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

#1 Paul Cunningham   Members   -  Reputation: 122

Like
Likes
Like

Posted 23 July 2000 - 08:00 PM

I''m trying to get a good explaination on what intuition is or how we could understand it. I''d like to work out how it can be artifically created and implemented. But its had enough just trying to understand how it works. Definition of intuition is the ability to make desicions or conclusions without the use of knowledge on the subject. I love Game Design and it loves me back. Our Goal is "Fun"!

Sponsor:

#2 MadKeithV   Moderators   -  Reputation: 971

Like
Likes
Like

Posted 23 July 2000 - 08:22 PM

quote:
Original post by Paul Cunningham

Definition of intuition is the ability to make decisions or conclusions without the use of knowledge on the subject.




I don''t agree with the definition
So I''ll start by stating my interpretation of intuition:

Intuition is the ability to make decisions or conclusions without the conscious use of knowledge on the subject.


How do you simulate it? Hmm, I''m not sure you can, but you can EMULATE it. That''s what expert systems were designed to do. They "capture the implicit knowledge and intuition of an expert into an automated system."
The problem with it is, that you have to know WHAT you are trying to store intuition about.

What exactly is it that you want the computer to be intuitive about?


Give me one more medicated peaceful moment.
~ (V)^|) |<é!t|-| ~
ERROR: Your beta-version of Life1.0 has expired. Please upgrade to the full version. All important social functions will be disabled from now on.

#3 StrategicAlliance   Members   -  Reputation: 122

Like
Likes
Like

Posted 24 July 2000 - 09:11 AM

I think the important word with intuition is ''association''.
Because you never have NO knowledge at all about something.

Suppose you have some sort of creature in an RPG and you want to give it intuition. At one moment it becomes hungry. If it has never encountered this state and if it would have no knowledge at all it might try sleeping, but it wouldn''t work. Talking wouldn''t work either. Now if this person has some form of association patterns (based on neural nets for example) like

HUNGER ~ FOOD ~ EAT

then it might do something useful. This way you don''t have to program the fact that it has to eat when it is hungry. You only say that FOOD is associated with HUNGER, something we do a lot every day. The exact actions might be going to a bakery and buying food, then eating it. Again you might build a pattern like

STORE ~ BUY ~ MONEY ~ FOOD

After time the creature would be sufficiently trained once it has found the correct associations. (The intuition is verified and it becomes knowledge).

Maybe my explanation looks a little cofusing but I hope you get the idea.


******************************
Stefan Baert

On the day we create intelligence and consciousness, mankind becomes God.
On the day we create intelligence and consciousness, mankind becomes obsolete...
******************************

#4 avoden   Members   -  Reputation: 122

Like
Likes
Like

Posted 24 July 2000 - 01:36 PM

In the example above when creature is hungry it may try as well eat itself. In that case that whould be correct intuitive solution except it will create new problems to solve:pain and missing part of the body. I think intuition can''t be effective without experience, experience without mistakes and ability to learn from them.
If, to be more specific, I have RPG creature which has n possible actions to choose from (some of them it never tried) and creature given some task, it can: see if it ever tried to solve that task before and if not try to go through every possible combination of actions to see if that whould solve a problem. It similar to game of chess where the easiest solution whould be to go through every possible move and choose the best one. Of course that whould take long time but here is fun part where you can programm AI to make judjment. May be like that:
does trying action will give similar result as if trying actions[k] + action[j] +.... In other words it can try to group things in categories.

I have no words of wisdom.

AI v.2000.07.24.17:36


#5 Paul Cunningham   Members   -  Reputation: 122

Like
Likes
Like

Posted 24 July 2000 - 07:38 PM

The goal i''m tring to achieve by using computer intuition is more a solution to the problems arising from using conventional ai routines for games. The primary problem I see with the current ai system is: That the player in "A" game will eventually/usually adapt strategies and skills which are more derived from weaknesses and predictabilities in the computer ai rather than using his/her creative and logical smarts to beat the computer in a game.

Although i am mainly speaking from my own personal experience. The one problem i have with playing games which rely on computer ai is that once i realise a way to beat the computer then i know that i can always beat the computer at this game. So i''ve come to the conclusiion that the AI in these games relies more on evironmental forces to make there decisions and less to none on their internal forces (computer intuition).

To wrap up, i''m tring to work out a way that the computer will behave differently / unpredicatable which each new start of a game rather than just being beaten because of the computers ability to calculate faster than me.

I love Game Design and it loves me back.

Our Goal is "Fun"!

#6 MadKeithV   Moderators   -  Reputation: 971

Like
Likes
Like

Posted 24 July 2000 - 09:27 PM

What you could do is give the entities in the game an "alphabet" of basic actions. These are probably very similar to what the player has, things like "Move, Take, Use, Drop". Then you might have some basic "words" with definitions such as "Eat[ Use Food ]".

To generate new kinds of behaviours you might train the entity to try out different things, and it could ask you to name them if you consider something good... or it could evaluate its own actions and keep learning during the game, but then you''d have to set up an evaluation system as well, that could be based on pain or something.


Give me one more medicated peaceful moment.
~ (V)^|) |<é!t|-| ~
ERROR: Your beta-version of Life1.0 has expired. Please upgrade to the full version. All important social functions will be disabled from now on.

#7 avoden   Members   -  Reputation: 122

Like
Likes
Like

Posted 25 July 2000 - 06:29 AM

Since we talking here about game with hardcoded rules, there is not much space for intuition but rather computer has to have library of various strategies. Strategy should be based, like some wrote above, on the basic actions. Also you need satisfaction criteria - how id chain of basic actions more successfull then other. Lets consider example:
Unit A has no personal experience buffer, library of precoded actions (set by programmer) and some formula for calculating its success (some units have different goals and thus different formulas) in range from 0.0 to 1.0. Unit A tried randomly or selected scenario from library (say here is chain of actions: A,B,C) and achieved satisfaction result of say 0.7. Unit A writes that chain of actions with satisfaction result in experience buffer. If human adopts then satisfaction result will eventually fall down to say 0.5. Then Unit A will start looking for another strategy, itterating process. On top of that should be another AI spesializing in processing of existing experience buffer.

Cheers

#8 ahw   Members   -  Reputation: 262

Like
Likes
Like

Posted 26 July 2000 - 09:42 AM

Dictionaries are wonderful sources for reflexion. Intuition means in LAtin "to look,examine carefully".
Paul, you seem to think that ituition is sixth sense that give you the answer to a problem. you seem to imply that it should come "from the inside".
I''ll agree partly on this. According to the dictionary again, intuition is a mean of understanding/solving a problem by sheer empirical method. That is, without using reason, logic. You''ll agree with me that it''s a bit the opposite of what a computer is.
So I was gonna stop my reasoning here, but I thought that maybe we could emulate this somehow.
I agree a bit MadKeith in that intuition is a non conscious phenomenon if you will. We find a solution but we don''t seem to know where it came from.
The way I see it, we have 5 senses, we receive continuous electrical stimuli from the outside, our brain is constently processing. But all this raw data that we receive is/has to be filtered. And Reasoning, is such a filter. Instead of reacting to stimuli directly, we add a step of processing, that usually brings a solution without the experience of trial and errors that a non-reasoning species would encounter (a bird desperatly trying to pass through my window to fly outside, instead of reasoning that, if it gets stopped suddendly, for sure there must be an invisible obstacle ahead).
Know intuition is a shortcut to the normal "human" process. The raw information suddenly "connects" its solution without having to go through the normal pipeline. We duck when we hear a frightening noise, we extend our arm to prevent a fall if we are pushed backwards (try not to), these would be physical intuitions, wouldn''t they ?
More intellectual intuition would be found in the domain of human relationship. For instance the way someone says "hello", the way to look at you. The eyes are the mirror of the soul, they say, and I firmly believe in this. You know someone is lying to you, but how do you tell ? Because of the tone of the voice, the way the eyes look at you, you don''t really know, but you KNOW. That''s intuition. You don''t explain it because it''s not reasoning, it''s perception to its best.

Now how do you emulate this ? Well, I hope I have given enough keywords to give ideas to you, but here are mine.

I think the main problem we have is that our AI are done in a way where they just CAN''t have intuition, because they process ALL the data they receive, since they only receive the data they want to process.
What I am saying is that the main problem is that the AI should be listening to the greatest number of things possible. For example, we only "listen" to the position of the mouse cursor, because we only deal with this. But know, if we where "hearing" the position, the speed of the mouse movements, the frequency at which the player clicks, the frequency at which the mouse is moved, the fact that the mouse is moved for nothing, or to issue an order, we could discover things like the nervousness of the player, the impatience (typically, I move the mouse in circles when the system begin to slow down), or the fact that he is doing pretty much nothing ...

We already have the filtering part, the brain, the reasoning. But we don''t have much to filter. Everything is meaningful. If we begin to have more *RAW* data, then we can have things like intuition appearing.
We also need a memory to our AI.
Say we have a very well coded AI for a RTS game. We have some harvesters (resource collectors) in a field, working peacefully. Enemy tanks are passing nearby, approaching, at close combat distance, then they start to attack. The superbly coded AI decide to retreat its harvesters and call in support. OK, I think we can have this already.
Now if we add intuition. The next time the tanks pass nearby, or any enemy for that matter, the harvesters AI, if it had access to data such as the direction, the apparent level of aggresivity of those units, the nature of those units, would intuitively deduce an incoming attack (association of the "ENEMY" concept with the "DANGER" concept) and either flee (the direction of the attack is towards them), or warn commanding units, passing the direction and nature of the enemy units (the AI would then decide upon an appropriate answer to the attack).

I think this is a bit what avoden offers in his first answer, actually. Make association when the create consequences (hence the need for a memory), and try new association when you have the time/opportunity.

As well, you need rules to make correct associations between concepts.
The association of DANGER and ENEMY would provoke the civilian units to run away. Which is an intuitive answer. But a soldier shouldn''t !
Maybe follow the "pyramid of needs" system (I forget the author''s name, Chomsky ?). First a human has to survive. Then he has to feed, which helps survival. Then it has to find shelter (it helps feeding, if you have a fixed source of food nearby). Then there are other need but I forget them. The last one being the need for self accomplishment.
A soldier will prioritize the protection of his nation rather than its survival. Of course, morale issues can change this behaviour drastically.
But I''ll stop going off tracks now... I hope my ideas are a bit helpful. At least, this thread just gave me new ideas

youpla :-P

(ps : I know I am a bit lengthy, but I believe I answer your questions Paul ... well, I hope)

#9 Paul Cunningham   Members   -  Reputation: 122

Like
Likes
Like

Posted 26 July 2000 - 07:15 PM

I was going to write a reply yesterday to this but everything i wrote was crap, don''t you hate that. Ahw, i couldn''t have wrote a better post than your last one i wish it was mine.

What about treating Artif-Intui like a plugin. We know the goal but to make it happen maybe we could develope plugins to achieve it. Could the first plugin be "Deduction"? The controlling logic that handles all the plugins could revolve a "Relativity logic" which could work such as: The "Deduction" plugin is used in "Relative" to the "Size of the computers army"?

The more plugins the better the artif-intuition.

I love Game Design and it loves me back.

Our Goal is "Fun"!

#10 ahw   Members   -  Reputation: 262

Like
Likes
Like

Posted 27 July 2000 - 06:10 AM

quote:
Original post by Paul Cunningham
I was going to write a reply yesterday to this but everything i wrote was crap, don''t you hate that. Ahw, i couldn''t have wrote a better post than your last one i wish it was mine.



Now, THANKS =) *blushes*

But I have to criticize your proposition. Well, maybe I am just being picky. But as I said, Intuition by itseld IS the basis, then Reason (logic, whatever) IS the plugin !

What we have at the moment is :
-filtered inputs
This prevents us from having intuition because it prevent us from having things (data) we don''t expect.

what we need is more input data, and a better way to filter it, and a mechanism that can do relevant and fast connection of ideas.
The decision to switch to Reasoning comes when nothing conclusive shows up during the low level (instinctive level if you wish). Then we make complex association of ideas, which takes more time, but can be more directed.

Maybe if we take the example of the harvesters again.

Our harvesters are on the field for the first time. When they see enemy units they get info such as "appearance of units", "heading", "speed", "distance". Which by themselves don''t mean anything.
Trying to connect Unit(unknown) with anything won''t be very meaningful. So the harvesters ignore those units. If those units come and attack, an association Unit(some very specific id)->Danger is made.

Fleeing in front of the incoming enemy is an Intuition for the harvesters, that they''ll refine after a while to include a set of ID (all enemy units, not jsut the ones that attacked).
Just switch to Runaway mode or something.

Fleeing and calling in support or transmitting the units positions is a Reasoning process, it takes a bit more time.

Maybe intuition is just a match of 2 ideas provoking a reaction.
While reasoning is a match of more ideas/concepts (ARTILLERY ENEMY I_AM_CLOSE_COMBAT_UNIT ARTILLERY_CANT_FIGHT_CLOSE_COMBAT) -> ATTACK.
using more complex operations such as induction, deduction, and other basic logic statements (OMG, my Maths lectures ARE useful ! )
RED_TANK -> DANGER, RED_TANK -> ENEMY, hence ENEMY -> DANGER.

Does that make any sense.

I am just thinkin that you are right Paul. We can make intuition be a plugin.
Vanilla AI jsut hear what they want to listen to. Which is very effective, but non-creative.
Add aplugin that listen carefully to much more data, make fast connections between ideas and make a sort of database of ideas, and there you go, you can have a more "creative" AI.

In fact, up to now our AI was Intuitive because it only connected data and executed actions, jsut like reflexes. Very branched, but reflexes nonetheless.
When we add the "concept matching" process, we get a creative AI.
Hence a Reasoning one.

I have to save that post, I am getting idea for my project

youpla :-P

#11 Paul Cunningham   Members   -  Reputation: 122

Like
Likes
Like

Posted 27 July 2000 - 08:28 PM

This is definitely my hardest debate yet. But to keep this on track i''ll stick to your havester example.

I think if we''re truely tring to create a very artifical intuition then what we are tring to achieve is allowing the computer to have instinct without prior knowledge. A kind of wisdom if you like. So for these harvesters to have intuition then they would have to make desisions before learning. All desicions made there after would come under normal AI.

What the game would have to do in my eyes is make some sort of profile on you the player. This hidden knowledge in Ram could be used as a way of instructing the computers units to make choices that the player would be unaware of, giving the player the feel that the computer is making presumptions, and intelligent ones at that.

Plugins would come in as i see it when the player saves his/her progress. That being when the player saves, their profile is saved with the file.

But this now creates the new problem of how a player profile is made. Should it be a psyc profile of the player which holds values that can be used relivently to the game? Your saying ahw that the plugin should be the logic/yes? So the profile is the logic. The logic plugin would then holds spaces for variables that can be altered due to the players nature. A measurment of the players aggressiveness, forgetfulness, and the like are placed into the control logic variables. Using this control logic will then make the computer units predict outcomes from player maneoveurs.


I love Game Design and it loves me back.

Our Goal is "Fun"!

#12 ahw   Members   -  Reputation: 262

Like
Likes
Like

Posted 28 July 2000 - 03:28 AM

Hello again.
I read my previous post, and indeed it''s a bit confusing, I was thinking while writing down, and it appears that my ideas changed during the post.
But that''s the fabulous thing about dialoguing, it makes you change your ideas

So I''d like to refine a bit what I was thiking about.

First of all, a "classic" AI is an AI that listen to a limited set of data compared to what is available, and makes decisions in a branched way (purely conditional, IF "this" appears, do "that").
You seem to like the idea of a plugin kind of AI. A basic one would be what we have today, and is NOT creative.
What we want is introduce Creativity in the mind of our AI.
To do this you need to introduce more data.

So the first step would be to listen to everything that is going on on the battlefield.
Then our improved AI would listen to everything (rather than a taget subset, like the normal AI), and make analysis on this data.
What we would be doing really, is some sort of pattern recognition. Make connection between events, state of the game, and other events; between concept and other concepts;
To build those patterns we need some sort of grammar. We need the tools offered to us by Maths Logic, boolean stuff, logic stuff.
Having those tools we can build reactions that improve over time, as more and more data is collected, and various situations produce various behaviours.
We add a Learning capability to our AI, which is the basis for what is to come next.

At this stage, you see that we need a way to store all this data, and to create the behaviours that results from situations collected. We need a scripting language, and the ability for the AI to write those scripts !

The next step is to add creativity. The first obvious way is the natural one : Trial and error. Very innefective, time consuming, but exhaustive. As well we could make a better analyst program.
A simple analysis is "Enemy units cause Harm, Harm must be avoided, hence Flee when Enemy units are seen".
With a good analyst program, if you give it an observation such as : "Enemy tanks take more damage at the back" it would come up with a strategy such as "Attack from behind enemy tanks" rather than a standard "Flee if enemy unit" behviour.

Well .. I think I am just saying the same thing over and over ain''t I.

So as you see, we need some sort of massive data collection / analysis. Doing this in a very ehaustive way during the battle seems out of the question. And the solution I can see is again taken from Real Life ™. During a battle you don''t do very complicated problem analysis, because it''s hard to think under fire, because Panic is a VERY problematic thing (more about that later), and so on.
That''s why we have debriefings !

Now imagine this (and I am rejoining some of your ideas here Paul).
You battle it out against the computers, you get ridiculously tricked into an ambush during the game. At the end of the game, the analyst AI doesall it''s data analysis, store its conclusion into a personal database. But it also analyses your battle, and shows you its conclusion : "Tanks without Close Combat units protecting them are very weak" or stuff like that. Yes, a good player won''t really need this, but imagine for the beginner player ! That would be a fabulously useful tool (remember the Counselor in Dune 2 ?).

Another thing I was thinking about was how to make personalities.
I mean, it''s nice and all to have a learning AI, creative as well, but how do you make the AI get new ideas, do you make it randomly or what ? I don''t really like the random choices ideas. Hannibal didn''t come up with the Pincers maneuver just because he tried it and it worked. It was part of his personality. He observed his enemies, saw weaknesses, then made choices taht lead him to try this new maneuver.
So we would have a single AI system, analysing, collecting data. But we would have different Commander files, that would contain the knowledge of different player, either virtual or real. Each with their own favorite tactics, their weaknesses (easily panicked, quick to anger, always trying to use his favorite unit, never willing to sacrifice his units, etc). Temporary data could be added, to simulate objectives for a single game (blow up this bridge at all cost, keep the enemy busy for some time then retreat, etc).
(All this should converge with yor idea of profile)

As well, I''d really like to see Panic, Morale of troops take a bit more importance in RTS.
It''s not that off topic really, because what panic and morale have to do with Intuition is that they change the way we use our brains.
My normal soldier obeys the AI orders, but when he panicks, the only thing he listens to is instincts, either try to take cover and stay there (shock state), either run away like hell.
Being with a great number boosts morale, but when things go wrong, it usually provokes mass reactions (crowd effect) rather than clever reaction : a massive crowd of orks attacks with bravery, but when the units see half of their men be killed, they all emulate each other and break formation, usually running away. A small group of elite soldiers loosing half of their men will retreat, tryingto take cover, and retaliating any attack to prevent more losses.

Anyway, it''s getting harder to think with my wee sister asking me to go play outside, so I''ll leave it for now.

I''d love to see new ideas on this post, anyone want to give it a try.
Also maybe try to generalize a bit, I think i am focusing to much on RTS here.

youpla :-P

ps : I think you have to stop thinking that Intuition is something different from Reasoning.
I believe that it''s just the same thing. Intuition is just a word to say that our brain found a solution without having to make complex thinking, and sometimes without us having the knowledge of the elements of the problems (what is it that makes you attracted by this woman and not that one ? there IS a rule, we just don''t understand the process).



#13 avoden   Members   -  Reputation: 122

Like
Likes
Like

Posted 28 July 2000 - 09:42 AM

Things are getting more specific in that thread
Id like to step back a bit. I think that intuition that we discussing here (based on the popular harvester sample) is not an intuition at all. Hannibal used Pincers maneuver not without learning a lot, thus making it simply a rational desision. Humans have awareness about sertain things and thinking is one of them. If I made move in game of chess, I judged (thus aware of) it on 2 moves ahead, but assosiative memory such as human brain does not require you to be aware of the thinking process in order to make desisions. So as result I could possibly made move 10 moves ahead except I wasn''t aware of it. So what we are discussing here is not applicable to computers since they are not selfaware. I read about monks and meditators "observing" their own process of thinking and they saw how many thoughts passing through their brain. I can describe thinking as: when I think I consentrate on 1 thought and trying to see if my brain give related desision. Its almost as if Im tuning to that thought and following all desisions made by my brain. We simply can selectively handle observation of the though or 2 but we can''t imagine how many little desisions we make without knowing it. We think we controlling brain? I see so many ppl on the streets who controlled/obsessed with one thought in the brain rather then the other way around. (Gotta stop here or Il drift away from the point )

The intuition that referenced by those monks is not related to the data analisys at all. They say it is an ability to tap into universal pool of knowledge. In fact it is required to stop thinking in order to gain access to that knowledge (what a controversy). After that you just know certain things. I dont see computers in the same category.

Cheers.

#14 Cadet Random   Members   -  Reputation: 122

Like
Likes
Like

Posted 29 July 2000 - 11:39 PM

Intuition is Pattern-Matching based reasoning.

Logic is Sequential-Processing based reasoning.

Intuition takes a static timeslice of the gamestate (possibly containing some simple time component like velcoity) and analyses that.

It''s the same system which emotions run on (well, that''s one opinion).

Neural nets would work nicely for pattern matching (it''s their bag, baby) but whether they can be implemented sufficiently large in a game...

Just some random thoughts,


Sam
Space Cadet''s DJ Random, Thinker, and Geek at heart.

#15 ahw   Members   -  Reputation: 262

Like
Likes
Like

Posted 30 July 2000 - 03:44 AM

Jeez, it seems I completly confused you avoden.
OK first off, I agree I am getting specific, but it''s to help give example, I still think in a general way.
I''ll try to reformulate my thoughts :
if you will, intuition and reasoning are the same. what happens is, you receive data, process it, generate a result/action.
Now the difference resides in the level of complexity, intuition is a sort of fast-processing, while reasoning "is called"/occurs when the intuition encounters a problem too complicated for it.
example : I am novice typer. I have never seen a keyboard. Intuitively, I understand that letters on the keys are the letters that are going to be typed. I have to look for the position of each letter on the keyboard everytime I want to type something (this is a conscious process, not intuitive if you will). And to type a "!" (on a US keyboard) I have to solve a problem ("there is a "1" and a "!" on this key, but how do I switch between the two) which is very much a reasoning process, not intuitive at all.
After some practice, I don''t really look for the most frequently used keys anymore, typing them has become an intuitive process, where I don''t consciously have to look for the key. Once in a while, I have to type a character which is not vey frequent, and have to stop to look for it.
Is it a clear enough ?

Now you say "I would be playing 10 moves ahead and not be aware of it". Either you are a genius, because it would mean your brain contains intuitive knowledge of so many situations that you have just "recognised" a situation and played the move, knowing the consequences 10 moves ahead... NO WAY on earth that can happen Or you are lucky, which is not intuition. Not in the case of chess, that is.
As for Hannibal, I don''t say he intuitevely came up with the Pincers during the battle of Cannae, I am just saying, that given the battle (I think the decision for the maneuver was made BEFORE the battle, which emphasizes my point that we should have somme sort of briefing and debriefing in RTS) Hannibal would have perhaps come up with another solution, but his personality made him choose this. As well, there are few other general that would have crossed the Alps with elephants, or crossed the Rhone (a french river), again with all the elephants, the way he did. Not that I am a specialist of Hannibal, its just that is he''s quite famous for a reason isn''t he. Because he was different from other generals, which lead to the question, how do you show this in game ?
For instance, after WWI how do you have some french generals deny the use of planes and tanks in warfare (considering them useless !!! And they actually were during WWI) and others like DeGaulle prophetising that the use of tanks would become a standard (no one listened to him ... except maybe Hitler :/ ). Was it intuition ?

To conclude, intuition is not data analysis neither. I just see it as a matching process (one operation) between and idea and the matching action to be done. If no match occurs, then the reasoning process takes place, trying to match ideas and actions in more complex ways, not existing already somewhere in the brain.
Intuition is just what what we already know, sometimes not knowing that we know
The data analysis I am talking about would be a process taking place at the end of the game, trying to analyse everything that happened, to create new patterns (that would now become intuitive). The data collection would occurs a bit like the Spy++ tool of Visula C++, listening to everything that occurs ingame and storing it for later use.

Youpla :-P

"Act without doing; work without effort." zen saying

ps : Space Cadet, this idea of using NN is good. As for the processing time issue. As I said somewhere in all my rambling, we can limit the "depth of search" during the game, and make an exhaustive search AFTER the game, during a debriefing phase.
After all, we don''t mind waiting a game to load, so why not wait a little bit as well after a battle ?

#16 dwarfsoft   Moderators   -  Reputation: 1210

Like
Likes
Like

Posted 30 July 2000 - 03:54 AM

That was a breath... Good post, nice topic cover. But how are we going to implement the NN into this (Ok, so I haven''t had much XP with NN''s). I think thought that I will have to say that intuition could be what you have said it isn''t and it may be something else all together. I think intuition is something like deductive reasoning. You see something then you imply the rest (oooh... that set notation and truth tables including x->y is looking very interesting all of a sudden). The example given before was if NPC X saw NPC Y with Enemy Z then NPC X deduces that Y is a nasty SOB and needs to be avoided at all costs. Simple but effective example I think.

I am wondering though if this is the kind of Intuition you are talking about or this deduction is something different alltogether?


-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 :)

#17 Paul Cunningham   Members   -  Reputation: 122

Like
Likes
Like

Posted 30 July 2000 - 07:11 PM

I really like the idea that intuition is based of patterns. I''ve had the thought that the subconsious mind recognises a pattern and sends a signal to the consious mind alerting it. The consious mind whilst recieving this signal has no idea how it came to know the knowledge included in the signal.

So if this was to be used in computers then the computer would have to recognise patterns (which i think has already been explained) instead of calculating a reaction.

Excuse me for repeating things you people have already said. A pattern (once recognised) would have to be given a value or purpose interpretation. This will allow this computer forsight in understanding it''s own destination if it doesn''t make changes when it recognises a pattern player. Which is what i''d really like

I love Game Design and it loves me back.

Our Goal is "Fun"!

#18 Cadet Random   Members   -  Reputation: 122

Like
Likes
Like

Posted 30 July 2000 - 09:14 PM

To implement NNs in a given application, you need to define 2 things:

(1) an appropriate input vector
(2) an appripriate output vector

A further thing you need in NNs is a set of training pairs, with the "good" input-output pairs.

This wouldn''t be too much of a problem in "the lab", but, if you want it to learn "in-game" (or between missions or whatever, basically take note of what the player''s done), you need a method of deducing the "right" response for a given situation, in hindsight.

The method for doing this really depends on the contents of your output vector. If it''s a simple warning bell type thing, then it wouldn''t be too hard - any unit that dies, take the last input-vector it had, and have a "run away" out-vector. Take any enemy unit that dies, and have an "attack" output-vector.

That''ll do for now.

Sam
Space Cadet''s DJ Random, Thinker, and Geek at heart.

#19 dwarfsoft   Moderators   -  Reputation: 1210

Like
Likes
Like

Posted 31 July 2000 - 03:01 AM

Just make sure that you have some kind of template you are basing all learning for the NN on. Then you can apply different Templates to different situations to get required results... Please don''t ask me to explain this at the moment, as I am half asleep as it is


-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 :)

#20 Paul Cunningham   Members   -  Reputation: 122

Like
Likes
Like

Posted 31 July 2000 - 04:43 AM

Erk, you people are probably all programmer''s yes? and very good one''s at that. I few terminologies have gone over my head here but i would really like to know what was said since my previous post. If you just explain input and output vectors then maybe/probably i should be able to understand the rest If you could just give and example and point me to an output vector then i should understand, i hope.

I love Game Design and it loves me back.

Our Goal is "Fun"!




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