Ai Dreams Forum
Member's Experiments & Projects => AI Programming => Topic started by: Mr.Fox on March 24, 2016, 09:23:28 am
-
Here is a recent white paper of some of my A.I thoughts, projects and more, take a look.
HERE (http://dream-walker.weebly.com/uploads/2/5/3/9/25390328/a.i_by_craig_fox.pdf)
Hope this sparks an interesting discussion!
-
Very nice review. It's always a plus to add a face to a name. Hopefully we'll enjoy some more experiments in this growing field. O0
-
Ok, so it's a sentence repeater bot. The code goes: on A reply B, right?
How about extending it to something "smarter"?
When a user says: "John walks.", we could parse this sentence as "A DoesX" and store it to a long term memory. Now we train the thing like:
"A DoesX" & "What do A does?" -> "A DoesX"
Now we have to replace "A" with "John", "DoesX" with "walks" and we get an advanced answer like: "John walks".
1. To make this happen, we need an english parser, so we can extract and analyze sentence fragments.
2. To make it a bit advanced, we could make an automatic trainer. When we automatically recognize responses of a user upon discussion, we can use the same responses from the machine upon later user input. The machine learns on the fly by communicating with it.
3. Going further without predefined English language, we could statistically analyze all input and responses, and automatically build parsing rules for a language that user used. After a considerable amount of input, we'd have enough data to extract rules of any language.
Points 2. and 3. give us true AI. Now, how hard would it be to program this behavior?
-
Very nice review. It's always a plus to add a face to a name. Hopefully we'll enjoy some more experiments in this growing field. O0
Thanks, you will, watch this space!
-
Ok, so it's a sentence repeater bot. The code goes: on A reply B, right?
<snip>
While I agree I think you kinda missed the point of the article, the software is just a simple repeater which is being developed on but that was just a side note really in the article. The point really in short I suppose is, artificial awareness/conciousness and the fact that, for A.I to be smart, it doesn't need to pass any turing test or be able to play games, or even speak English for that matter (or any other language for humans). It would merely need to be aware and have instincts.
-
sorry, i was just trying to help :-\
-
sorry, i was just trying to help :-\
I know it's cool man 8)
-
Hi Mr.Fox. Buddhists say we don't have 5 senses but 6. The 6th sense is the one that makes you feel what's happening inside your head. Try this:
Take your favorite language, say Lua, whatever. Wrap every function you'll use in a logging clone. The cloned function does the same as the original one, except it also sends an activity report to standard input. Tada! Self-awareness!
Just kiddin'
Welcome aboard ;)
-
There is not much real ai projects out there mr fox .
Noticing: most start with the same ideas...then forget actually its not as easy as they thought ....
Designing decision processes , are the next step .... As well as storing knowledge structures and retrieving that knowledge .
Often where most get stuck ...... Creating "sense" drawing conclusions , making further implications ..... Even generating a response instead of recalling some scripted answer.
Basic intelligence !
Then get to more higher intelligence levels .
Only in stages can it be done. The stages could be described as the secret behind creating A.I
-
Actually, I agree with Mr. Spydaz's assertion. Well put. O0
-
Yep I go with Spydaz too. It's the search for the philosophers stone again.
-
There is not much real ai projects out there mr fox .
Noticing: most start with the same ideas...then forget actually its not as easy as they thought ....
Designing decision processes , are the next step .... As well as storing knowledge structures and retrieving that knowledge .
Often where most get stuck ...... Creating "sense" drawing conclusions , making further implications ..... Even generating a response instead of recalling some scripted answer.
Basic intelligence !
Then get to more higher intelligence levels .
Only in stages can it be done. The stages could be described as the secret behind creating A.I
Yeah, I tend to agree also - some think this will be resolved by Machine Learning/Neural networks.
-
I would say that neural networks although solving some "predictions" based on past history (a decision process) cant handle conversation only maybe pick the best matched "scripted answer" (based on history of conversations) ... "I Read a paper on that" but it is a single multipurpose tool in the overall paradigm .
I have recently began building a "belief network" which when built will also evaluate which decision to make based on a decision tree / bayes probability . (Then the decisions can be related to personality type) depending on internal risk management strategy .
Again this is just for knowledge evaluation/ internally modelled decisions . But another step towards building a non scripted response ....
Which is one of my goals , the main being "understanding" and making all the collected knowledge "interogatable" ...
-
I would say that neural networks although solving some "predictions" based on past history (a decision process) cant handle conversation only maybe pick the best matched "scripted answer" (based on history of conversations) ... "I Read a paper on that" but it is a single multipurpose tool in the overall paradigm .
I have recently began building a "belief network" which when built will also evaluate which decision to make based on a decision tree / bayes probability . (Then the decisions can be related to personality type) depending on internal risk management strategy .
Again this is just for knowledge evaluation/ internally modelled decisions . But another step towards building a non scripted response ....
Which is one of my goals , the main being "understanding" and making all the collected knowledge "interogatable" ...
I'd like to see this.
P.S
A neural network using Unsupervised Learning could, in theory, with millions of neurons react in a non scripted manner - the problem I think mainly is hardware and algorithmic restrictions... I read a paper recently about a professor in the UK Manchester (yay, my home town) that is trying to do just that, I'll dig it up.
-
Glad to read it.
But i would disagree ... For many reasons ..but a very large response for here....
"It will be discovered that neural networks are not what they expected"
Would be nice if they could solve every problem .
But its like using a hammer when a screw is required (ai needs a multitude of techniques)
-
ai needs a multitude of techniques
I agree, but these techniques should all be hosted in the same homogeneous middleware.
There is not much real ai projects out there mr fox.
What you believe is what you get.
:jump:
-
hey art !
Interesting paper that you wrote.
It seems to be a deviation off of the Eliza program from way back.
I think that the idea is still usable though.
Obviously you want to expand it.
You hight try writing some Examples in HTML - probably would be easy and straightforward...
As for the Neural Coding,
Interesting ...
But, I think its a waste of time - just another algorithm.
As I like to say :
Its like trying to build a House out of Grains of Sand,
It might eventually work - but I don't want to try it !
- Look at how long the Human Brain took to Evolve from the Animal World...
-
@ jlsilicon,
It was Mr. Fox who wrote that paper, not me. Just setting the record straight! O0
-
Congrats to mr Fox !
(sorry)
Interesting paper that you wrote.
It seems to be a deviation off of the Eliza program from way back.
I think that the idea is still usable though.
Obviously you want to expand it.
You hight try writing some Examples in HTML - probably would be easy and straightforward...