[Mr.Fox]

So this is something I made a few months ago and has had some good feedback.

A.I Robot Maker





This software lets you, the user create and train your own robot to be artifically intelligent. You start off by naming and creating your digital robot and then you teach it things. You can give it your own personality, make it smart, sarcastic, anything you want really. You can even go all out and make it an impressive A.I bot potentially fooling humans, it's really up to you! The cool thing is, you can share it with other people too, and respectively they can share there's also within seconds. You may just want to have the enjoyment of having your own pet robot, and you being it's God - whatever floats your boat, and hey, it even talks! A great way to learn, be creative and generally have fun. It is very easy and fun to use from a younger to more technical audience. It has built in tutorials, examples and facebook group community making it easy to share with strangers and friends alike.

It requires the .NET 4.5 Framework to be installed, ~27MB ROM and very little RAM. It should work on most modern windows PC's. Just download the zip file, extract it and run the application.


DOWNLOAD

[Art]

Interesting.

Are the brain files saved as normal text files or added to a common file that keeps growing as more info is "learned" or taught to the bot?

Can it be "force fed" new info as in the form of either a text file or spreadsheet format?

Could it adapt to use any of the ALICE / AIML file formats?

Just trying to get a feel for the software before downloading.

Thanks!

[Mr.Fox]

Yes, it's essentially a text file with the extension .rbt for clarity, a short snippet of the format would be:

Code

Example Robot
1.0
25/11/2015
hi:QUESTION
hello pooh face
yo:QUESTION
hey man
hello sir:QUESTION
i am not a sir, i am a robot, but you have good manners i must say.
who made you:QUESTION
craig fox is my master
...

There is a tutorial for it. So yes you can directly open and modify this file is you wish just stick to the structure it's in. It's designed so you can easily share and run other peoples robots with one click, it's all done with user friendlyness in mind but trying to keep it robust. Soon, for instance, you can get it to do system commands, get information about the computer it's on, and some basic scripting.

It's not compatable with AIML because it's different in it's structure. You're essentially teaching it at runtime so you don't have to worry about coding and manipulating files.

[Freddy]

Very nice, real-time learning is something I am working on at the moment although in PHP.

Will be interesting to see where you go from here - swapping brain files etc is a nice touch. Can the brains be combined into one ? If so then a complex brain could be possible quite quickly from a repository or your system.

[Mr.Fox]

Very nice, real-time learning is something I am working on at the moment although in PHP.

Will be interesting to see where you go from here - swapping brain files etc is a nice touch. Can the brains be combined into one ? If so then a complex brain could be possible quite quickly from a repository or your system.

Thanks man, it took about 3 weeks to make and test. This is the "core" though and as aforementioned added features are coming. Already in fact though not officially published you can execute system commands and (nearly) reffer to the user as there logged in name on the PC which adds some personality, anyway I digress...

It would be fairly easy to combine files/brains, but the main issue I think would be conflicting questions and responses, and while cool, kinda takes away the indeviduality of your personal robot if you know what I mean.

[ivan.moony]

Mr.Fox, By my calculations you are closer to AGI than you think. Suppose you have inputs and responses parsed through natural language processing. You'd get a syntax tree of each input and its response. From there you could connect not only an input string to a response string, but to connect internal relations between the same.

So, starting from this

Code

input: The bird is flying.
input: What's flying?
response: The bird.

it is not hard to imagine that a response knowledge like this could be built:

Code

input: "X" is "Y".
input: What "X" is "Y"?
response: command (output ("X") from sentence that matches ("X" is "Y")).

You just need an English parser. That would be a good start.

With a little bit more polishing, this way it is even possible to extract all deducible knowledge by processing existing knowledge through classical logic deduction algo. that would be 1/3 of work.

From there, you can even process existing knowledge through inductive statistical analysis, and that would be 2/3 of work.

There are three types of questions:
1. yes/no questions,
2. Wh* questions
3. procedure as answer questions.

The first two wouldn't be hard to program and the third one, we could talk about it if you are interested.

Covering procedures too would be 3/3 of work and we could open a champagne then. There would be just a tiny thing to do and that is unsupervised learning. The solution would be to copy the behavior of users to the machine output.

I'm not trying to get you down. I just think that your app could rock a big time and maybe I could help with comments, or with some knowledge exchange. But maybe I'm pushing it too much. Maybe I should do it myself. I'll do it some day, certainly, but right now I'm busy with some other project. In a meanwhile, I'm just trying to help with ideas, if anyone cares.

[Mr.Fox]

Several updates done, please check it out guys!

[Art]

One thing is a particular thorn for most pattern matching bot and that is different ways of asking the same thing.

Do you know of a representative?
Do you know a representative?
Do you know any representatives?
Do you know of any representatives?
Do you know a rep?
Do you know any reps?
Do you know of any reps?

And so forth...

Does your bot allow for a quick inclusion for these different statements? Might there be another approach instead of typing all possible configurations of scenarios?

Does your bot allow for Wildcards like * @ % or whatever?

When testing your bot, I noticed that it is very specific in that the words and punctuation used to teach it are Exactly the ones it expects to have entered if it is to process them correctly.
Good morning! is not recognized if one states, Good morning
Hello! must not be entered as Hello

It appears that it is not case specific but definitely punctuation specific.

Demo only allows for a per session use and any entered work is summarily deleted when the program is terminated.
Kind of takes away any enthusiasm one might have in experimenting with it. Yeah, I know pay to play. That's cool!


Thanks Craig!

[Mr.Fox]

Soon, you can run the A.I against each other, should be interesting