Archived

This topic is now archived and is closed to further replies.

SFA

Goal : Reproduce Stone Age Behavior

Recommended Posts

SFA    122
Hi. I''ve just started my Ph.D in Computer Science (specifically the fields of graphics and AI), and after researching for going on 4 months now I''ve narrowed the focus of my research to have a long term goal of being able to reproduce stone age behavior in software (i.e before the advent of language to communicate). My supervisor (himself a doctor in AI) has advised me to be very ambitious and try to eventually create an ''Intelligent'' Agent in a virtual environment that has the ''strong'' agent qualities of rationality, learning, goal-orientedness etc. In his own words : i.e. the goal is to develop an I.A. that "thinks about" and reacts to its environment similar to humans. Then how does changing that environment change its behaviour? i.e. you could have it require food, clothing and shelter etc Now I''ve been reading a lot into emergent behavior, genetic algorithms, neural nets, fuzzy logic and so forth - but my main problem is that I can''t decide which style of AI representation(s) I should focus on first. Perhaps none of the above could really be used to develop so called ''strong'' agent qualities...or perhaps using them in conjunction could yield better results? I was just wondering if you guys had any comments, suggestions or recommendations for me as I start out on the long journey towards being a doctor! Any help would be great. Cheers, SFA.

Share this post


Link to post
Share on other sites
orbano    130
i think neural nets can be used to reproduce any kind of artifical intelligence/behavior. its just a matter of complexity. i think you should think about this, and the right ways to implement an evironment to teach your agent.
i hope you will succeed, im also planning to specialize my studies for AI an graphics, but i have a few more years left at my uni.

"Knowledge is no more expensive than ignorance, and at least as satisfying." -Barrin

Share this post


Link to post
Share on other sites
Infuscare    139
I''d personally probably use fuzzy logic for states like hunger, temperature, need to go to the bathroom, etc. Then feed that into a neural network and train to be what you define as being a leader.

I''d also just like to point out that I personally don''t have much experience with AI, and that this is off the top of my head...

dont flame me!

_________________________________________
"Why, why why does everyone ask ''''why'''' when ''''how'''' is so much more fun"
-Spawn 1997

Share this post


Link to post
Share on other sites
Guest Anonymous Poster   
Guest Anonymous Poster
Stone age civilizations had language...even monkeys have vocal communication...

Anyway, it might be more important to focus on the problem representation instead of how you are going to solve the problem. If the agent requires food, clothing, shelter, how does it get these things? How complex is the simulated world that it will have to navigate? Then to behave rationally, the agent has to take its observations and choose actions...genetic algorithms/random searching could be used to evolve an AI that survives well in the environment, of course there also needs to be a representation for the AI... it sounds like it should have some sort of memory, to remember what to do in different conditions as well.

Share this post


Link to post
Share on other sites
Timkin    864
You should definitely go out and get a copy of the game Creatures (I,II or III) and read up everything you can find online about Norns. Then hunt down their creator and read anything you can about his research work.

Norns are the little creatures in the game that learn and interact with each other and the environment. They can actually teach each other and learn to perfom a whole range of tasks in their environment. They have desires and needs (like food, shelter, sex, etc) which are governed by their physiology. They actually have a simplified neurochemistry which alters their perceptions and reactions and which is affected by their activity, food intake, rest, sexual activity, etc. All of the biological information in the norns is encoded in their chromosomes and when they mate, they actually pass on genetic information to their offspring, so the norn race actually evolves over time as well.

It sounds like this is the sort of ecosystem you want to achieve. How you implement it is up to you. It''ll almost certainly require a combination of AI tools and you''ll need to do some research on basic physiology.

There''s plenty of literature around on artificial life simulations... Norns are just one example of them.

Good luck with your research... it''s going to become hard, tiring and you may get depressed about ever finishing it... but it''s all worth it in the end. The sense of pride you have when you look at your thesis is very special.

Good luck,

Timkin

Share this post


Link to post
Share on other sites
ahw    264
quote:
Original post by Timkin
Good luck with your research... it''s going to become hard, tiring and you may get depressed about ever finishing it... but it''s all worth it in the end. The sense of pride you have when you look at your thesis is very special.



Gee, you dont know how familiar that sounds...

Anyway, SFA, one thing I know is that you shouldnt let the tools dictate which way you want to go.
See, the thing is, they are TOOLS, which mean they get used.
A Neural Network is not some sort of panacea that will solve all your problems, nor should you try to think of everything you try to solve in terms of "how can I use a NN to do this".
Think about what you want, think about the tools at your disposal, THEN choose the right ones for the right job.

And dont limit yourself to AI reads. There are a lot of things out there, and sometimes, ideas come from far places.
Read Richard Dawkins, for instance, if you want to see his point of view about the origin and "goals" of chromosomes.

Read Jean M Auel''s saga if you want a deeper (and verbose, and researched) look into stone age behvaiours. There isnt enough room here to tell you how much details she puts into her books, but if you are interested in Nehandertals and their cousins Cro Magnon, you ll want to read her work.

anyway, you are just at the beginning of the journey, so good luck, and hang in there !


Sancte Isidore ora pro nobis !

Share this post


Link to post
Share on other sites
JD    208
You probably want to accelerate the system thus you need to intervene in the simulation to create situations that aid your creatures to develop skills faster. This AI is possible to do but it will be exhaustive effort with moral implications. I don''t like the moral issues that will come out of this. You can accomplish anything that you dream of but watch out for the consequences of your actions.

Share this post


Link to post
Share on other sites
Guest Anonymous Poster   
Guest Anonymous Poster
Uh, what exactly are the moral issues involved with applying domain specific knowledge to training an AI agent?

Share this post


Link to post
Share on other sites
Raghar    96
Have you listened about Chomsky? From what age would be your simulation? A Stone age is a very wide term.

It might be funny to try a normal life simulation.
Imagine presidnet in his office and you install a cooling system. How would he react? Shiver, yell at others, use secretary.
And what about someone that is more used to frequently changing environment? Shiver, use more clothes, use his wife, shoot down the cooling system becose his wife was too cold.
Alas system that is simullating behaviour is disapoiting without simulation of myths.

quote:
Now I''ve been reading a lot into emergent behavior, genetic algorithms, neural nets, fuzzy logic and so forth - but my main problem is that I can''t decide which style of AI representation(s) I should focus on first.


So you are in situation: You read alot, however you don''t know how to use that together?

Share this post


Link to post
Share on other sites
pentaphobe    122
Steve Grand (the inventor of the guts of "Creatures") wrote a book called Creation: "Life and How to Make it." (http://www.amazon.com/exec/obidos/tg/detail/-/0674011139/qid=1075709610//ref=pd_ka_1/104-9853923-9506312?v=glance&s=books&n=507846)

It doesn''t go deep into code specifics, but he gives a nice overview of the philosophy (important in this case) and design concepts behind creatures.

His website has documents that go into it a little deeper (http://www.cyberlife-research.com/)

Although the creatures were quite simplistic, they were quite a raw representation of the inner workings, so you could provide additional layers of filtration on the world data in order to come closer to the "Cave Men."

Well worth a look, try to read his non-technical documents as well, it helps in some way..

Share this post


Link to post
Share on other sites
quote:
Original post by SFA
Hi.

I've just started my Ph.D in Computer Science (specifically the fields of graphics and AI), and after researching for going on 4 months now I've narrowed the focus of my research to have a long term goal of being able to reproduce stone age behavior in software (i.e before the advent of language to communicate).


I am assuming that you want to animate this behavior? Is this correct?
quote:

My supervisor (himself a doctor in AI) has advised me to be very ambitious and try to eventually create an 'Intelligent' Agent in a virtual environment that has the 'strong' agent qualities of rationality, learning, goal-orientedness etc.

In his own words :

i.e. the goal is to develop an I.A. that "thinks about" and reacts to its environment similar to humans. Then how does changing
that environment change its behaviour? i.e. you could have it
require food, clothing and shelter etc



This is good. I've been working on something like this for my game for some time from the design standpoint, not the AI POV, however, I have been thinking about these character's behaviors for some time.

quote:

Now I've been reading a lot into emergent behavior, genetic algorithms, neural nets, fuzzy logic and so forth - but my main problem is that I can't decide which style of AI representation(s) I should focus on first. Perhaps none of the above could really be used to develop so called 'strong' agent qualities...or perhaps using them in conjunction could yield better results?

I was just wondering if you guys had any comments, suggestions or recommendations for me as I start out on the long journey towards being a doctor!

Any help would be great.

Cheers,
SFA.



Well, I am not a programmer or much versed in AI topics, but I have a good handle on primitivistic behaviors in modeling interactions these kinds of primitives would have with my avatar/player.

The first behavior I designed for my early humans interaction was with the most fundamental purposes of life itself, survival, so they had constant use of, and continuing articulation with, the sense of smell and sight, to smell or see if anything beneficial or harmful was present or approaching or leaving or paralleling, and this was conjunctive with the flight or fight instinct. They knew if they were downwind, that keeping a low profile would be helpful, and that making sound was a boo-boo so they kept still but alert. Think of men on the battle line and you get the picture.

Since the behavior for survivalistic mentalities is both perceptula in the senses (sight, hearing and smell) and physical (poised to fight or flee, yet, ears forward and nose high for optimal detection, simlutaneous with body hair positioned in the crosswind to detect changes in thermal mass and wind speed/direction due to another proximal physical mass {either the predator or the prey, or another of it's kind not part of the family unit already nearby and in the same position}).


In addition, the defensible location or exposed position is going to effect the degree of alarm in the sensory/physical response, and, the number of other primitives in the party, whether hunters or women, children or elderly, is all going to play into the individual response.

It is tribal as well as individual, so I would think it best to start with fight or flight, detect or be detected, hunt, reconnoiter or threat detection behaviors.

These will all rise or fall, change individually based on the collective environment, are we in a defensible area? Are we exposed in the open, is it half and half, are we a group of males armed and hunting, or a migrating pack or collective packs forming an entire tribe? What age are we in? Extremely primitive, where learned behaviors about predator/friendly/indifferent life form types are new experiences not yet generationally passed effectively or are we late stone age, where we walk taller with skins of pride on our backs because of inherited, learned or discovered skills are slowly making us master of the planet and we gnosis this?

So the quadrangle you are going to want to think about is: what level of sentience/development pysiologically and socially, what environmental familiarity or not, what physical characteristics of the individual (old, young, dominant alpha prime male) and what comprises the immediate proximal societal profile (a hunter alone behaves far different from a party of mixed humans migrating or feeding). Behaviors are going to spawn or emerge from these primary influences. The games we played then were learning tools for challenges to come, like they have always been. This is also true in the animal kingdom to a reasonable extent. The young play practicing fighting for their mature years and rites/priviledges.

In studying this to find engaging and fun ways I could make interaction for a modern consciousness (the user behind the avatars interaction) and the primitivistic NPC AI, I found myself observing human packs in the outdoors from an objective distance. Places like parks on weekends, outdoor backyard gatherings (I found that indoors and collective things like sports events were too socialized to let the primitivistic behaviors we all have so deep inside us to emerge with validity, there was always some sort of mental control going on that spoiled the sample, as it were.)

What I found was that more naturalistically designed playgrounds, with children and mothers, were good places to observe these survially oriented detection and tribal life skill teaching management behaviors, and, I also remarkably discovered that females do this all the time in great sophistication and subtlety. Males will just puff out the chest and stomp around mostly, basically because their tribal behavior is pretty simply defined; protect, provide and inseminate, so the scope is not wide, yet it is detectable, quantifiable and catalogable. Yes, I'm a writer and I bend words alla da time.

So watch the females, they practice this everywhere. Especially when not with the children and when there are men around, not in a gathering, but in a transitory place like a supermarket or bus/rail station. The female is constantly under threat assessment because of men nd their motives.

Were you aware that the first eighteen non verbal communications between a man and a woman were all designed so that the female can determine whether the man is a threat or not? This flies right in the face of the smooth talker theory more than you would think. They possess myriad, subtle and powerful behavioral indications that go back for as long as females have been persued, which in this case, is as far back as reproduction goes, which we can safely say is somewhere on the order of tens of thousands to hundreds of thousands of years old. Thus, it is some impressive bag of tricks, from elusion to evasion to confrontation. Just wait until you see a woman back a man off without anything but body position. Even a hulking dolt gets the message.

Study it long enough, and you will pick up some tricks of body language that will improve your love life in most interesting and rapid ways.

Please let me know how this is going, because I have great use for this in one of my adventure game aspects.

HTH,

Adventuredesign





[edited by - adventuredesign on February 2, 2004 4:58:29 AM]

Share this post


Link to post
Share on other sites
Fruny    1658
quote:
Original post by SFA
I''ve just started my Ph.D in Computer Science (specifically the fields of graphics and AI)


Welcome to hell, you''ll be in good company here.
*Puts head in hands and weeps.*


“Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it.”
— Brian W. Kernighan (C programming language co-inventor)

Share this post


Link to post
Share on other sites
ahw    264
Hey Addy ! Tell me, as a writer and fellow watcher of mankind, did you read Jean M Auel''s Earth''s Children saga ?
I read the books last year and I must say I was at once amazed at the sheer quantity of details she managed to put into the books. It kinda detracted from the literary quality, IMHO, but it certainly made for an instructive read.
anyway, I just wanted to point to it again as I am pretty sure I was largely ignored (the problem of making long posts all the freaking time, eh)


Sancte Isidore ora pro nobis !

Share this post


Link to post
Share on other sites
Guest Anonymous Poster   
Guest Anonymous Poster
Basically there are only 4 things your entity needs:

At least one sense (sight, smell...)
Memory of some sort
Recollection of some sort
Stuff to do with recollected stuff

Once you figure out how to have it ''see,'' ''remember,'' ''recollect'' and ''do,'' youv''e got it made!

Share this post


Link to post
Share on other sites
BrianL    530
What is the difference between memory and recollection?

It seems like you need:

Sensor(s) => Storage => Decision Making => Effector

The sensor basically puts information into storage. Storage can act as a percept history for the object. Based on this information, a decision is made which drives the effector.

Share this post


Link to post
Share on other sites
quote:
Original post by ahw
Hey Addy ! Tell me, as a writer and fellow watcher of mankind, did you read Jean M Auel''s Earth''s Children saga ?


I read a little bit of it. I was swept up in a new script and it was memory intensive at the time, so I tend to ignore everything during those times.

She did go into good detail, which a writer and designer should do - extrapolation to the point of all probabilities considered for the sake of realism or surrealism in a complete, self consistent story world- depending on the theme and goals, but I have been working with this concept for some time, since the late 80''s when, as a project manager for the oldest ranch in california, the archaeologists we hired from UCSB found something that fundamentally rewrote california history, and, I immediately tried to capture it in literature with a radio drama I wrote for The Actor''s Workshop of Santa Barbara.

quote:

It kinda detracted from the literary quality, IMHO, but it certainly made for an instructive read.


It absolutely can, I an regard it more as due diligence research (some writers just need enough words and compositional skill isn''t working for them, so they put silk stockings on the research and trot it out as content), but, it is a vast, often lifesaving resource to draw upon when the story is losing legs.

quote:

anyway, I just wanted to point to it again as I am pretty sure I was largely ignored (the problem of making long posts all the freaking time, eh)


So that''s been my problem.... guess I''m just a novelist in a sound byte world.. Good to hear from you, hope the writing is going well.

Addy




Share this post


Link to post
Share on other sites