• This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn more.
  • Notice the new look? We've updated to a new version of Xenforo, please read our latest post.

suggestion AI assistant

Status
Not open for further replies.

Tommonius

Necromancer
Backer
Dec 15, 2012
616
247
400
29
Wolverhampton
#1
Okay I got a bad history of making threads that people already discussed but a glance at the pages of the suggestion forums does not show AI assistants like we had in DK1 (if it does sorry about this)

Did people like the assistant managing parts of your dungeon like making new rooms? Personally I LOVE managing my own battles/dungeon economy and hero conversion but a few people in the past in discussion have noted annoyance at micro managing and making the same dungeon each new level.

So do people want AI assistant or not? would it be possible to put in wTFO? and if so what functions could it manage?
 

Lord of Riva

The Lord
Founder
Dec 29, 2012
2,786
923
485
30
#2
nope nobody suggested that :D

and to be honest i think this feature sucks badly, the Ai wont probably help me getting my tactics executed, irrelevant if it is good or not... a machine cant think... ;)

anyway adding this ;)
 

Fireeye

Augre
Backer
Dec 30, 2012
1,155
687
515
Ze Germany
#3
I do not think this would make for a good feature because of three reasons:

1.) While having to micro-manage your dungeon may at times seem mundane, it is still an integral part of the game. Handing this role over to the AI would not only drastically limit the variety of different strategies you could employ, but could also give you an unfair advantage over players who chose to micromanage the dungeon on their own, as they effectively have to divide their attention between fighting you and keeping and eye on their own home.

2.) Most AIs in RTS games are frustratingly stupid; a result of the fact that they can only follow certain program routines and are largely incapable of adepting to new situations. As such, an AI agent may very well carve you a nice dungeon, but it will more than likely fail to acknowledge the problems a specific mission in the campaign or a specific human opponent in multiplayer games will offer. There is, simply put, a reason why winning against an AI is usually remarkably easy when compared to winning against a human player of the roughly same skill.

3.) If people dislike micro-management, this simply is the wrong game for them.
 

Cultist Joris

Ember Demon
Founder
Dec 30, 2012
1,555
840
545
24
Belgium
#4
I never used computer-help in DK anyway, except by accident (it ruined my whole dungeon design... :()... It don't think people will need this, especially when there will be difficulty levels...
 

Tommonius

Necromancer
Backer
Dec 15, 2012
616
247
400
29
Wolverhampton
#5
I actually agree with the sentiment and am playing devils advocat here a bit. I think it was the old timed levels thread that people brought up micro managing a dungeon especially when time was sensitive, also some indeed did dislike remaking the dungeon (shocked to be honest the best bit is making a dungeon for me AND modifying it for the needs of the level)

I suppose in timed levels you could simply be given a few standard rooms if it made people happy. also:

I never used computer-help in DK anyway, except by accident (it ruined my whole dungeon design... :()... It don't think people will need this, especially when there will be difficulty levels...
There will be easy and hard modes? WOOOOP! I am going to love hard mode from the start, really get the sense the heroes are throwing everything and the kitchen sink at you... in fact a kitchen sink catapult trap would be nice :cool:
 

Mozared

Juggernaut
Founder
Feb 17, 2013
1,132
836
520
27
#8
It's an interesting topic. Gut reaction based on DK1 and the argument that dungeon management is an integral part of the game immediately says no.

That's not said the whole idea and concept becomes uninteresting after that point, though. I have a feeling something could be done with helpful AI's that doesn't make the game less game-y and doesn't ruin your own dungeon design. The question is, what?
 
Likes: Amon

Amon

Ember Demon
Apr 14, 2012
2,264
811
545
22
Somewhere
subterraneangames.com
#10
Hmm. Not liking the idea; the problem I could see emerge here is the strategy of the player not being 'in sync' with the AI's approach. That being said, I think it could be nice if the AI could pose suggestions; I can see new players and casuals benefit from certain hints.
If that makes it in, then it should be an option that you can turn off. I would leave it on, but other people...
 
Jan 3, 2013
3,239
739
495
30
#11
People probably never used assistant the way it was intended and assistant itself wasn't tailored properly, not to mention it wasn't working properly on all maps. However, assistant in DK1 ( Pick Only ) could deploy all your troops and use spells in combat in a second, making things much easier. DK2 pseudo-assistant is purely for testing ( game was unplayable because AI assumed full control ).

In short - feature can be quite useful, if player can turn it on/off and configure it the way he needs in some simple way.
 
Jan 7, 2012
446
171
340
22
#12
The ai builder was very fast in dk, the mechanic could be abused by using it as a starting help for multiplayer and campaign levels were time matters.
Maybe a cheap hint mode would be enough, click on tiny button on the far left to get a text message with a room priority or building order to help you in easy mode.
 
Likes: Amon

Medjay

Shadow
Backer
Aug 23, 2012
1,095
491
490
23
Canterbury
#13
I rather like stretching myself to the limit of my micro-managing ability. I mean, in essence this game is about which player is the better micromanager (domestically and in combat), so I can't see how making any of this process automating would make the game any more fun. Sorry :)
 

Ben Chandler

Impassibly Cool
Backer
Jan 23, 2013
311
87
275
#14
I think on occasion if everything was secure I used it in dk1 while leisurely waiting, with the setting so it basically only slapped lazy imps and slow trainers.
 

Zieluowaty

Dwarven Worker
May 19, 2013
9
0
20
28
Łódź, Poland
#15
I want say about AI.

If you want to make really good campaign in this game you must make AI which can learn. I know that it's hard to implement but look at the SC2 AI. It's learn all the time and play with it is always new expirience! Think about it! Campaign where your opponents are stronger from mission to mission!

Secondly you should do something with workers AI. When I tell them to dig, they don't want to do nothing else! Player should has some options which gave him opportunity to tell how many percentage of all workers should dig, build walls, capturing new lands and else...

ps. I see that you choose Unity, that's a great graphical and physical engine! Good luck!
 

Zieluowaty

Dwarven Worker
May 19, 2013
9
0
20
28
Łódź, Poland
#17
Supreme Commander 2, not Starcraft 2, I forgot that when i write SC2 someone can thought about Starcraft xD

ANd I know what mean learning AI, beacuse I must to learn about it on study :p
 
May 16, 2013
201
82
170
UK
#18
Huh, I never played Supreme commander, so I can't comment, but I'm surprised that any game has had a true learning AI, let alone a game that came out three years ago. For one thing most people's computers wouldn't be able to handle it, even people with gaming computers might struggle because they often focus on GPU to classify a computer as a 'gaming computer'.

Then again I suppose it might just be a limited/stripped down learning AI. It doesn't take much to be able to classify your game's AI as a learning AI when compared to every other game out there. Unfortunately AI technology has a history of over-promising and under-delivering, mostly because it tends to make a very beefy system to do anything even remotely intelligent. Intelligence is hard it seems :p .
 

Zieluowaty

Dwarven Worker
May 19, 2013
9
0
20
28
Łódź, Poland
#19
I'm not speaking about full AI which can pass the turing test
Only about the system which can collect data about the opponent's strategy, how he builds rooms, trains monsters, collects gold.
In old games like, for example, 'Dungeon Keeper' or 'Knights and Merchants' opponents had many resources, nearly unlimited and no inteligence at all. They trained troops and sent them to our bases.
If you raised difficulty level the only thing that changed were special multipliers for speed of training troops, their strength and everything like that. When you find some way to force a computer player to do something stupid, you could easily win all missions (for example Stronghold: Crusader Kings).
 
May 16, 2013
201
82
170
UK
#20
To the best of my knowledge there is no game that doesn't increase difficulty, at least at the higher ends, by giving the AI more stuff (more resources, convert resources to units faster, etc). It's a phenomenon known as "The AI is a cheating bastard", which also covers games where the AI doesn't have fog of war.

I'm also not talking about AIs that pass the turing test. The most successful AIs for this kind of application (that I've heard of) have actually used evolutionary models (Take a 'base' AI, generate 10000 copies with random variants, have them fight thousands of battles, take the most successful ones and generate thousands of copies and have them fight, etc). Obviously that's unsuitable for making an AI practice against a human because even if someone played solidly for a week you wouldn't even get through one 'generation' of AIs, so practically no improvement. My experience with this kind of AI development is periphery at best ofc, while we did have two courses on AI development at uni they tended to mostly cover fundamentals and the mathematical basis of different approaches for AI development for any field, so I don't know what research groups at the cutting edge have done, and their successes tend to take a few years to filter through to actual products.

EDIT: Huh, I rambled a bit, but anyway I imagine that the greatest hurdle to making a 'learning' AI is how you make it recognize actual strategy. How do you get it to analyse it's opponents actions? AIs (that aren't generational) are generally terrible at adapting to emerging situations, emergent gameplay, that humans are very good at adapting to. The best you would probably manage with a learning AI is some kind of random but weighted system based on previous success, but that would still be unable to deal with 'trap' situations because that kind of system 'learns' good unit compositions, not good pathing, not good strategy, so you could still lead them into 'the valley of death' repeatedly. Systems for avoiding 'the valley of death' do exist ofc, take random approaches, use units with different mobility options (eg, flying units) and the AI can use them, and you can weight the different approaches, but honestly the success of those different approaches tends to be very situational. Attacking through the front door (say in starcraft 2) is probably going to succeed in the first few minutes, but not after that, at least for the first base and the expansion. But it again becomes a valid strategy when attacking later bases because often they go undefended by any static defense so the only important thing is to get there while avoiding the army, weather or not you are a flying unit or a ground unit doesn't matter.

ugh, rambling again.
 
Status
Not open for further replies.
Top Bottom