Jump to content

  • Log In with Google      Sign In   
  • Create Account


Artificial Intuition


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
57 replies to this topic

#41 Paul Cunningham   Members   -  Reputation: 122

Like
Likes
Like

Posted 07 August 2000 - 12:11 AM

It sounds like a very good way to go to me pax. Because it''s feasible. Hmmm

I love Game Design and it loves me back.

Our Goal is "Fun"!

Sponsor:

#42 pax   Members   -  Reputation: 122

Like
Likes
Like

Posted 07 August 2000 - 06:21 AM

Woohoo, somebody liked my idea!!

In trying to develop good AI, I look at myself. Why do I do things? What''s my motivation and what external influences do I experience? How can I represent that on a computer?

In any situation, there''s a mood, hence the mood determination net (can also be used for selecting appropriate music?). Then using the mood and other input (each of the relationship outputs and maybe one for the location) I determine what I want to do. This can be weighted by my goals, schedule, priorities and personality/values.

The problem with AI is that there are so many factors to realistic behavior, both internal and external.


Pax

#43 ahw   Members   -  Reputation: 261

Like
Likes
Like

Posted 07 August 2000 - 06:35 AM

Well, I was watching a fascinating program last night on the evolution of the idea Science has had about the importance of the brain. And they showed that lobsters for instance, have a dramatically simple nervous system, and yet has to deal with 8 legs plus 2 arms when walking onland. Just to say that there seem to be some hope of simulating all this in a simple way. Of course, it will only be simple when we figure it out.

Maybe if we deal with tanks being very simple beings (I just can''t forget my first program, an ant simulator in BASIC, very nice emerging behaviour, that''s probably what started me in studying all this computer crap rather than Fine Arts)

maybe we could do something. So the question is, what kind of input and stats do we give to a tank.
I am also thinking, for my MAsters, about seeing all this as a multi scalar problem. The tank is a being, but on another scale, the tank is just a "cell", an arm, of a greater being, a group for instance; which is itself a part of a greater body, etc.
Can all this speed up some processes, this is what I expect to see.

So any idea how we could model a "unit" on the battlefield. For the sake of argument, let''s take the RTS context, unless you see a better context ? (NOT RPG. There is way too much stuff to take into account)

#44 egerlach   Members   -  Reputation: 122

Like
Likes
Like

Posted 07 August 2000 - 06:45 AM

Finally! I can post! (I''ve been having trouble postsing ever since they made the upgrade to the message boards)

Pax, I like your idea, very similar to the one I suggested not 5 posts earlier. Correct me if I''m wrong, but I think you''re suggesting that you use an ANN for every unit, and then a master one for everything. Each of those nets on their own is feasible, as Paul suggests, but as a whole, when there''s 200 units on the board? You''d probably see that ANN routine slowly eat up your processor time as the game got bigger. But I do like the ANN determining the mood; if you look at my suggestion buried deep within another one of my posts you''ll find they''re very similar. I like your term better than mine too.

Guess great minds really do think alike.

Eric


===============
"Or a pointed stick!"

#45 egerlach   Members   -  Reputation: 122

Like
Likes
Like

Posted 07 August 2000 - 07:04 AM

quote:
Original post by ahw

Well, I was watching a fascinating program last night on the evolution of the idea Science has had about the importance of the brain. And they showed that lobsters for instance, have a dramatically simple nervous system, and yet has to deal with 8 legs plus 2 arms when walking onland. Just to say that there seem to be some hope of simulating all this in a simple way. Of course, it will only be simple when we figure it out.



ahw, you (and everyone else) should check this out:

http://www.frams.poznan.pl/

It''s Framsticks, an AL simulation. I can''t do it justice explaining it here.... you''ll have to check it out for yourself.

Perhaps there''s something in it for unit AI?

Here''s a good question: Does better unit AI make the system more intuitive with emergent properties???

Eric



===============
"Or a pointed stick!"

#46 ahw   Members   -  Reputation: 261

Like
Likes
Like

Posted 07 August 2000 - 02:04 PM

No one ... can tell you what the Framsticks is. You''ll have to see it for yourself. LOL

God, it''s that time of the night again.

#47 pax   Members   -  Reputation: 122

Like
Likes
Like

Posted 08 August 2000 - 06:03 AM

ahw: Your multi-scalar idea is called heirarchal AI and is being used or at least investigated for RTS games. The way they use it is that there is an AI for the player, an AI for each group of units and an AI for each unit. The player AI determines overall strategy and assigns tasks to the groups. The group AI determines how to best complete its task, and the individual AI determine how they complete their own part of the task.


Pax

#48 BeanDog   Members   -  Reputation: 1063

Like
Likes
Like

Posted 08 August 2000 - 08:10 AM

I think that the only feasable option in using a NN for RTS AI (how''s that for acronyms?) would be as an overall strategy-controller. From my (1 hour) research, NN''s use a good deal of processing power. Individual unit AI should be built in at the most basic level. For instance, in my RTS-under-construction, in the damage-dealing routine, I check for any units that got hurt and tell them, based on their power and the attacking unit''s power, to attack that unit or retreat. Then I take any other unit that could see the unit that got hurt, and repeat the process for them. So, if you attack a group of marines with one marine, that marine and all the ones near it will attack your unit back immediately. The same thing goes if the computer attacks your units.

Where the NN would come in would be in telling the computer how to best build a base, when to go beyond friendly territory and build a new base, when to make an attack, which units to build given certain conditions, etc. What I''m talking about is a mostly factory-trained AI-NN, and letting it learn on the players'' computers, too. So, if the standard set of strategies fails miserably, the computer would start experimenting, and when something starts working well (this is where NN''s are awesome), it would slowly become the natural thing for the AI to do. And if the player changes his strategies again, well, it''s no problem for more than a few games.

BTW, is there a tutorial out there to build a neural net in C++? I''m not sure I fully understand the underlying concepts. I have a book, "Neural Nets in C++", but it is from 1992 and WAY outdated before the ANSI C++ standard, so the coding is a bit wierd. Plus, the associated source is on a 5 1/4" floppy...

~BenDilts( void );

#49 silvren   Members   -  Reputation: 122

Like
Likes
Like

Posted 08 August 2000 - 09:25 PM

Beandog:

Hi, personally I prefer to read through algorithms and implement them by myself.
No need to buy books like "....using C++" gfx/NN whatever.

They are usually quite badly written.

Anyone has the same opinions?



/Mankind gave birth to God.

#50 egerlach   Members   -  Reputation: 122

Like
Likes
Like

Posted 09 August 2000 - 05:35 AM

A good reference for ANNs is the follwing, although it's not really a tutorial.
com.ai.neural-nets FAQ
This, on the other hand, is a good place to learn algorithms (implementation not included):
Backpropagator's Review
Another link, explaining a different kind of ANN, perhaps more useful:
Self-Organising Maps

===============
"Or a pointed stick!"


Edited by - egerlach on August 9, 2000 12:39:23 PM

#51 Abramski   Members   -  Reputation: 122

Like
Likes
Like

Posted 12 August 2000 - 02:49 PM

I have two questions . . .

Why is a guy in Canada speaking Japanese?

and wasn''t the original topic how to make a computer opponent SEEM to be acting intuitively?

I''m just a newbie, so don''t flame me too hard, but couldn''t you just have the conputer run whatever scenario from a certain point once in a while, like the space cadet seemed to be implying, as a simplified version of the game you''re playing and play for both sides, then prepare for whatever horrible thing the player has in store for it? Or, better yet, run the game about 3-5 minutes ahead and see where all the pieces are and react to any upcoming threats, thereby seeming to anticipate it??

Abram

#52 dwarfsoft   Moderators   -  Reputation: 1209

Like
Likes
Like

Posted 12 August 2000 - 03:49 PM

1) No idea what you are talking about
2) Because the topic is about Intuition, simulated/emulated Artificially and therefore is quite deserving of its title. I think it is catchy,

*Pulls out a flame thrower* Any more statements?

-Chris Bennett ("Insanity" of Dwarfsoft)

Check our site:
http://www.crosswinds.net/~dwarfsoft/
Check out our NPC AI Mailing List :
http://www.egroups.com/group/NPCAI/
made due to popular demand here at GDNet :)

#53 Abramski   Members   -  Reputation: 122

Like
Likes
Like

Posted 12 August 2000 - 06:24 PM

1. egerlach

2. I agree, it is a cool name for it, but . . .

quote:
Original post by dwarfsoft

But anyway, Now that we are all agreed that NN''s are gonna suck for what we were discussing its use for (I hope we all agreed right?) what can we use instead?



. . . not being a [strong]programmer[/strong], I was coming up with the best idea I had. Not that I have anything against programming or anything, I just don''t think visual basic or Excel functions qualify me to be giving any kind of opinion on the effectiveness or appropriateness of neural networks. I am a little confused though, because the topic did seem to jump from simulating/ emulating intuition "How''d it know I was going to do that?" to advanced strategy and tactics "How''d it do that?".



Abram


I''m on FIIIIRRREEEEEEEEE[/I]

#54 Abramski   Members   -  Reputation: 122

Like
Likes
Like

Posted 12 August 2000 - 06:25 PM

I hope I've made my point about my programming ability.

Edited by - Abramski on August 13, 2000 1:43:11 AM

#55 dwarfsoft   Moderators   -  Reputation: 1209

Like
Likes
Like

Posted 12 August 2000 - 06:32 PM

gee... how crusty and old was that last post you dug up of mine? It was from like 2 weeks ago I haven''t really posted since then, as most things started going over my head.

Anyway, if it is programmed correctly (the hard part) then it is good (the hard part) if it is programmed incorectly (the easy part) then it ends up being crap (the easy part). It isn''t an easy task to accomplish... I am going to return to lurking now

-Chris Bennett ("Insanity" of Dwarfsoft)

Check our site:
http://www.crosswinds.net/~dwarfsoft/
Check out our NPC AI Mailing List :
http://www.egroups.com/group/NPCAI/
made due to popular demand here at GDNet :)

#56 Paul Cunningham   Members   -  Reputation: 122

Like
Likes
Like

Posted 13 August 2000 - 07:44 AM

If i may but in here for a quick question that''s been on my mind that i''d like to bounce off you peoples. In terms of a game development mechanism, how do we now seperate Intution from intelligence? I''m thinking that intelligence is about learning and intuition is about guessing? That''s my opinion and i think that the computer could be made to guess by balancing up the odds such as time and distance a player is away from the computers home base if you get my point. Where as learning is more about gathering and using information.

I love Game Design and it loves me back.

Our Goal is "Fun"!

#57 Anonymous Poster_Anonymous Poster_*   Guests   -  Reputation:

Likes

Posted 08 March 2001 - 03:49 AM

If you want to know what intuition is, read the celestinion promise (if I spelled it correctly), the first 9 insights.

Do you want to emulate it? You have to play for God/soul team

#58 DWard   Members   -  Reputation: 122

Like
Likes
Like

Posted 08 March 2001 - 06:51 AM

OK, it looks like it''s about time to try to fuse everyone''s best ideas into one stagedy and call it my own...

What IMHO would be the most efficient way to implement a system that learns, using some form on fuzzy logic (or intuition) would be this:

Start the AI with a general stragedy template. This is no different from most games today. Next, from this general AI, it passes commands down to the Commander AI level (let''s say that there are n commnaders), they have a preset way to deal with these high-level commands. Then the commanders will pass mid-level commands to the units. The units have a template of responses as well.

Each of these levels of input -> responses/subcommands is stored in script.

Whenever there is excess processing power (perhaps between levels), then starting with the lowest level, they process a NN with their new inputs and come up with some generalizations/observations. These get passed up to the next level of command. Then, command would pass changes to that lower level''s scripts.

Here''s how that would work. You have 3 harvesters in a field. They get attacked by a tank, which kills two of the tanks. The third tank escapes due to a misaimed shot. The surviving harvester returns to base and runs it''s debriefing while being repaired (if injured, then get repaired is a predefined script fo the Harvester). It comes up with the observation that if a tank appears then you''ll lose 3/4ths of your harvesters (2 dead + 1 injured). That observation gets queued up for the harvester''s commander.

At the end of the level, the commander get the time to go through it''s queue of observations and sees the one about Harvesters getting killed by Tanks. He in return issues a new command to all Harvesters: If you see an enemy tank, then head home. The commander also passes up a message that the enemy tanks can do at least 20 HP damage (the HP of the harvester).

The Game AI sees the mesage about the tanks capabilities and tells all commanders that they need to increase the armor on all units. The next level, the game AI may realize that it''s had an increase in killings of it''s light units (like scouts), and order their armor lessened to increase their speed.

This could limit your loss of processing time by using the NNs, by using low stress time to do that processing. Then dynamically determine when there''s not a lot going on.




Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS