Here are some things I noticed about the AI, through my tests, and observing others.
1. The AI tends to go after easy targets. If you leave a VP city open, they go for it over a heavily defended area. Sometimes this is good, other times it leaves them open to be surrounded and destroyed. Players can set traps for the AI, easily cornering them and destroying them by manipulating their goal towards undefended 'important' areas.
2. I believe the AI prioritizes provinces on a number of factors, VP level, depot, harbour, city level, etc. It seems the 'better' the city, the more the AI wants it.
3. The AI does not seem to remove embedded commanders from brigade command. For example, Holmes and Bonham will most likely be brigade commanders the rest of the game, even though they would better be served as divisional commanders. I believe the AI 'can' and sometimes possibly 'does' (although I have not seen it directly) remove them from command, but, for the most part keeps those assigned to brigades (and most likely divisions and HQs) in their assignments. So, beware if you assign scenarios to have lots of Brigadiers assigned to Brigades (I have modded Brigadiers into the stack, but not directly in command of a brigade, giving the AI and Player the option to do so).
4. The AI does sometimes attach leaders to silly units. One was Holmes (whom has been modded not to be embedded to his brigade) attached to a cavalry regiment who was just about to enter a 'death ride' behind enemy lines. A waste of even a poor leader. I personally have modded out leaders from being attached to cavalry regiments to stop the AI from attaching and killing off their command structure.
-
McNaughton
- Posts: 2766
- Joined: Wed Mar 21, 2007 8:47 pm
- Location: Toronto, Canada
(1) and (2) true, but the hunter-seeker algorithm can supercede that. If it don't triggers, it means there is not a good enough candidate for the kill (of one of your stack). But the evaluation can certainly be improved.
3. true. to be improved.
4. She does that only with excess leaders (theorically...). And it is not quite a waste, because leaders are extremely difficult to kill, so even if the cav is dispatched, the leader will sneak toward a gathering region.
3. true. to be improved.
4. She does that only with excess leaders (theorically...). And it is not quite a waste, because leaders are extremely difficult to kill, so even if the cav is dispatched, the leader will sneak toward a gathering region.
Hofstadter's Law: "It always takes longer than you expect, even when you take into account Hofstadter's law."
Pocus wrote:(1) and (2) true, but the hunter-seeker algorithm can supercede that. If it don't triggers, it means there is not a good enough candidate for the kill (of one of your stack). But the evaluation can certainly be improved.
As I see it, this can be an issue if the AI knowingly leaves a great defensive spot for a VP city, and puts itself in a position to be pummeled. Of course, VP cities are key to the AI and the game itself, so I don't know how you code around something like that. Maybe if the AI knows it is in a numerically-inferior position (greatly) to nearby enemy units? That would be a case where I'd say, stay in place (if in a great defensive position).
The thing I don't like, of course, is that the AI can be lured into a position and then sucker-punched. This was true of actual leaders, but the AI has less of a logic filter than a human. But I suppose gamey players can never be truly accounted for in AI coding.
Yes I know she can be lured, and this will be difficult to alleviate but not impossible (difficult as 'it asks much time to do'). The AI switch gradually to a cautious mode depending of the perceived enemies in the area, so she will only try to catch one of your stack if somehow superiority is achieved... but the player can alter the ratio by being just inside the FOW range.
Hofstadter's Law: "It always takes longer than you expect, even when you take into account Hofstadter's law."
PBBoeye wrote:The thing I don't like, of course, is that the AI can be lured into a position and then sucker-punched. This was true of actual leaders, but the AI has less of a logic filter than a human. But I suppose gamey players can never be truly accounted for in AI coding.
That's entirely true. No AI can ever play as human and that's why Kasparov could beat Deep Blue. (will he also beat Putin? lol)
Look at America's Army game.
There's not even the option for SP (only on Xbox version). For marketing purposes, it's hard to launch a game where you MUST play online MP and no SP but if you look closely there are many games doing so...WOW for example, ultima online, wolfET just to name a few...
The complexity of AACW is astonishing...managing to program an AI for it is even more astonishing and the results so far are so nice...yet i can't wait to begin my MP experience with AACW
-
McNaughton
- Posts: 2766
- Joined: Wed Mar 21, 2007 8:47 pm
- Location: Toronto, Canada
I think that the AI in AACW is superb, but was posting what I was noticing in order to help modders 'change things' or not 'change things' in order to help or affect the AI. Keeping leaders out of brigade command in starting scenarios, having VPs help aid the AI in attacking the right places, etc. will help improve the AI through modding. I am well aware, and really appreciate the effort AGEOD is putting forward to helping make this already good AI even better!
Who is online
Users browsing this forum: No registered users and 2 guests