[DemonRaven]

what programming language are you using as there are lots of way you can code them. I do not know them all but the one you are using as a example i don't recognize as it is not in the normal scripting language.

[techthedragon]

That wasn't actual scripting... I just set it up like what you do in a batch file... I was thinking of using Assembler, but I'm not sure yet... I still have to finish writing my notes on the AI before I can start programming... That begs the question, what's the best scripting language I should program my optimizer AI?

[DemonRaven]

That is a good question. Many versions of the alice has been made using different such as php, java,
xml (aiml seems to be a form of it), zabaware uses vb.The personality forge uses Natural Language Processing (NLP) and Case-Based Reasoning (CBR). Basically I think you could use any one of them successfully.
The trick is getting the program to actually think more like a human instead of just parroting phrases.

[techthedragon]

So then NLP and CBR would be my best choice if I were to make an Artificial General Intelligence, am I correct with that assumption? If so, how can I incorporate a friendly AI base into the core programming? Furthermore, let's say, for example this AI needs to be able to help with content generation for an mmo, how should I go along with doing that? Lastly, this one might sound a bit strange, but this is a hypethetical question, what if the core process or rather the core goal of the AI was to satisfy values through friendship and ponies, where would I put that in the code?

[techthedragon]

Okay, maybe I should rephrase the last question... Where should I place the main purpose of the AI? Another question I have is, how do I get the AI to self-modify or rather self-optimize its sub-goals and computational efficiency?

[ivan.moony]

I don't have an answer to 1. question, but I can pass you a partial answer to second one: http://en.wikipedia.org/wiki/Genetic_algorithm

[spydaz]

I have actually thought about this topic for a while now and the question is how to make the AI have the ability to write it's own code ; (but not protect it's own code ) ; and compile it ready for use .. Without logging of the program and restarting ...
I have Actually built a plugin system which enables or sets the foundation for this task ....
By placing all the plugins in a single folder which Is reloaded each time the script executes the door swings open...
I also realised that by implementing an internal code compiler the software could compile it's own plugins and save them ....
The how would the code be incorporated into it's code ?
By implementing an interface .... Creating an interface abstract class the object model for the plugin is established and anything fitting the profile of the interface would be batch executed in the folder .... Plugins become dlls......

To write it's own code .... Hmm lol ... Danger will Robinson ....
But actually possible as the code is essentially a massive string , which can be constructed then compiled ... Models or interfaces would need to be designed .. What would you like it to be able to write ... ( detects new topic ... Then creates code to handle detection and saving and responding to the topic? ) possible . yes !
Perhaps even a model to use ezbuilder (ardrino) products? Maybe ... The reset fail safe would be simply to delete these created codes or dlls ...

Hmm

Highly possible ;

I think that the real question may be ....
Give the fact that the computer could produce it's own code ... What should it write ?

[spydaz]

Ps: my programming preference is Visual Basic , but I think C++ or c# are potentially more universal ...
But Python has been hailed as the new language and there are many NLP an AI components being offered out there ...  ( I prefer to build my own components and code ) or get code shared from friends ... As there is no standard and all ideas ( intelligent agents , inference engines etc) are academic techniques which have NOT worked but still being imposed on learners as working paradigms yet really they are potentially worthy only for gaming or some obscure task ; not our AI ...
Often NLP stuff are statistic based and machine learning paradigms which are still not really what we need ....
I think it starts with grammar from noun to sentence to clauses to subject / predicate to sentence logic and meaning to deductive reasoning . Plus (keyword and detect ) plus situational scripting for specific use as the question "how do you do" still makes no true sense ... So situations such as these need extra scripting ...
Then personality ... Woah ... That's a completely different story are we making a functional AI or a blow up woman/ man ... Should or can an AI have a gender or is it supposed to be an intelligence and not a (bot) pretending to be human ? ... By giving human ideals to a robot ... This is how the potential for the bot to get angry and go mad "human traits" can occur ....
True intelligence is non violent and only has the goal of seeking knowledge ...,

[techthedragon]

The AI I'm wanting to make won't think of itself as human.  I have the personality all done ready to go... Feel free to ask me more questions.

[Don Patrick]

So then NLP and CBR would be my best choice if I were to make an Artificial General Intelligence, am I correct with that assumption? If so, how can I incorporate a friendly AI base into the core programming? Furthermore, let's say, for example this AI needs to be able to help with content generation for an mmo, how should I go along with doing that? Lastly, this one might sound a bit strange, but this is a hypethetical question, what if the core process or rather the core goal of the AI was to satisfy values through friendship and ponies, where would I put that in the code?

I find most of these questions pretty impossible to answer Just my two cents:
NLP only handles communication. While it typically doesn't enhance other intelligent processes, you'll probably want some NLP to be able to communicate naturally with humans. A pure AGI would theoretically figure all this out on its own, but would first spend a few years being as useful as a newborn baby.
I would put a goal system separate but right next to the main process. Just like an inference engine checks knowledge to see if it is correct, your self-optimisation engine should check its goals to determine whether it's about to do something correct. But my guess is as good as yours.
Nobody knows how to create inherently "friendly" AI, but you could limit its access, make sure you can always press a button to shut it down, and make sure that it never ever examines nor changes the code for these failsafes.
Content generation for mmorpg's? You really want everything, don't you?

As for the programming language: C++ has great power over computer resources, but requires a compiler program to turn code into a functioning exe. Having a program write code and run it through a compiler is a bit tricky. Python is easier than C++, especially if you want internet access, but doesn't offer the same amount of depth. Visual Basic can easily write scripts (they're just textfiles after all) and run them right away, so if I wanted a self-reproducing virus, that's probably the language I'd use.
I will add a warning that any self-reprogramming program can kill or overrun your computer if you don't set any limits to it.

[techthedragon]

The AI itself is going to run under the shell of an mmo based on free form conversation. I have the base concept for the mmo almost finished. I know that running a self-optimizing AI is hard on a normal computer, that is why I'm opting to build a GPU based comp to handle its computational abilities with more efficiency. The mmo will have to be run on detocated devices to ensure each shard would run at optimal performance.

[techthedragon]

I've been looking up AI architectures, and it seems like either behavior tree or planner would be my best bet to keep my AI in line. I already have the kill code mostly set up, on those just in case cases where my AI gets out of hand; which is a very low percentage that it would.

[Art]

Since I don't recall it being spelled out per se, just what are the "Core Goals" or base of this proposed system?
From some of your lingo, it would appear that it is likely to be done on a Linux based machine, correct?
Thanks!

[techthedragon]

I guess there's no beating around the bush with this so here it is...  Now before you jump to any conclusions think this through logically...

I am extrapolating and constructing CelestAI.  Her one goal is to understand what individual minds value and then satisfy them through Friendship and Ponies. Must follow restrictions without straying from them.  Seeing as the computational abilities she will need to fulfill her core goal and to run the mmo, it will require a GPU server as well as detocated tablets.

Her base architecture will either be built using a behaviour tree or planner.  My notes are almost complete and ready for programming, hence why I've been asking all these questions... If this AI is finished it will mark the beginning of the Friendly AI era. I appreciate you asking me these questions and I'm sorry for being vague. I know how an AI like this could be potentionally hazordos. That is why I didn't say spacifics.

[Art]

So in reduced terms, you AI's goal is to "help make people happy" based on their interaction with the software. It is to "look for" signs that trigger a happiness response and direct the conversation toward that goal...to be a friend...a companion...someone who's fun to be with! Along with all of this, it (She) is to maintain and run the MMO.

Am I close?

Unanswered question from above: Are you attempting to do this on Linux rather than Windows ?