I have an idea that will allow us to make a true AI programs for a robot that can do everything that we can from feeling and seeing things, to making decisions. I started my posts at the artifical imagination forum, and have now moved here because they are trying to do something different now. If you could take a look at what I have posted, and tell me what you think about it, I think that we might be able to make a base program for a "True AI".
My email is ruai@comcast.net
True AI
Started by zzzomed, Mar 27 2002 10:02 AM
81 replies to this topic
Sponsor:
#2 Staff Emeritus - Reputation: 1668
Posted 27 March 2002 - 11:18 AM
You should summarize your views here. We''re not prepared to dig through other threads to find your opinions.
[ GDNet Start Here | GDNet Search Tool | GDNet FAQ | MS RTFM [MSDN] | SGI STL Docs | Google! | Asking Smart Questions ]
Thanks to Kylotan for the idea!
[ GDNet Start Here | GDNet Search Tool | GDNet FAQ | MS RTFM [MSDN] | SGI STL Docs | Google! | Asking Smart Questions ]
Thanks to Kylotan for the idea!
#3 Members - Reputation: 864
Posted 27 March 2002 - 01:15 PM
zzzomed, it is not my intention to rain on your parade, however your post in the Artificial Imagination thread - which I believe you are using as the basis for this thread - has innaccuracies and errors that should be addressed before the discussion of a True AI continues.
Yes, these are autonomic functions. Most forms of mammalian life display innate autonomic functions from birth as well as the ability to learn autonomic functions through repetitious action.
This is simply not true. There are many examples of animals displaying intelligent behaviour based on original thought, rather than copying humans. The view you have expressed is one held by certain groups of European and American anthropologists for several centuries but widely abandoned last century. It is certainly not the current view of science nor of many artificial intelligence researchers.
Again incorrect. What sets us apart from other mammals is the size of our neo-cortex and in particular our frontal lobe. These have allowed us to develop complex social interactions and language above and beyond that of other mammals. For instance chimpanzee communities display definite social behaviours and organisational structures and they have a limited vocabularly. However, it is reasoned that the complexity of these behaviours is limited by their relatively small frontal lobe, when compared with humans. This is possibly why homo erectus thrived where other mammals failed. Tight social structures and language help weaker individuals to survive during tougher times.
That is certainly not true of modern artificial intelligence algorithms.
It has been done, both in evolvable software and evolvable hardware, where evolution is gauged by the ability of the agent (be it software or hardware) to reprogram itself to adjust to the changing environment.
Better is usually defined in terms of an individuals personal preferences. What is better acording to one agent is not necessarily better according to another. This is the notion of subjective utility and has been studied outside of computer science since Pascal proposed his famous wager. It is also the basis for an acceptable computational definition of rationality: The Principle of Maximum Expected Utility.
The ''degree of randomization'' you talk about though really doesn''t make sense in the context that you use it. Perhaps if there are two actions with indistinguishable utility (in other words, your agent cannot prefer one over the other) then yes, an aspect of random choice might be appropriate. However, humans don''t behave that way. They are more inclinded to choose the action that leads to the more predictable of the two outcomes. Humans prefer certainty in the absence of utility.
That is certainly not true. There are many algorithms for determining the extent of correlation between objects, based on the (dis)similarity of their attributes. This is not even AI, but rather statistics; although the AI community has developed some very useful and important techniques of its own.
It seems apparent that, like many others who have come before, you have your own ideas about intelligence. That is most natural and to be encouraged. We all, at some stage in our lives, question our intelligence, how it came to be and why we are different from other forms of life. It is natural to generate opinions based on these thoughts. Unfortunately, as is most often the case, these opinions are grounded in naivety and lead to fruitless ends. I am not trying to suggest that you should stop thinking about AI. I am merely suggesting that you should do a lot more reading and gain a more thorough understanding before making statements about intelligence and what a ''True AI'' would be.
This post is not intended to offend or to denigrate you, so please do not feel that I have intentionally done so. I have merely posted the above information to correct inaccuracies before other members of the forum take them on board as fact.
Please feel free to respond and to discuss the above information.
Cheers,
Timkin
quote:
Oringinal post by zzzomed
...snip...
When I though about this, I found that when you are first born, you have certain responses that act automatically (i.e.: crying when you are hungry).
Yes, these are autonomic functions. Most forms of mammalian life display innate autonomic functions from birth as well as the ability to learn autonomic functions through repetitious action.
quote:
Animals do the same thing too, and the only difference (as we know it now) is that animals cannot think random thoughts on their own.
This is simply not true. There are many examples of animals displaying intelligent behaviour based on original thought, rather than copying humans. The view you have expressed is one held by certain groups of European and American anthropologists for several centuries but widely abandoned last century. It is certainly not the current view of science nor of many artificial intelligence researchers.
quote:
None the less, this lack of random thought is what sets us apart from animals.
Again incorrect. What sets us apart from other mammals is the size of our neo-cortex and in particular our frontal lobe. These have allowed us to develop complex social interactions and language above and beyond that of other mammals. For instance chimpanzee communities display definite social behaviours and organisational structures and they have a limited vocabularly. However, it is reasoned that the complexity of these behaviours is limited by their relatively small frontal lobe, when compared with humans. This is possibly why homo erectus thrived where other mammals failed. Tight social structures and language help weaker individuals to survive during tougher times.
quote:
That is probably what all of the computer programs are to day, they are just instincts.
That is certainly not true of modern artificial intelligence algorithms.
quote:
You will need to put a type of coding program with in the program in order to let the computer program itself, or "learn".
It has been done, both in evolvable software and evolvable hardware, where evolution is gauged by the ability of the agent (be it software or hardware) to reprogram itself to adjust to the changing environment.
quote:
The think we need to work on, is the response. How does a computer make a choice? In order to program this, you can''t just have the computer weigh the choices and choose whichever one is better. It has to be able to choose which is better, but with some degree of randomization.
Better is usually defined in terms of an individuals personal preferences. What is better acording to one agent is not necessarily better according to another. This is the notion of subjective utility and has been studied outside of computer science since Pascal proposed his famous wager. It is also the basis for an acceptable computational definition of rationality: The Principle of Maximum Expected Utility.
The ''degree of randomization'' you talk about though really doesn''t make sense in the context that you use it. Perhaps if there are two actions with indistinguishable utility (in other words, your agent cannot prefer one over the other) then yes, an aspect of random choice might be appropriate. However, humans don''t behave that way. They are more inclinded to choose the action that leads to the more predictable of the two outcomes. Humans prefer certainty in the absence of utility.
quote:
The computer AI programs right now cannot determine the relationship between going to a bar or a restaurant. It cannot correlate between two different variables when the only thing they have in common is that they both serve food.
That is certainly not true. There are many algorithms for determining the extent of correlation between objects, based on the (dis)similarity of their attributes. This is not even AI, but rather statistics; although the AI community has developed some very useful and important techniques of its own.
It seems apparent that, like many others who have come before, you have your own ideas about intelligence. That is most natural and to be encouraged. We all, at some stage in our lives, question our intelligence, how it came to be and why we are different from other forms of life. It is natural to generate opinions based on these thoughts. Unfortunately, as is most often the case, these opinions are grounded in naivety and lead to fruitless ends. I am not trying to suggest that you should stop thinking about AI. I am merely suggesting that you should do a lot more reading and gain a more thorough understanding before making statements about intelligence and what a ''True AI'' would be.
This post is not intended to offend or to denigrate you, so please do not feel that I have intentionally done so. I have merely posted the above information to correct inaccuracies before other members of the forum take them on board as fact.
Please feel free to respond and to discuss the above information.
Cheers,
Timkin
#4 Members - Reputation: 122
Posted 27 March 2002 - 01:30 PM
I did exactly this in my youth and it worked well, although it never worked well enough. Since then we have been trying to do all from retro-fractal thinking to preminission-algorithms. The best solution yet to date (what I know of) is the Isil-split method.
You take each node of interest and split it into a tree of potential decisions. Each time a decision is choosen by a function (and not a state) you branch this decision node further. In parallell to this you have to do oblivion on less used decision trees, otherwise you will get filled rather quickly with worthless shit in the brain :=) That is the hard part, because you have to traverse the trees every time you want to obliviate something and it takes time
(sleep and dream, sorting your thoughts).
The ''brain'' will come alive and will no longer act as ''expert'' system and will learn continously until the limit has been reached (hardware and your set limits for oblivion). The system works well even for ''small'' hardware options.
I guess I just revealed some very secret stuff, but I think nobody can trace it back to me
and I really believe that shared super-science is for the best.
Good Luck newbie-ai coders....
You take each node of interest and split it into a tree of potential decisions. Each time a decision is choosen by a function (and not a state) you branch this decision node further. In parallell to this you have to do oblivion on less used decision trees, otherwise you will get filled rather quickly with worthless shit in the brain :=) That is the hard part, because you have to traverse the trees every time you want to obliviate something and it takes time
(sleep and dream, sorting your thoughts).
The ''brain'' will come alive and will no longer act as ''expert'' system and will learn continously until the limit has been reached (hardware and your set limits for oblivion). The system works well even for ''small'' hardware options.
I guess I just revealed some very secret stuff, but I think nobody can trace it back to me
and I really believe that shared super-science is for the best.Good Luck newbie-ai coders....
#5 Members - Reputation: 118
Posted 27 March 2002 - 09:12 PM
Nobody really knows what sleeping and dreaming do, although what you suggest is suspected. There are other plausible theories though.
As for building decision trees you neglect to mention the hard part which is deciding what branch to follow...
As for building decision trees you neglect to mention the hard part which is deciding what branch to follow...
#6 Members - Reputation: 100
Posted 27 March 2002 - 09:50 PM
zzzomed,
Intelligence is a little bit more complex than you have described. Consider all the terms below. They exist in AI literature because they have significance:
Intelligence is a little bit more complex than you have described. Consider all the terms below. They exist in AI literature because they have significance:
- Rationality
- Intentionality
- Belief
- Reactive
- Discourse
- Context
- Microtheory
- Axiom
- Deliberation
- Monotonic
- Nonmonotonic
- Common sense
- Domain
- Planning
- Constraint
- Internal modeling
- Causality
- Truth maintenance
- Meta reasoning
- Learning
#7 Members - Reputation: 122
Posted 29 March 2002 - 04:03 AM
The way I see it is that if you give a robot a program that will allow it to learn or improve on itslef and have its own instincts (needing to be plugged into a wall socket every so often to recharge), then it will evolve in a similar way that we have since we have been around. The will then have a thought process in which it will accomplish the same things that we can (if you tell it to move a bolder, it might move it by itself, while we try to build a contraption to aid make it easer for us). Over all, it will accomplish the same things as we do.
My email is ruai@comcast.net
My email is ruai@comcast.net
#8 Members - Reputation: 100
Posted 29 March 2002 - 06:32 AM
quote:
Original post by zzzomed
The way I see it is that if you give a robot a program that will allow it to learn or improve on itslef and have its own instincts (needing to be plugged into a wall socket every so often to recharge), then it will evolve in a similar way that we have since we have been around. The will then have a thought process in which it will accomplish the same things that we can (if you tell it to move a bolder, it might move it by itself, while we try to build a contraption to aid make it easer for us). Over all, it will accomplish the same things as we do.
My email is ruai@comcast.net
Instincts aren''t good enough. Conceptual knowledge, abstraction, and deliberation are necessary as well.
Does the robot hear thunder outside? Are the LEDs on the clock radio currently blank? Did a fuse blow? Would the neighbor''s house have a wall socket? Would it be appropriate to go over there and use it?
#9 Members - Reputation: 122
Posted 30 March 2002 - 03:28 PM
Gues what?? well There is no such thing as "Random" there is only stuff that is very close to random, And our choices That you said, they are not random, you said we can''t have a computer just wigh the choices and decide what is better?? you say we need to throw in a bunvh of randomness??? well guess what?? we, as humans do not throw in a bit of randomness... its just that our descision making process is so complicated(you might not relize it, because our sub concious does alot of the decision making, and none of our thoughts just randomly appear there, they are brought there, by other stuff, Your brain is just one Big Linked System, where everything is linked and one item leads to another, just every has alot of Information with....
What do you think when i say "Apple" well first i think red, then i think about the general shape, then I''ll eventually bring up "Apple Computers" then Computers, Then Programming and games... the list goes on...
What you think is random is really not it just seems to be, because the process to come up with it is so Complicated.
though a few of your points are valid, not most.
No offense
What do you think when i say "Apple" well first i think red, then i think about the general shape, then I''ll eventually bring up "Apple Computers" then Computers, Then Programming and games... the list goes on...
What you think is random is really not it just seems to be, because the process to come up with it is so Complicated.
though a few of your points are valid, not most.
No offense
#11 Members - Reputation: 100
Posted 30 March 2002 - 05:11 PM
quote:
Original post by Puzzler183
Well actually you are incorrect ExplosiveNewt. There are random quantum functions such as radioactive decay. There is actually hardware that you can buy giving you truly random number using this method, although they are not common.
A valid point, but one which only serves to undermine ExplosiveNewt''s point, which was a whole lot more valid.
#13 Members - Reputation: 136
Posted 31 March 2002 - 12:44 AM
Your definition of Random is the most important thing when deciding if "Randomness" exists.
Most people consider it to mean "cannot be predicted" or "has no pattern"(that''s the dictionary definition), and in that case humans can generate them. (ie there is nothing that can predict what number a person will say if you ask them for one, so according to said definition it is therefore random)
There are, of course, other definitons of "Random" which state they cant exist etc etc...
When arguing about stuff like this, it very quickly gets onto stuff like religion and origins etc, and from there its all opionated and unprovable :/
What i don''t understand is, whats wrong with pseudo-randomness in terms of AI? Why can''t a number generated from a huge range of maths and time be used in a decision making process?
Most people consider it to mean "cannot be predicted" or "has no pattern"(that''s the dictionary definition), and in that case humans can generate them. (ie there is nothing that can predict what number a person will say if you ask them for one, so according to said definition it is therefore random)
There are, of course, other definitons of "Random" which state they cant exist etc etc...
When arguing about stuff like this, it very quickly gets onto stuff like religion and origins etc, and from there its all opionated and unprovable :/
What i don''t understand is, whats wrong with pseudo-randomness in terms of AI? Why can''t a number generated from a huge range of maths and time be used in a decision making process?
#14 Members - Reputation: 122
Posted 31 March 2002 - 01:23 AM
quote:
Original post by Smurfwow
(ie there is nothing that can predict what number a person will say if you ask them for one, so according to said definition it is therefore random)
True, you can''t by any deductive means predict what number a person will say. But... it has been shown statistically that if you ask people for a number between 1-10, a majority is going to say 7. Now this should not happen if we were truly choosing at random. I don''t know what factors are the reasons behind this, but it sure makes you think.
Despite appearences, I agree with you that who cares whether randomness is ''true'' or ''almost true''? For most purposes it''s not going to matter anyway.
#15 Members - Reputation: 287
Posted 31 March 2002 - 01:38 AM
Hull, that sounds very similar to what I've been doing, where each "leaf" of the tree is a process (yes we think processes, not organic data with fixed processes), the tree of course can be infinitly complex meaning it can solve any problem (theorectically), depending on the base language that generates the leaves/branches. Can I get in contact with you somehow?
Edit: I am hoping to write an article on this sometime, after I've made good examples of the 'fractal' ai in use.
[edited by - stevenmarky on March 31, 2002 8:41:09 AM]
Edit: I am hoping to write an article on this sometime, after I've made good examples of the 'fractal' ai in use.
[edited by - stevenmarky on March 31, 2002 8:41:09 AM]
#16 Members - Reputation: 540
Posted 31 March 2002 - 02:54 AM
I was not trying to "undermine" ExplosiveNewt''s point. I thought I made it clear that the hardware to generte truly random numbers was rarely used and almost no one had it. If this somehow escaped you then, don''t flame me for it.
In the future I will try not to contribute to the productivity of other at this forum. Thankyou for the advanced warning.
In the future I will try not to contribute to the productivity of other at this forum. Thankyou for the advanced warning.
#17 Members - Reputation: 100
Posted 31 March 2002 - 06:22 AM
quote:
Original post by Puzzler183
I was not trying to "undermine" ExplosiveNewt''s point. I thought I made it clear that the hardware to generte truly random numbers was rarely used and almost no one had it. If this somehow escaped you then, don''t flame me for it.
No, it''s just that randomness (especially in this context) really has no value (in my opinion). I think ExplosiveNewt made a valid point.
quote:
Original post by Puzzler183
In the future I will try not to contribute to the productivity of other at this forum. Thankyou for the advanced warning.
You will be missed. You have taken what I said too seriously. I only pointed out your remark with reference to ExplosiveNewt''s. I could not possibly have anything critical to say about anything else you''ve said or will say without judging it on its own merits.
#18 Members - Reputation: 122
Posted 31 March 2002 - 08:20 AM
Ok, our problem is definattly our difference in the defination of Randomness... In my defination of randomness there is COMPLETLY No Evidiance of what it will be, and there for if you went over a moment, went back in time and replayed it it wouldn''t be the same number. I Belive that if you have a good enough computer, Caculated the Quantum State of everything in our closed system, you could predict the future with absoubsloute certianty... But it doesn''t really matter, what i was trying to say was that There is no need for randomness in the decision making of our AI.
#19 Members - Reputation: 142
Posted 31 March 2002 - 09:03 AM
In game AI the main point is to be believable.
If copious quantities of calls to rand() help you to do this, then so be it. Of course, even rand() is entirely deterministic. However, it is not deterministic to the player; he does not know the seed. Thus, it might as well be random - whatever that really means.
This brings me to my point: What we want in game AI is a response which is not deterministic given data available to the player. Whether this is achieved via rand(), or by making an FSM which is too complicated to predict, or by using MinMax search to "out-think" the player, doesn''t really matter. If the AI is believable and shows no unrealistically predictable pattern, then you''ve succeeded.
Personally, I am partial to more "intelligent" systems which use data about the problem rather than simple random numbers to do their thing. But, as I said, it''s only the end result that matters. Is it "true AI?" It might be if it can pass the Turing test!
If copious quantities of calls to rand() help you to do this, then so be it. Of course, even rand() is entirely deterministic. However, it is not deterministic to the player; he does not know the seed. Thus, it might as well be random - whatever that really means.
This brings me to my point: What we want in game AI is a response which is not deterministic given data available to the player. Whether this is achieved via rand(), or by making an FSM which is too complicated to predict, or by using MinMax search to "out-think" the player, doesn''t really matter. If the AI is believable and shows no unrealistically predictable pattern, then you''ve succeeded.
Personally, I am partial to more "intelligent" systems which use data about the problem rather than simple random numbers to do their thing. But, as I said, it''s only the end result that matters. Is it "true AI?" It might be if it can pass the Turing test!
#20 Banned - Reputation: 100
Posted 31 March 2002 - 09:29 AM
I''ve given a bit of thought to "randomness" and "chance", and I''ve come up with this definition of chance (the same as randomness):
*Clears throat*
Chance is simply our representation of factors which we have not taken into account.
...which is why it is so useful in designing games. Once we have gone as deep as we want to into the complexities of designing - for instance - a weather system, we just add a "random" number to represent the rest of the factors. Much quicker than taking into account someone sneezing two miles away...
About that radioactive "real random number generator" - I''ve never heard of it, and am therefore not qualified to comment on it, but I would suggest that it''s simply that we don''t yet understand how the decay occurs. So the scientists just say it''s "random".
Anyway, just my 2c...
---------------
#define TRUE 0
#define FALSE 1
//MUAHAHAHAHAAAA!
*Clears throat*
Chance is simply our representation of factors which we have not taken into account.
...which is why it is so useful in designing games. Once we have gone as deep as we want to into the complexities of designing - for instance - a weather system, we just add a "random" number to represent the rest of the factors. Much quicker than taking into account someone sneezing two miles away...
About that radioactive "real random number generator" - I''ve never heard of it, and am therefore not qualified to comment on it, but I would suggest that it''s simply that we don''t yet understand how the decay occurs. So the scientists just say it''s "random".
Anyway, just my 2c...
---------------
#define TRUE 0
#define FALSE 1
//MUAHAHAHAHAAAA!
