Couldn't hurt!

Very interesting, I like it. I could see myself using something like this on occasion. Although I'm traditionally a big picture kind of guy and don't micromanage unless necessary.

IT-land is profoundly polluted with bad slash-compounds, so it's understandable that he'd use a slash instead of "and." The bulk of the OP is about champion assignments, so quests seems reasonable to include in the title to me.

And I definitely like the idea of seeing AI help the players in this sort of situation. (I'm also very greedy and like even more the idea that these assistant-AIs would be able to notice when their zone of action has gotten weird, turn themselves off, and tell you to come look.)

Probably ‘quest’ is the wrong word for this, ‘Order’ maybe more appropriate.  
Robbie, my approach is different to yours.   All orders do not involve AI’s speculation at all.  Even if there are a few pre-requisites for the Stable building, the game will queue up the pre-req in a sequence determined by its building cost (or some other quantifiable criteria).   The game UI explicitly tells how it is prioritized.  The UI shows the queue and allows gamer to adjust it; nothing is hidden compared to AI governor, or try giving AI a goal to achieve.

Even when some pre-req building requires tech research, the game will queue up the tech research, which the sequence is also adjustable by gamer.

Here I am not asking SD devs to create AI that defend the city well, or do something that requires AI’s judgment.  There are a lot of mouse clicks that can be saved by my suggestion, for simple things like building a structure or order a hero around, all sort of mandane tasks.

This UI feature allows Macro-management without the use of AI governor (which will surely disappoint).  For those for when to mirco, they can queue up all production the old fashion way, one by one.  For macro guys (or when your empire is have 20+ cities), he probably doesn’t mind building new structures in a slightly inefficient way.

The same principle applies to orders given to heroes.  All orders are clear cut & well defined.  Whenever the heroes encounter something requires attention in its LOS, he writes a log entry.

This suggestion should reduce a lot of Micro what most of us don't wants!

I think your idea is too lofty. Don't get me wrong, if they did it it would be awesome, but I just don't see it happening. Have an AI capable of doing the kinds of things you want while taking the 'bigger picture' from the player via some method of communication still seems like sci-fi to me, unless you get a billion dollar DoD contract... And the problem is, if they try for something like this and fail, they end up with a lump of worthless coal.

I disagree. If it were easy or simple in any way to make an AI able to converse meaningfully with a human, then AI governors - not to mention diplomacy with AIs - and the like would be so much better than they are. Getting an AI of today to actually understand human intentions and goals and strategies seems like way too much to ask.

But then, whether or not I disagree on its feasibility is irrelevant because I'm not a professional programmer or game developer or anything like that, and I wouldn't be the one tasked with the job

I agree with Post #9 to 11. 

Robbie, your idea is at one higher level of the micromanagement spectrum.  The lowest form of the spectrum starts at micro everything, aka traditional TBS.  Mine is the next higher one, automate everything possible without the use of AI, the AI never make any decision on behalf of you.   Yours is a version of AI governor.  Then the highest level of the spectrum is AI play the game (basically) without gamers’ input, aka MOO3.

Goodmorning all,

I know that the system i am proposing is ambitious,  but i don't think it is overly ambitious.

Consider, half the system is almost a requirment. the A top down AI desission making process will be needed.

Some part of the AI will have to desided what it's top priorities are, and baced on those top priorities devise secondary goals, and from those what direct actions it will need to take.

At this stage all i'm adding is that those choises contain strings, expressing explicately what the meta, goal, and subgoals are, and since the AI must have some way of deciding to do B because of A, then have a string expressing that,


Now you have a system where you can ask the AI what it is upto.

so you make that list, and parts of it, Espionoge stealable,  you make parts of that that list also accesesible to users when they turn on Govners and managers of cities/caravans/whatever.  Still free.

Now if you wrap that all into one AI governing aid-de-campe.  You've got a very powerful too, highly mod-able, to the extent that users can Program thier own Complete AI's to compeat against one another.

Now the hard part, Since players can Steal AI plans,   AI's need to be able to steal User plans. 
Thats harder.  but not impossible, Consider if the AI tools are powerful and useful enough, people will volentairally keep them mostly accurate.  "build me a stables, and weapons and armour, to equip Bear calvary" [more accurately, subgoal, Build: Bear Calvary, Actions: Build: Stable,weapons(of bear calvery),armour(of bear calvery)].

Now you have the AI fill in the Meta goals, by looking at all the goals, and guessing what Meta goals the user is going for [unless the AI is directly told ofcourse]  for any set of exlicate or obvious subgoals the goals can be guessed moderately accurately,  and the Meta goals from those goals.  At first it would be inaccurate, but if it is capable of remembering which meta goals seamed favoured by the user, eventually it would learn.

The last step IS hard, and it IS going to be more work, probably involving at least a small neuronet like proccessing stage. I do not belive it is beyond Stardock, personally.  and even if THAT part of the AI isn't perfect the rest is easy enough and WELL within Stardocks abilities I am confident in that. If they choose to, they could. 

It would add a LOT to gameplay to have an AI govnor who isn't doing blind secret stuff behind your back and always needs to be corrected for building the wrong buildings or making units you don't need can't afford. (like every other 4X 4E game ever).

and it is possible.
I'm pritty sure i could find at least 3 undergrads graduating this year who could do it, in just my collage alone. at least all but that last step. that one is hard(er).


Take care all

Robbie Price

While yes any form of diologe would be over the top, and parcing sentence structure to extract battle plans, would be undoable. The AI will have a limited list of goals, meta goals, avalible to it.  You add to each of them a string, and you can communicate to a user.   Then you show the list and the user chooses from it. 

No real diologe is provided, it's build by lego. Unless somebody wants to Phython script a NEW meta goal, or a new goal, then you work with the lego you've got.

To do all but the guessing what the user is doing, at a functional level is not hard, as long as you expect the humans to be able to play by the AI's rules. which we can.   Most users will use little more then the city manager AI, and maybe the caravan AI. Those have sufficently limited variablity that the AI interactions can be made much more pleasent then say SMAC's city managment, or CIV's.   IN those you have little more control the to say, 'you can't build these types of units, you can't build these types of buildings.' and give VERY basic commands to new units. 

This isn't about making something the US army is going to want to buy from stardock this is Show the users what your going to do, Show the user why,(Both you need to have in the desicion proccess somewhere, so show it), then let the user tell you your wrong. The user has to use the AI's language sure, It's programming not stratagising Yes, But it's three steps above anything anybody else is offering.

All this is is an AI with a UI.

take care

The hard thing would be for the opposing AI's to be able to espianoge steal your plans.


There are two ways to handle that, you can have them not be able to steal them/steal more bacis information, this unit was sent there, this town's got this on it's build list.  or you can something populate a plan for the user, both so that they can keep it up to date [allowing them to build stratagies, check things off as they get them finished, remind them of things they were trying to do... when playing at night(pronounced early in the morning before going to bed) having a handy turn by turn checklist would be nice for my sleep addled brain],  and so it can get stolen.


It's hard, but to do a 40% job of it, isn't.

Consider a 3 win game sanario.  You can win by diplomacy, concour, or magic spell.  An AI would know that if IT was trying to win by Concour it would have a big army, modest number of spells, and not very much trade, and a few good wars.  If IT was trying to win by Magic it would have a LOT of reasearch, a LOT of spells, modest army, and good trade relations, and if IT were trying to win by diplomacy it would have LOTS of trade, smaller armies, and a decent but not scary spellbook. A real AI would also have Expand and Explore as meta goals, and probably more things too, but simple demonstration of concept.


If the player fits into one of those sets,  the AI would set a meta goal for the player as that.  IT might be wrong, infact it quite probably would be wrong 30-40% of the time. but it would be better then nothing, and it would be enough that the user doesn't have to populate ALL the fields by themselves to make a resonable plan if the user wanted.

For every Meta goal the has goals and subgoals it thinks are best to meet those,  By looking the Subgoals and goals you've achived lately it decides the most probable goals meta goals your working towards.

The last thing a User wants to have to do is input manually all the goals all the proceedures, So the AI populates each field with a guess, a semi intelligent guess, but still a guess.  If the User  opens the AI planbook and changes a goal it changes all the subgoals under that goal to match. so that the user can then look at those subgoals and deside that they are sane, or change them. 

The user isn't required to follow those plans, and if the user doesn't the plans will change to match the actions the user does take so that the user doesn't have to Update it all manually when they return, if they return.

I think i've made the system sound much more powerful then ever i intended it to be.  The AI is not guessing your thoughts,  It's matching your  actions to what goals it would have to have to take those same actions. Or more acturally finding the most likely goals it would have to have to have it take those actions.  Having made a guess at what goals you were trying to meet, it then builds a plan for what goals IT would meet in the next 15 turns if it were you, and it had the same goals as it thinks you have.


It's big, but it's not impossible.

i hope that's clearer.

Robbie Price