[lrh9]

I don't know. I've never read any of his work and I don't know anything about his latest book. I'm more into stuff by Robert Jordan myself.

Right now I'm reading a book named Moral Machines. Maybe you've seen my post about it.

I finished watching the movie about Helen Keller. I just have to go over my notes one more time to formalize the ideas that I need to analyze. I'll probably be watching L'enfant Sauvage soon. I finally got a copy yesterday.

I probably won't be programming any more of my project at least until I finish watching L'enfant Sauvage. I'm pretty confident I got the mechanics of learning - at least insofar as they can be applied to computers - correct. However, I should have paid more attention to the method of learning. What I discover about learning from these two movies might prove that my project - as it stands - will not succeed. I do not want to take the effort of creating something that is supposed to learn if I cannot teach it.

[lrh9]

I finished researching my material. Now I need to evaluate which direction I want my a.i. project to go in.

[Art]

Have you given further thought as to what method you will employ for the bot to learn or will there be Options from which the user can choose?

Will there also be a provision for it to learn or at least draw information / answers from searching the Internet?

Does any of this "learned" information go into a database or table for future reference?

No offense but let's hope this bot is not another typical AIML / Pandora / scripted type of bot! There's no real learning unless the "botmaster" assists and IMO, that's NOT learning!

If I can be of assistance don't hesitate to ask!

[lrh9]

I disagree with your opinion about human assisted learning. Without human assisted learning (from our parents, teachers, and peers) we'd barely survive and even then we'd just be scratching around in the dirt. The majority of human knowledge is learned by rote. I suppose one could make a semantic distinction between obtaining or generating original knowledge and acquiring the knowledge in another person's or item's database. Regardless, human assisted learning is vital to rapid individual advancement and advancement of a group's knowledge base.

However, from the potential abilities I've noticed of the program, there are certainly some that suggest it will be more advanced than previous chatbots and desktop assistants. I won't go into all of them, because some are more certain than others, and some may be taken out, and some may be added, but one that really excites me is that the program may have the potential to be symbol grounded.

Basically, symbol grounding is how words get their meanings. When we know that "apple" represents an apple, then the symbol "apple" is grounded to the physical object. One thing almost universally agreed necessary for symbol grounding is a sensorimotor system. Basically the ability to interact and obtain data about objects. Our body is a sensorimotor system. We obtain data about the apple such as its color, taste, shape, and texture and then we can assign a word to it. This program could accomplish the same thing given sufficient sensorimotor mechanisms. I'll admit, there is some debate as to whether having sense data on an object is sufficient to give a word a meaning. One hypothesized vital component is consciousness. We'll have to wait and see how it turns out.

[Art]

Symbol grounding...

You might look into the ASL for some additional insight.

My grandchildren were taught a lot of it at a very early age
and became quite adept at communicating with it way
before they could even talk.

And yes, you're certainly entitled to your opinion no matter
how far removed it may be from mine.

I merely expressed the idea that the scripted bots were
not the learning type of bots. The late KnyteTrypper and I
often had lengthy discussions of scripted vs learning and
while each type has it's place, we both agreed that the bots
that could learn unaided or unassisted were by far, more
interesting in their sometimes unexpected conversational
responses.

To each their own.

[lrh9]

Just how does something learn something useful unaided and unassisted?

[Maviarab]

Trial and error?

[lrh9]

How does something recognize when it has encountered an error?

[Bragi]

Just how does something learn something useful unaided and unassisted?

Perhaps an othother way besides trial and error, you can try to find an algorithm that is able to find structure in the incomming data.

[lrh9]

Just how does something learn something useful unaided and unassisted?

Perhaps an othother way besides trial and error, you can try to find an algorithm that is able to find structure in the incomming data.

That is assisting the program.

[Art]

I think what some of us are trying to get you to see are programs similar to the John Lennon or Jack-the-Ripper or AIML bots that are almost "expert systems" in that they contain a finite amount of data regarding a particular subject. Any additional information or data must therefore be added by a diligent botmaster.

Other bots might very well absorb new data gleaned from conversations and filed away only to be recalled later, perhaps next week or when the bot deems it to be an appropriate "link" from one subject to another, etc.

Some bots can be asked to "read" a passage of text. Though individual words and phrases are not selected by a botmaster as such, the bot can and will again use said information at its disposal often surprising the user.

Though I had long forgotten a conversation I once had with a bot, months later it brought up a small facet of my family that actually made me wonder how or where it had learned that information. Upon reflection, I realized that it must have been part of an earlier conversation but the bot "saw fit" or at least chose to save that info for later use.

The word "learning" in the context of a chatbot often takes several forms and is accomplished by as many methods but we all know it's just "managing" the words, phrases or paragraphs and filing away some key bits of information like subjects, actions, relationships or proper nouns, etc.

Regardless of the internal mechanics of the bot, it is ultimately the output that either impresses or annoys the end user.

There is no right way to construct a bot and while none are perfect they all offer us a small glimpse of how our own minds work.

[Bragi]

The word "learning" in the context of a chatbot often takes several forms and is accomplished by as many methods but we all know it's just "managing" the words, phrases or paragraphs and filing away some key bits of information like subjects, actions, relationships or proper nouns, etc.

Regardless of the internal mechanics of the bot, it is ultimately the output that either impresses or annoys the end user.

Exactly: it's just a trick that manages to impress us (if we're lucky).

[lrh9]

I just hope I'll be able to construct an a.i. that can have an actual understanding, at least more of an understanding than current a.i. programs do.

[lrh9]

I have taken a small vacation, but otherwise I fully intend to continue my work.

Right now I'm still learning the intricacies of the Python language and modifying my abstract design of my a.i. agent framework as necessary. I'm trying to be as simple and lazy as possible, in the good way. Basically I want to reduce the complexity of the design and reduce the amount of code. However, those goals conflict with other goals such as conceptualization - designing and using the program around a concept. While there might be a simple way to do something, it isn't exactly the way it should be thought of. For instance, I could import ability modules into the agent using a simple import statement, but that doesn't exactly fit with the concept of messaging. I need to find a way to reconcile these differences.

I've really only had one legitimate person interested in the project, and it has been some time since we've had some contact, but I'll do all of this by myself if I have to. Indeed, that's what I foresaw from the start.

[lrh9]

I've worked some more on my program.

Using Python's object oriented features, I've made progress on the design of the agent itself.

I've written an agent class, and I added a property for its unique identifier, which was supplied from Python's uuid library. I'm glad for that, because I would have had to write an algorithm or implement an algorithm myself.

Code

#Python 3.1

from uuid import uuid4 as _uuid4   #Provides access to a function for creating a universally unique identifier.

class Agent:

    def __init__(self):    #This function is executed each time an agent is created.
        self._uuid = _uuid4()    #Sets the agent's universally unique identifier

    def _get_uuid(self):    #This function returns the agent's uuid.
        return self._uuid

    uuid = property(_get_uuid)    #Defines a general attribute with a get() method.

#Test code. Delete the quotations to run.
"""
agent = Agent()    #Creates an agent.
print(agent.uuid)    #Prints the agent's universally unique identifier.
"""

I'm also excited by the possibility of using Python's import feature to utilize ability modules, instead of the messaging system I was previously considering. I thought about this idea one night, so I did some tests and determined that Python object's - such as agents - can have their own instances of modules. The most obvious benefit of this is that it would save a lot of development time by using an existing system instead of one made from scratch.

Modules provide a place to store function and class definitions, so any agent importing the module would have access to those. That's one of the obvious functions of ability modules. I'll just have to consider how a Python module could communicate with hardware - because eventually this system is intended to work with hardware - and I'll have to consider how object's could share a module.