dstaltman

It's working great so far from what I've seen. The friendly AI can actually be depended upon to pursue a good plan of action, not just mingle about.

One problem I've seen is when the AI decides to pursue someone with extremely low health. Normally they will pursue that person a long way, and not ignore it which is what they do if you're not attacking them and at >40% health or so and running away. Anyway, sometimes they will pursue a person a long way into the enemy base, "hoping" to get the kill. So far so great. But sometimes it fails to get the kill, and gets stuck inside the enemy base. Instead of trying to escape, it will stand and fight towers until it reaches <20% health, and then try to run (which is often too late for most of the game).

A "run away!" plan is needed in case they are alone against a large amount of towers and going below a certain percentage of health (ideally significantly higher than the normal "run away!" trigger).

Edit: I would also like to see AI demigods play more with other friendly demigods. For example, they could literally follow human demigods around when it makes sense.

Interesting comments.  I was taken by surprise alright the first time the AI took my valor flag while all the towers were still up.

Any chance of making the AI remember and change based on player actions?  Right now the same tactic can be used over and over again.  An example of this is laying down mine fields as Regulus between the AI's citadel and one of their portals.  After the dying twice in a row it would be nice if they would check for mines the next time they try to get their flag back.  I would like it if the AI could learn where the most likely spots for my mines are and check for them instead of charging in. 

The ideal single player AI would learn between battles in the cmpaign so the job of ascending gets harder as you play.  However, the only AI I know of that adapts the more it is played was made for TA Spring.  A 'learning'  AI would make Demigod standout from all other games.  In theory 'super' AI's could be developed by giving them data on how the best Demigod players play.

If learning AI can't be put in the game now, then it might be a worthy goal of the year of free updates or in an expansion pack.

lol, I agree with every one so far even kitkun! Also ty for talking to us about stuff and plans.

Is this why my Yetis dance around in a circle stuck or doing nothing with enemies insight? All you have to do it play Senda once to see this. May be the AI isn't done just yet.

Really interesting. 

Something to note - At high levels of play in a 2 human v 1 human and 1 nightmare AI - the team with the AI will always lose.  This is because the AI has no parameters which say "drop what I'm doing and go by the quickest route possible to where my ally is getting crucified by two opponents" or "somewhere on the map an opponent DG has less than 1k health, I should snipe them for the kill." 

I.E. A good team of 2 humans will beat 5 nightmare AI because the nightmares do not tagteam or synergize their abilities effectively. 

The self-preservation between this Beta and the last one is much better, but as HF says, if the AI is in a grim situation when priorities shift they can't exit effectively.

Nightmares chose great items like Heaven's Wrath for example - but they trigger lightning at game start on my citadel - not gonna do much for them.  They'd be alot more effective if they used it against grunts early for the xp and shifted to citadel targeting only when within a given radius (about to win, basically).  It seems like alot of favor item usage follows this pattern. 

Kestrel hit the nail on the head alright. Another similar example is that the AI will chase a low hp Demigod regardless of whether they can kill it + they aren't very wary of towers.

i think the ai's are getting smarter. I've seen them use potions, teles, and caps lock. They seem to be getting smarter keep it up. Why are u talking about Erebus and Tb last time I checked we couldn't play them.