How was AI created so terribly?

The AI wasn't built with a very strong Strategic Core. I've heard talk of opening up the AI so that it can be modded by a true strategist familiar with LUA scripting and RTS games.

I can think of one person who can pull this off: Sorian.

He modded the AI in Supreme Commander to make skirmish games with bots fun all over again. The community would play and give lots of feedback, which he would use to improve the AI further.

Having your disconnected teammate turn into AI is the worst thing that could happen to you.  I'd rather that he joined the other team as an AI so I could farm him indefinitely.

Well its just some basic things with the AI... like Im no programmer, but you'd think its pretty easy to make AI realize running into range of towers chasing near full HP enemy who is just a bit faster then you is suicidal....

Then programming the AI to just teleport when its low HP instead of actually realizing, I cant win this fight or this enemy will beat me if I try to fight it so maybe Ill play passive....

According to the developers journal:

http://forums.demigodthegame.com/337968

The AI is suppose to be smart.

I don't think the AI is any worse than is found in other games like this but I agree that it has no place in competitive matchups.

Any AI is going to be relatively predictable once you figure out the triggers - even Sorians SupCom AI (which I loved, by the way) but it certainly could be worse.  If nothing else, I think it's lightyears ahead of Dawn of War 2's AI in most cases.

I'd say it's supposed to execute plans aimed at fulfilling goals and they TRIED to make it smart... but it's a multi player game, so AI probably didn't get anywhere near enough loving to be competitive against the online community

at least give them income x4... maybe that way they can get buffed up

The AI, oh the AI. I know LUA well enough and I've done AI tweaks before but I lack the time to do it all by myself.

What it needs to do is assess risk better. Humans are doing it, but it seems that it is only working off of what is happening in the very moment, which causes it to get easily confused. Why teleport when you can run? Why wait until you are killable in 3 hits to retreat? Why ignore enemy towers? I could go on.

But overall, the AI is not horrible. It tends to handle itself very well in terms of taking on a human. Its when certain situations arise that it seems to lacks the programming to preform properly. The biggest problem is the fact that  you know an AI will be a feeder, regardless of the skill level of the humans, simply because the amount of HP to trigger a retreat is entirely too low. The retreat method is also improper for playing against humans. Instead of running away, which in a lot of cases can take you out of immediate danger, they will teleport to the health crystal. While this is a viable and pretty intelligent move, you can't do it in melee combat with less than 1000 hp at anything but very low levels.

The AI needs to pay attention to a few things, and I believe its survival rate will be improved quite a bit:

• Where am I in relation to enemy towers.
• Can I survive this current DPS? (Enemy HP is visible. Will the AI be able to out DPS his attacker? Simple math really. I do it, so an AI can for sure.)
• Keeping with the DPS: Can I survive for the 2 seconds teleport will take?
• Keeping with teleports: Have they used an interrupt ability before? The AI needs to keep track of this on a DG by DG basis. If they haven't used an interrupt on it as of the moment of the teleport, they should assume its safe. Otherwise they need to run.
• How many DGs are near my target? (as in not running into a 2v1 situation....)

Im sure there is more but I don't feel like making a comprehensive list at the moment.

While I agree with the gist of what you're saying there proximity, that simple math call on the DPS, well it ignores a lot of extra  factors... well ok, 2 factors

Minions and powers.

both of these can change DPS substantially and while it's possible to do a worst case / best case assessment, doing a proper risk assessment on the use of powers and minions is probably not just basic addition

Proximity:  I don't know LUA at all, and it's finals week, so I don't have the time to look atm, but do the computer agents have local state?  I know a lot of the time, AI doesn't need to remember things, especially if it only needs to analyze what's currently happening.  If there's no local state, it would be extra hard to track enemy DPS, whether they've used interrupts, and even the location of the towers, depending on what sensors have been built in (if it can only see what isn't in the fog, and has no memory, it wouldn't know if towers had already been destroyed.

I was wondering, what if when an AI took over for a human player, it only gave 50% of the normal XP/gold (or more, or none at all) to the enemy.  Or even if that rule was in place during any MP game.  If I knew that the AI would not be feeding the opponents, I would welcome it instead of going "Oh NO, sure lose now".   It's the "feeding" the enemy that kills you, I think most could survive having an AI partner if not for that. . .

dont worry man , rook is the worst ai , noone thinks your bad losing that game , its an uphill battle which is very hard to win , just feel proud the times you do

It is because the AI between demigod and supcom is VERY different.

In supcom (and normal RTS), all AI got to do is build order. If anything lacking, you make more of it; then AI send them forward when they detect enemy base (where building location stay in fog of war) from plenty of scouts.

In demigod, these AI need to walk around search for players, AND not see them anywhere on the map. That means the AI need to be able to compute what he is detecting. In a sense, demigod AI would need to be closer to a FPS bot.

Supcom AI is already not that great, now demigod AI need to be even more complicated, naturally it will suck even more.

I've yet to see the AI of any game be anything above crap. It's completely incapable of thinking on anywhere NEAR the level a person is. It doesn't evolve, it doesn't gain experience from past games. Once a player plays the AI a few times they know exactly how it'll react and then they use that knowledge to beat the piss out of it. The AI can never do that, in any game.

Have you ever noticed in 90% of games the AI is a joke, and in the other 10% the only way to get it even remotely competitive they add levels above "hard" where the AI uses cheats to gain an advantage. The AI isn't actually any better, they just get to play outside the box you get to play in to give it hope of winning.

I don't think it's really a problem of the devs. The AI is limited by the fact that it is AI. That's the ultimate limitation. I wouldn't say the DG AI is any worse than any other AI, they're all worthless. I think it's a god damn joke the AI is part of multiplayer in any way though.

All I have to say about the AI is: Run, Forrest! RUUUN!!!

Seriously: Make it so the AI buys speed items instead of teleport scrolls and let it run away earlier. I think this might at least help with the feeding a little.

Ya, I've seen the AI chase my guy through and into all my towers many times.  It just does not seem to take towers into account when in battle.

I've suggested it before and agree with PurplePaladin that killing bots in multiplayer should award no gold (or maybe 10% of the usual amount or something). That way they wouldn't be an active hindrance as currently it would often be better to be 1v2 than having an AI pal in play.

As an anecdote, I played a game of Pantheon as Sedna with an AI Rook against a human UB and an AI Torchbearer. I actually managed to heal and babysit the Rook enough for him to score a kill against the UB  But only once and that was the only time I've seen AI do any good as a Rook.