

Good AI (artificial intelligence) is not good AH (artificially human). Good AI is the intellgent choice. "A straight line between two points" which is too predictable.

Good AH is almost achieved by "random selections filtered thru intelligent choices" but that takes alot of code. Even for ONE of these nations such a set of rules would be huge to do it with any variety for replayability.

The best bet WE will ever see are more designed AIs for each nation being submitted to the SemiRand. Or a really good mod builder creating an AI.

"AI programmers consider it a giggle that to go from AI to AH requires dumping some intelligence in favor of randoms. There must be a comment there about the human race."

For more reading: the Association for the Advancement of Artificial Intelligence
http://www.aaai.org

[Tollund]

Originally Posted by Gandalf Parker

There are pros and cons to the AI doing things totally randomly.
On one hand, it usually gets it wrong. On the other hand it can surprise us.

How can it surprise us? By doing exactly the same thing in every game?

Good AI (artificial intelligence) is not good AH (artificially human). Good AI is the intellgent choice. "A straight line between two points" which is too predictable.

And yet, the best chess programs in the world work in exactly that way.

Good AH is almost achieved by "random selections filtered thru intelligent choices" but that takes alot of code. Even for ONE of these nations such a set of rules would be huge to do it with any variety for replayability.

I'd rather worry about competence before I worried about replayability. The AI is currently incompetent at anything beyond pushing armies around the map. And yes, I mean it when I say that it's incompetent. It only has a chance against a non-beginner when the odds are more than three to one in its favour.

The best bet WE will ever see are more designed AIs for each nation being submitted to the SemiRand. Or a really good mod builder creating an AI.

How, exactly? No mod can change the AI behaviour in the game. Better pretender designs won't fix the lack of coherent pre-designed plans for the AI. Allowing the AI to cheat by giving it massive starting advantages isn't a solution either.