I’ll be updating this thread with my ideas on how to go about creating an AI that “understands” concepts or appears to understand.

Context is a big issue with AI chabots. Will look into this too.

So, my current AI works by matching user inputs to preprogrammed responses in it’s databse, then outputs the response. Very basic, if user says: “hello”, it will match the word/phrase to “Hi user” and that’s it.

Now, I want to make an AI that can use logic. Which means I have to program a logic database for it to use in order to work out similar situations.

The AI should be capable of utilising the logic behind cause and effect, this would require prior knowledge of physics in this world.

Hmmm, purpose too.

Let’s program our AI to get to the other side of the road. It has a purpose now. But, it has to fulfill it’s purpose. There are cars driving on the road, and if it gets in the way of the car, it will not make it to the other side, therefore mission failed = purpose not fulfilled = must avoid failing mission. If we can program the AI to predict, then it can work out cause and effect, and vice versa.

Looks like we need modules.

Purpose processing - multiple sub modules - long term and short term purpose
Task list - long term and short term
Computer Language - prolog, c++, etc integration.
Visual - general tracking - advanced hardware and programming required
Audio - Will be very hard
Text>Direct
Language processing - audio, visual and direct text
Logic Processing - further sub modules
Prediction
Knowledge & Information database - further Sub modules
Creativity Processing.

Any other modules we need?

Are you using Prolog for your Project?

Not sure. C++ is the core language, with prolog for the logic module etc.

http://www.chatbots.org/ai_zone/viewthread/839/

Yes, Prolog is rather good at what it does. You may want to have a look at Logtalk if you want to add OO features to your Prolog code.

Excellent, now I just need a team of skilled programmers.

So atm, I’m using C++, with Logic Paradigm (http://www.mpprogramming.com/cpp/) and libxl (http://www.libxl.com/).

I can use C++ for imperative code and algorithms, but I can also use declarative code for facts/logic programming, and I can use excel spreadsheets for the database.

I’m trying to create an AI program that can “understand” or appear to “understand” concepts. So, if the user stated: “I shot an elephant in my pajamas”, it would create a mind map like so:

User -> shot (action) -> elephant (object) -> in pajamas (description of either elephant of user) -> information incomplete -> request more information to confirm understanding.

The AI needs a purpose to be able to display a reasonable understanding of basic concepts. For instance:

Walk in front of car travelling at 10mph+ -> risk (dangerous) -> Objective failed -> termination
Avoid car -> objective not failed ->  proceed

Similarly, we can program a logic module:

If man = human
and human = mortal

then the AI can use substitutions to work out that man = human = mortal, and by omitting the results linking the two associtations we get: man = mortal.

Creating a strong AI is kinda like solving a sudoku puzzle, you gotta start somewhere and make changes as you go along.

> http://www.meta-guide.com/home/bibliography/google-scholar/tasks-of-natural-language-processing

Tasks Of Natural Language Processing

> http://www.meta-guide.com/home/bibliography/google-scholar/modules

Modules / Modular Programming

> http://www.meta-guide.com/home/application-programming-interfaces

Application Programming Interfaces (APIs)

> http://www.meta-guide.com/home/artificial-intelligence-middleware

Artificial Intelligence Middleware

> http://www.meta-guide.com/home/open-chatbot-standards

Open Chatbot Standards for a Modular Chatbot Framework

Some very helpful links, thank you!

Would be able to help me understand how to go about the language processing aspect of chatbots? Having a hard time…

I was also hoping to integrate a visual aid for the AI, or something like a minds eye. So if I told it imagine a man walking across the road, and to show me car running the man over, it would show me the simulation (thinking of GTA IV style graphics and physics). Likewise it should also be able to associate words with real life objects/contexts/concepts, and further find associations between them.

> http://www.meta-guide.com/home/bibliography/google-scholar/computational-metaphorics

Computational metaphorics concerns the study of metaphor analytics.

> http://www.meta-guide.com/home/bibliography/google-scholar/computational-dreaming

A number of basic computational models for dreaming have been proposed.  Very likely metaphor analysis will prove to be the bridge between natural language technology as we know it today and the realistic brain functionality of tomorrow.

> http://www.quora.com/Linguistics/Is-it-possible-to-explain-metaphors-using-discourse-analysis-frameworks-or-only-with-cognitive-linguistic-frameworks

Is it possible to explain metaphors using discourse analysis frameworks, or only with cognitive linguistic frameworks?

How about a context module? Like during a conversation the AI would create a mindmap that it can refer to if necessary.

So in the beginning of th conversation if I say: “I went to the shop yesterday.” It would create a mind map:

User -> went to (action) -> the shop (object) -> yesterday (time description).

Later in the conversation I ask it: “Where did I go yesterday?”

The AI would refer to the mind map it created and find the solution:

Where (information request - purpose: provide information) -> Did I go (reference to past action - *information incomplete*) -> Yesterday (time description - lookup mind map for yesterday’s activities: “did I go” = “went” match found, object = shop) -> shop -> confirm with user -> objective completed.

> http://twitter.com/mendicot/status/206776160447770625

@Wolfram_Alpha .. re ‪#SystemModeler‬ => would like to point at XML ‪#sitemap‬ and get MM ‪#mindmap‬ back ..

==>> http://www.wolfram.com/system-modeler/