Artificial Inteligence

Just out of curiosity, hypothetically, if one were to set out to create a large program for the purpose of mimicking the thought processes of the average human being, What language would you use? It would have to be robust, fast, and have the ability to quickly interface with one or more massive databases. and it would have to have the ability to alter its own progamming. I would think the best way to write such a program would be to seperate it into a large number of small subroutines and functions and link them together with a common interface. What would be the best Language to do this in? I only ask because I don't have much experience in programming.

I'm no expert, but my own fumbling investigations lead me to believe that our thought processes are complex because our brains are multithreaded and referential... in other words, each input triggers pathways to multiple nodes (or pattern clusters) and as the patterns reference each other they become more dominant until they trigger response patterns, and we react. This is of course a gross oversimplification (flinching under imagined flames from neurosurgeons everywhere :P )

The language is immaterial, perhaps something as efficient as possible because you will have a lot of commands to run through very quickly, to even start mimicing our thought processes... the database design is the hard bit. Ideally something with a weighting system, to trigger responses, perhaps activated or cleared with a heartbeat (A timed run through the database), and a way for the database to build and edit itself (without data loss... not an easy task). You're talking a very stupid, slow AI, even with current tech.

That's my best guess, anyway.

i guess u post is related to natural language processing.

well, as far as implementation is concerned, i think the language which most appropriate is c++ with embedded precompiled modules/codes of prolog/lisp, For visual interface, vc++ and for databases, oracle are the best try.

hope u`ve got what u asked.

I figure It would first be necessary to basically write a program with the ability to construct a sentence, meaning you would first have to create a huge list of english words, catagorized by types of speech(n,v,adv,adj, etc.), then writing rules for constructing different types of sentences, (declarative, interrogative, etc.) from different word types. (basically teaching it english) then expanding the database of english words to include synonyms, antonyms, and such, that way when the program detects a certain type a sentence, like interrogative, It knows the user is asking a question, and knows it must provide a declarative sentence in return, and constructs one with relavent words as dictated by the database's list of synonymous words and, you know, just work on it, later inserting points of logic and allowing the program to learn from normal speech. It may even be possible if you create a program that can understand and interpret speech to allow it to search and learn from the web (yikes, that could get scary)

Above post continued.........

Searcing through the database to get the optimum solution , is the most important aspect of any artificial intelligent system to be efficient.

There r many searching techniques which find out the optimum soln`s within severe time limits among a number of soln`s. Game playing (like chess, tic-tac-toe) uses searching techniques to much extent.

so i guess if ur planning to make any such system with human like intelligence then to start with make a game like chess in which ur opponent will be computer who`ll respond to ur moves, those moves should be the best moves among all available moves. u`ll get the logic of such search techniques in almost every good on AI.

moreover if ur interested in Machine Intelligence then Neural Networks is the key to ur question.

wich u all the best.

I don't really have any specific plans, just curious really. but the more I think about it, the more the idea facinates me. I human thinking is basically the same as computer calculations, with the right code ( and probably a hell of a computer to process commands) I think it may be possible to make a learning program with the ability to understand and interpret language, which is pretty much AI, which can become potentially more intelligent than humans. (whoever does it better be carefull, though, look at the terminator and matrix trilogies!)

You're still thinking too logically. People rarely construct a sentence in their head. Rather, we tend to respond by rote to situations.

Example: A child is taught 'An apple is red'. The concept of what an apple is is stored as a series of patterns: taste, shape (visual reference), and now colour. Red is also stored as patterns, mainly as a visual reference, but it's now also linked to the 'Apple' clusters. The next statement arrives 'What is red?'. The pattern of the statement is compared to stored patterns and the most dominant reactions call responses... 'Red' is fired off, triggers 'Apple' and the link, which is also linked to the phrase 'An apple is red'. The phrase pattern is presumably linked to concepts of speech and the associated motor controls, which fire in turn, and so on until the correct sequences are produced to utter the initial statement 'An apple is red'.

Again, I'm oversimplyfying, but I think the database would have to be a collection of concepts (of which a phrase is one concept) which are compared against inputs and given weightings. As more input is applied that links to concepts, the input is stored as further concepts and given the associated links.

Links must also have a lifespan, depending on their reinforcement.

Concept weighting must also have a lifespan... as input that requires a response is entered, the patterns that are compared that match are given a weighting. The most dominant weighting is decided to be the appropriate response, and the weightings slowly die away over time with the links.

I once wrote a full artificial intelligence program that saved the whole world from a computer virus then became self aware and started building robots who could travel through time and kill people in the past.

No , wait that was terminator 3. Damn, I need more coffee.

Wrayth, I understand where you're coming from, but realize that a computer doesn't have a sense of sight, taste, smell or touch, It's entire world is based on it's input devices, so you must first create a program that learns from an interface it knows, such as text.

acid, yeah, I know what you're saying, but you still need to think in terms of concepts, not text strings. As a silly example, the string 'red' has no associations other than perhaps the letters 'r' 'e' and 'd'.
However, the concept of 'red' within the brain has limitless associations... wavelengths, red objects, memories, the above-mentioned string, and so on, each in turn regarded as a concept with associated links, not as an individual component.

We need to get the computer to always think of any object in terms of its links, not as a single unit.

the fact is a computer is not a human, so it seems foolish to try to program it to think like one, my opinion is you must teach it first to learn from stimuli and then interact with it like it is a human, teaching it not to think like a human, but to interact with humans as if it does.

nah, then you get another dumb robot. Lots of those around, just check any university website.

Sequential processing isn't fast enough, so you need to fake referential networking. the complex bit is the continual self-adjustment of the database to develop subgroupings to filter the concepts through.

Yep, My T-X does all these things and it also travels through time and kills people.

I suspect that one day it will get to big for its boots and start a thermo nuclear war or something

I am not purposing that writing a program that can interpret language is writing a program that is intelligent, I'm simply saying you need an interface before you can teach it anything.

*nod* granted, but you need the interface to process the information correctly, as well as store it correctly. That way, it will end up teaching itself. Otherwise, you've hit a dead end. You can easily end up with a huge vocabulary, but it wont have any way of referencing inputs... other than a very slow sequential search through the database, for each component. My way, you're already referencing complete concepts.

thats, just it, how do you teach it to build INTELLIGENT sentences, not just sentences. that is the real challenge

By not building sentences word by word. It's like building a brick wall by making each brick from scratch. Rather build a store of complete bricks and assemble the bricks as you need them.

We do a lot less base thinking than we realise. Most of our interaction with the world is pre-set responses. 'Learning' is just accumulating a store of those responses and linking them.

ya, I read somewhere a long time ago about someone attempting to create an online artificial intelligence by feeding massive amounts of logical statements into a search-engine type program, such as "apples are red" and "when you die you stay dead" and hundreds of very simple little logical statements a day for a few years, and soon they had a program that was actually rather coherent, but not artificial intelligence, I suppose that after one writes a program that can build correctly formatted speech, the next step would be to feed alot of logical concepts into it, to sort-of give it a sense of coherence.

(going to bed, be back in a few hours)

Well, the next step is the app has to continually edit the database, somehow with the links... essentially building a filter system to clump the concepts (at least, from the outside)

Perhaps a front-end (sort of like the motor cortex) to cut input into manageable chunks. And to make sure the output is logical.

Sleep well.

Mmm i already have artifical intelligence, My Brunette girlfriend died her hair blond(and that would be :q!)or maybe esc,tab,:q!?

I think that in order to get a fully operational form of artificial intelligence we have to retrieve the chip from the model t101 in the factory where sarah conner crushed it.

Then we can work out its logic and reasoning and build super strong ento skeletons and cover them in biological material to make the perfect killing machine or AER.

So, Who's nicking down the factory to get the arm and the chip back then.

Nah, I nipped down there yesterday but they already threw it out. You gonna have to go dumpster diving again.

(You're really not helping, you realise this? :P)

LOL,

O.K. I will contribute something worth while.

Do a google on AI programming languages. you will find several languages dedicated to AI.

P.S also do a google on "weapons of mass destruction" and hit the i'm feeling lucky button. Its funny.

An AI program would not usual modular or object orientated languages but rather a complex system of neural nets and complex matrices in order to 'learn'. Nodes or synapses would have to be formed and linked and be able to stengthen or weaken the weight of pathways through experience. I do not think computers as they are designed today(digitally) are capable of reproducing the functions of an organic brain. They funtion on very different principles, ie analog and digital. Humans do not calculate things, they just DO them because they have learned through a vast amount of trial and error.Where our decisions are made from past experiences and constantly changing, a machine uses logic structures which do not change. Also text would have absoloutley no meaning to a machine as it is used solely for our benefit not the machine's, its all 1's and 0's to them. You must create a brain which can learn to crawl before it can walk, we aren't born reading and neither will a machine be. An AI would have to spend time as an infant unable to communicate just like us and learn first to organize the data recieved from its senses before it can begin to make use of that information. Most likely as silicon based technology becomes obsolete we will see an integration of existing technology with more biological based machines, and computers will become far more complicated and powerful as they are able to grow and adapt to their environment like organic lifeforms.
-Maestr0

Sorry, but i couldn't resist...

Quote:

---

Artificial intelligence is no match for natural stupidity.

---

i found that in the TechHumor forum :D

A widely used language for artificial intelligence is LISP, which stands for List Processor. Programmers like it so much, because it can handle complex data structures so well.

You might want to visit this excellent site: http://www.cs.berkeley.edu/~russell/ai.html

They have an upper year AI course at my university and it is taught using the Scheme dialect of LISP that peace_on_earth mentioned.

Quote:

---

Scheme is a statically scoped and properly tail-recursive dialect of the Lisp programming language invented by Guy Lewis Steele Jr. and Gerald Jay Sussman. It was designed to have an exceptionally clear and simple semantics and few different ways to form expressions. A wide variety of programming paradigms, including imperative, functional, and message passing styles, find convenient expression in Scheme.

---

Check out more info on it here:
http://www.swiss.ai.mit.edu/projects/scheme/

do you people have any idea of "CYBORG" Project.


what is it all about?

Why it hasn`t been completed successfully, continual research is going on it since past decade?

Maestro, that's exactly what my proposed database structure should emulate. It's not going to create a genius, at least not with current processing power, but at least it'll be an interesting way of modelling the development... incidentally, there's been some huge advances on bio switches for processors.

AlcatraX, never heard of it... links? More info?

another subject you would need to take into consideration when writing a program for imitating human thinking is psychology. To an extent, it also is based on rules, just like language, these two areas are where I think one should focus. Because they are based on rules, they can concievably be imitated by computer programming.

I was trying to create my own AI program last night.

The results are in the photo

There are a lot of things to consider before choosing a programming language. But just
so you know, I don't think there's a quick enough programming language and/or computer
to do this sort of thing. IF you want to set up a cluster of machines, then the database
searching routines (which would take a LONG time on a normaly comptuer unless a new
EXTREMELY fast DB structure is invented), then it would work. At the moment, the only
way a fully functional AI robot (with all of the features I have in mind) could work is if
it had a wireless network attached to itself to connect to HIGH (VERY HIGH) speed internet
connection, which in turn would connect to some sort of cluster over the internet. Though,
we I could be wrong here...the government has all kinds of cool stuff that we don't know
about :)
The first thing I would do is work with the language processing routines...they are very
complex. And while doing this...I would work with the learning routines. The problem with
making an AI bot is a few things: 1. Humans can hardly ever agree on what absolute truth is on some things and 2. Any useful AI bot would have to be a lot smarter than the creator in
its ability to learn, because as humans, we even have our limits.
If we beat #2...we could literally have soemthing like "The Matrix" happen. Well...I mean
I doubt we could get free & stuff...but if we could.../I/ would be "the one" :)
3. Personality....thats part of what makes us human...and also part of how we learn
4. Emotion....same as above
The AI bot would literally have to go through a growth process...deciding its own personality and what it wants to be. Emotion is really tricky...I don't know that we have any perfect words that describe a lot of our emotions. And as far as learning....we all learn differently based on environment, personality, and emotion....do we all reach the same conclusions? No. The AI bot would literally have to have its own life. Unless you just want a servant...but
that in itself would be easy. Creating something to understand commands is a lot easier
than creating something that gives back truly intelligent feedback. Dude, if I made an AI bot, I would want it to be able to talk to me about things...but how could it do so when emotion is so hard to define? How could we do so when each word has so many meanings depending on context, past experience, learning abilities, and personal connotation? Dennotation is worthless here...I mean...if we want flat out dictionary responses, that would be easy...but
a true AI bot would be able to learn & understand slang. Its VERY tricky to make an AI bot like a human. And to make an AI bot anything other than like a human would seem pointless to me. But let me ask you anyways: How human-like do you want it and/or could you describe your AI bot a little bit better?

-Lanica

Ooops....one more thing:
The key to this is to knowing more about what it is to be a human and just exactly how we work. There are A LOT of things we still don't know yet.

<<<<<< Now officially ignoring mark_boyle2002 >>>>>>>>
(just kidding)

I think we're going at this all wrong. how did humans become the 3rd most intelligent species on earth, (right below dolphins and mice)? EVOLUTION. I think the secret is to have a large team work on different ways to perform a small funtion of an artificially intelligent being. then put them up against human feedback. for instance, have several different groups of programmers work on different speech processors, put all of them to the test with human beings, the humans speak into microphones, the computer gives output of what it THINKS the person said, and the person selects the parts of the speech the computer got right, the most accurate program is then put under review from the ENTIRE team to finds different ways to make it better, then the new, improved versions are pitted against one another yet again... repeat a few times and what do you have? the best speech processors possible! repeat for other things, such as image processing, facial recognition, and plain-old language coherence, eventually the end products are grouped together as one artificially intelligent DEMI GOD!!.. well, you know what I mean.

Well, personally my goal is simply to model the brain in order to get the multilayered and referential processing abilities... emotions are not really necessary, and personality is simply your individual collection of defined responses, almost a side-effect of the learning process.

Speeding up the searching routines is simply a matter of making the filters more complex. That puts more pressure on the front-end than I'd like, but otherwise it'd take days for a single thought.

As far as making it human... I wouldn't want to limit the AI that way.

I think speech recognition/processing & image recognition/processing would be the easy part to this project. I know it sounds crazy...but all you have to do is to program the AI bot to retreive & intepret information by using various inputs...you would have to write quite a few search routines, but the idea is that it would learn & understand the functions of everything on its own. We would end up with a lot of broken things though. :) Its kinda like how AcidReign learned things on computers.... :) heh, but I did the same :) LOTS of Trial and LOTS of error :)

So basically Wrayth....you just want a more intelligent search engine, eh? :)

well, if you think about it, that is what the basis of the program would have to be, if you realize all the information humans accumulate over time, it would take a massive database to store it all, so a good portion of system resources would have to go to searching that for relavent data. that having been said, I think that the big thing that would have to be tackled on an AI project would be coherent conversation, (I know, I keep coming back to that) but really, that is how humans interact with eachother, through conversation, so I think that would be a very vital step in the creation on an artificial intelligence.

I'm trying to stray away from the concept of search engines, actually... people still think of those proceeding sequentially through components. My concept is similar, but follows links rather than some kind of sequential order, and weights concepts to emulate the conceptual dominance.

The search process is secondary, though... the real trick is to emulate the patterning process of the human brain. In other words, it only matters how you get it, once you've figured out how to store it. We're still approaching it from too compartmentalised a perspective... we have to think, like, holistically, man :P

Conversation is a side effect of the learning process. Need to walk before you run, remember :)

We need to create the equivalent of a baby's mind. Once that is done, we simply add more pathways to create more complex understanding... but the secret is the storage and processing, not the input.

what about a relational linkage system with, like, different degrees of relevance, linked items with lower degrees of relevance are ignored during more intensive searches. like, you keep bringing up an apple as an example. The search term "apple" would return a lower degree of relevance for "red" and a higher degree of relevance for "fruit".. how would the degrees of relevance be determined by the program, though?

(when accumulating new information, i mean)

Relevance is determined by personality. What do you think of first, second, & third when you see an Apple? Do you think the word? Do you focus the image in your mind? Do you see yourself eating the apple? Do you think of the taste? Its not the same for everyone. How could we possibly weight the interpretations? There needs to be one or the other of these things:
1. No weighting...multiple outputs
OR
2. Personality to allow weighted responses.

Yes, links would need different 'strengths', and they also need to slowly decay.
Concepts (stored elements) also have different search 'weightings' (which also decay)....

Basically a search starts, and each time a concept is referenced by a search and scores a hit, the concept increases its weighting. The concept with the highest weighting is put forward as the resultant output from the input.

We could give the AI a moral sense by artifically weighting 'bad' concepts with a negative result, if we wanted ;)

Basically, the link strengths would be enhanced by repeated additions and reinforcements... if we told the AI 5 times that 'apples are red' and once that 'apples are fruit' the AI would think it more important that apples are red. When asked what an apple is, it would reply that 'an apple is red'... sounds wrong, but teach a 3 year old the same way, you'll get the same result.

As far as teaching goes, I suggest repeating the same idea as the AI bots that are already out there... get the basic structure formed, and then let it chat to the people on the web. It'll be silly at first, but get smarter as more people talk to it.