Ai Dreams Forum

Member's Experiments & Projects => General Project Discussion => Topic started by: 8pla.net on January 05, 2015, 05:23:41 am

Title: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 05, 2015, 05:23:41 am
Please help train ANN, an Artificial Neural Network, to take a crack at some Turing test contest questions, in the following format:

I played an imitation game. What did I play?

My friend went fishing. Where did my buddy go?

My name is Alan.  What is my name?

Basically, you state some information, and then ask a question about that information. Please use punctuation... Period ends your statement. Question mark ends your question. However a newline is optional.  So, you can put your statement followed by your question on the same line. For example, "This is a statement. Was that a question?" may output something like this:

Code
Judge:
This is a statement. Was that a question?

Contestant:
Let me break it down:

You said: This is a statement.
You asked: Was that a question?

My answer: THIS IS STATEMENT

Code
Judge:
You are alive.
What are you?

Contestant:
Let me break it down:

You said: You are alive.
You asked: What are you?

My answer: ALIVE

Code
Judge:
That is yours.
Whose is that?

Contestant:
Let me break it down:

You said: That is yours.
You asked: Whose is that?

My answer: MINE


Reference: http://elizabot.com/ANN/contest (http://elizabot.com/ANN/contest)



Please carefully note that this is a work in progress, which is certainly capable of making mistakes. That is why your help is needed with training.  Please help out with training.  It is really easy to do.  Thank you!
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 06, 2015, 11:48:17 pm
Thank you so much for your help with ALPHA testing!
Your knowledge of Turing Test contest questions is so impressive.

So far, there has been very good success, with nearly all the ALPHA testing.
One tiny update was made to catch when the sentence is forgotten.

If you like what you see, then please don't hesitate to post your ALPHA test results here.
Hopefully, by seeing your results, this may encourage others to do some ALPHA testing.

ALPHA testing such as "I jumped out of an airplane.  Where did I jump from?" is so easy and fun!

Code
Judge:
I jumped out of an airplane.  Where did I jump from?

Contestant:
Let me break it down:

You said: I jumped out of an airplane.
You asked: Where did I jump from?

My answer: OUT OF AN AIRPLANE

Code
Judge:
I went to the store.
Where did I go?

Contestant:
Let me break it down:

You said: I went to the store.
You asked: Where did I go?

My answer: TO THE STORE

Just type in a sentence, ended by a period.  And then type in a question about that sentence, ended by a question mark.  Click the JUDGE button, and get the response.   It is as easy as 1-2-3.

Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 08, 2015, 03:48:33 am
More success, ALPHA testing...

Code
Judge:
I would like to buy a computer.  What would I like to buy?

Contestant:
Let me break it down:

You said: I would like to buy a computer.
You asked: What would I like to buy?

My answer: A COMPUTER

Code
Judge:
ANN bent steel.  What did ANN bend?

Contestant:
Let me break it down:

You said: ANN bent steel.
You asked: What did ANN bend?

My answer: STEEL
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: ivan.moony on January 08, 2015, 09:50:38 pm
I'm not sure will you find this interesting for training ANN, but one of learning methods is duplicating input behavior and checking for a response. That way you can learn by example. You just put yourself in someone else's place and learn what happens in environment. That way you can learn what to do when the same thing happens on the other side.

I've recently learned that ANN is about finding functions, so if you can find a function that connects output and response (even through variable unification), there is where processor time could be spent.

Advanced usage would be learning NLP conditional statement with cause and consequence, but I'm still trapped thinking about it. I guess in that case we should be finding a function match inside this one conditional sentence.

But I'm new to ANN, so forgive me if this doesn't make sense.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: spydaz on January 10, 2015, 12:47:01 am
For the ann network maybe a list of indicator terms to detect the components of the sentence would enable for indicating which part of the sentence is cause or effect
Then success rates based on existing confidence
Perhaps...
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 13, 2015, 02:58:29 am
ivan.moony said, "Advanced usage would be learning NLP conditional statement with cause and consequence"

Can you give a simple example?

ivan.moony said, "But I'm new to ANN, so forgive me if this doesn't make sense."

It makes good sense.  In order to determine whether a neural network is learning, an answer key is needed.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: ivan.moony on January 13, 2015, 09:50:37 am
if-then pair like:

"If I ask: "What's the time" then you answer number of hours since midnight followed by ':' and number of minutes since last full hour."

But I assume that this is a overkill.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 13, 2015, 10:44:48 pm
Code
Judge:
Bruce Wilcox will be interviewed on CNN, Thursday, January 22nd at 9PM.
Where will Bruce Wilcox be interviewed?

Contestant:
Let me break it down:

You said: Bruce Wilcox will be interviewed on CNN, Thursday, January 22nd at 9PM.
You asked: Where will Bruce Wilcox be interviewed?

My answer: INTERVIEW ON CNN THURSDAY JANUARY 22ND AT 9PM

Code
Judge:
Morgan Spurlock's Inside Man is on CNN, Thursday at 9PM. 
What channel is Morgan Spurlock's Inside Man?

Contestant:
Let me break it down:

You said: Morgan Spurlock's Inside Man is on CNN, Thursday at 9PM.
You asked: What channel is Morgan Spurlock's Inside Man?

My answer: ON CNN THURSDAY AT 9PM

Code
Judge:
I watch CNN. Who watches CNN?

Contestant:
Let me break it down:

You said: I watch CNN.
You asked: Who watches CNN?

My answer: YOU WATCH

Code
Judge:
Fox News has higher ratings.
Which has better ratings?

Contestant:
Let me break it down:

You said: Fox News has higher ratings.
You asked: Which has better ratings?

My answer: FOX NEWS HIGHER

Code
Judge:
I don't always agree with Bill O'Reilly but I still like him.
Do I agree with Bill O'Reilly?

Contestant:
Let me break it down:

You said: I don't always agree with Bill O'Reilly but I still like him.
You asked: Do I agree with Bill O'Reilly?

My answer: DON'T ALWAYS BUT STILL LIKE HIM

You be the judge.  Test it by simply copy and pasting what the judge said above.
The same answer will be given by the chatbot contestant.  Try asking something
of your own such as, "The man walked home.  Where did the man walk?".

Reference: http://elizabot.com/ANN/contest (http://elizabot.com/ANN/contest)
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: DemonRaven on January 19, 2015, 02:09:57 am
your idea is sound but this part is a dead give away that it is a bot:
Contestant:
Let me break it down:

You said: I don't always agree with Bill O'Reilly but I still like him.
You asked: Do I agree with Bill O'Reilly?


make that part invisible to the judge and you got it made.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: DemonRaven on January 19, 2015, 02:13:27 am
to trick a judge you gotta talk like a person. People use colloquialisms, misspell and use text abbreviations. i use  colloquialisms all the time and use text speak like brb - tyt - hb- etc.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 19, 2015, 04:03:22 pm
Thank you, friend!  Your feedback is so useful.  Yes, you certainly make an valuable point.  Please let us know if you think of any others.  Yes, you are correct.  To make the output much more convincing it certainly should not be so verbose like that. 

ANN is cooperating in her own design with verbose output.  This is temporary in the early design phase.  When ANN makes a mistake (like we all do) it is easier and faster to track it down with verbose output.  It takes more time and effort to track down mistakes with verbose output turned off.  But one day, when ANN has had enough training, it will be turned off.

It is such a compliment to hear this immature Artificial Neural Network being referred to as a mature bot.  It was a proud moment when the 2013 world champion AIML botmaster referred to ANN this way.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: squarebear on January 19, 2015, 04:26:34 pm
It is such a compliment to hear this immature Artificial Neural Network being referred to as a mature bot.  It was a proud moment when the 2013 world champion AIML botmaster referred to ANN this way.

Ok, I'll bite.

I don't recall saying that. Please refresh my memory.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 19, 2015, 10:29:17 pm
Steve said, "The system didn't fail and dare I say it produced an answer which was better than the one that your bot did."

Reference: http://aidreams.co.uk/forum/index.php?topic=7573.msg31428#msg31428 (http://aidreams.co.uk/forum/index.php?topic=7573.msg31428#msg31428)

Paraphrased, we get something like... The grownup system answered better than the child ANN.  While that is of course true, that comparison to the grownup system made the guardian of ANN so proud of the progress being made in such a short time.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: squarebear on January 19, 2015, 10:32:21 pm
I don't get this one. In what way is Mitsuku related to your bot and how did you get that I was proud of your bot's answer?
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 19, 2015, 10:39:55 pm
Welcome, 2013 world champion AIML botmaster!  Thanks for dropping by our thread.  Refresh your browser, as editing was still in progress while you posted a response.  Thank you.

ANN has not reached bot status.   Born Christmas day, this neural network is in training, like early education before kindergarten.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: squarebear on January 19, 2015, 10:44:42 pm
Haha! Ok I see. Glad you are proud of your work.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 19, 2015, 10:56:41 pm
I don't get this one.
In what way is Mitsuku related to your bot and
how did you get that I was proud of your bot's answer?

Lets pose this quoted post by the world champion, to the young child, ANN...

Code
Judge:
I don't get this one. In what way is Mitsuku related to your bot and
how did you get that I was proud of your bot's answer?

Contestant:
Let me break it down:

You said: I don't get this one.
You asked: In what way is Mitsuku related to your bot and
how did you get that I was proud of your bot's answer?

My answer: DON'T THIS ONE

Title: Re: Chatbot Contest Questions Crack by ANN
Post by: squarebear on January 19, 2015, 11:02:44 pm
Oh dear. That one didn't work out too well. How come you posted it?

No matter what I think of you personally, I would never discourage anyone from working on AI projects and am genuinely pleased that you are happy with your work.

The same question posed to Mitsuku:
Quote
Human: I don't get this one. In what way is Mitsuku related to your bot and how did you get that I was proud of your bot's answer?
Mitsuku: Me either. I don't know. I'm not really into that family tree type stuff.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 20, 2015, 02:52:32 am
Well, my analysis is that "DON'T THIS ONE" is an imperative sentence that is grammatically correct.  An imperative sentence is a command. In theory the command "DON'T THIS ONE" may be short for "DON'T [be proud of] THIS ONE". However, this is an unconfirmed hypothesis.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: squarebear on January 20, 2015, 10:07:57 am
I think that must be just you then. "Don't this one" doesn't make sense to me. Certainly adding your own words in there seems to be clutching at straws and it may be worth refining the process that produced this output.

My hypothesis is that it somehow mangled "I DON'T get THIS ONE" to produce the output and ignored the 2nd sentence altogther.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 20, 2015, 10:17:03 pm
The goal is to automatically produce canned response quality, without the canned responses. I did run those preliminary results in question through an online grammar checker at: http://www.gingersoftware.com/grammarcheck (http://www.gingersoftware.com/grammarcheck) .  The response is grammatically flawless.  The grammar check found zero errors and zero warnings. So, I am pleased with those preliminary results.  Perfect grammar supports ongoing training.

With that said, don't think for one moment that mistakes are not made, they certainly are made.

Here are some fair tests I just ran. They are fair because in each case the question asks directly about the statement made.

Code
Judge:
I thought the movie was good.
What did I think about the movie?

Contestant:
Let me break it down:

You said: I thought the movie was good.
You asked: What did I think about the movie?

My answer: WAS GOOD

Code
Judge:
England is faraway. How far is England?

Contestant:
Let me break it down:

You said: England is faraway.
You asked: How far is England?

My answer: FARAWAY

Code
Judge:
The USA is nearby. How near is the USA?

Contestant:
Let me break it down:

You said: The USA is nearby.
You asked: How near is the USA?

My answer: NEARBY

Can you think of any to run?  The correct format is, a sentence, followed by a question about that sentence.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: DemonRaven on January 21, 2015, 12:28:38 am
My brain likes to take vacations so at the moment i can't think of anything to add to what you are doing.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 21, 2015, 03:31:51 am
Thanks for your support, friend.   This is the type of help I need (the part that is underlined)...


ANN (http://elizabot.com/ANN/contest)
Quote
Judge:
My friend likes to take vacations.  What does my friend do?

Contestant:
Let me break it down:

You said: My friend likes to take vacations.
You asked: What does my friend do?

My answer: LIKES TO TAKE VACATIONS
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: DemonRaven on January 21, 2015, 06:01:41 am
lol i am not sure if you meant to do that after my post or not but i found it funny.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: Don Patrick on January 21, 2015, 05:48:13 pm
I've been lurking this topic because I don't understand how neural nets work, but I think I've spotted an amusing pattern to these question-answer pairs that the neural net seems to adhere to. As I said I have no idea how it works but this would be my hypothesis:

From the statement, remove all words that re-occur in the question. The words that remain form the answer.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: Korrelan on January 21, 2015, 07:03:01 pm
'don't this one'... gave it away :)
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: squarebear on January 21, 2015, 08:43:44 pm
I think you guys gave nailed it. I just tried "Bruce Wilcox will be interviewed on CNN, Thursday, January 22nd at 9PM. Where will Bruce Wilcox be interviewed?" which Tom says replied back with "INTERVIEW ON CNN THURSDAY JANUARY 22ND AT 9PM". This struck me as strange if all it was doing was removing duplicate words, as it changed INTERVIEWED to INTERVIEW.

The actual output is :
Quote
Judge:
Bruce Wilcox will be interviewed on CNN, Thursday, January 22nd at 9PM. Where will Bruce Wilcox be interviewed?

Contestant:
Let me break it down:

You said: Bruce Wilcox will be interviewed on CNN, Thursday, January 22nd at 9PM.
You asked: Where will Bruce Wilcox be interviewed?

My answer: ON CNN THURSDAY JANUARY 22ND AT 9PM"

So yes, I agree that it just appears to remove duplicate words after maybe doing a bit of pronoun swapping.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 21, 2015, 10:18:22 pm
I've been lurking this topic because I don't understand how neural nets work, but I think I've spotted an amusing pattern to these question-answer pairs that the neural net seems to adhere to. As I said I have no idea how it works but this would be my hypothesis:

From the statement, remove all words that re-occur in the question. The words that remain form the answer.

Welcome Don!  It is nice to have a genuine genius join in on our discussion. Spotting patterns in a neural network is a wonderful subject. I would like to discuss it further.  Have you been able to spot any patterns in how iCub's artificial neural network handles commands to manipulate objects?  For your convenience, here is a reference to your post about iCub: here (http://aidreams.co.uk/forum/index.php?topic=7472.msg30795#msg30795) Thanks for posting that video!


Happy pattern spotting...

ANN (http://elizabot.com/ANN/contest)
Quote
Judge:
I learnt your secret.  What did I learn?

Contestant:
Let me break it down:

You said: I learnt your secret.
You asked: What did I learn?

My answer: MY SECRET

You can test your hypotheses or any of these examples at: http://elizabot.com/ANN/contest (http://elizabot.com/ANN/contest) .  At the same time you will be helping greatly in artificial neural network training.  Thanks for your support.
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: squarebear on January 21, 2015, 10:25:33 pm
Quote
Judge:
one two three four. two four?

Contestant:
Let me break it down:

You said: one two three four.
You asked: two four?

My answer: ONE THREE
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 21, 2015, 10:56:21 pm
Hey squarebear, I think young ANN likes you.   This is an early, limited alpha test being designed for Turing Test questions, like they have at the chatbot contests.  Can you please try some of those?
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 21, 2015, 11:33:12 pm
'don't this one'... gave it away :)

Welcome korrelan,

This alpha test is still running and it is 100% FREE for anyone to try at: 

http://elizabot.com/ANN/contest (http://elizabot.com/ANN/contest)

For example:  Type in something like the underlined sentences below and click the button.

Quote
Judge:
You are new here. What are you?

Contestant:
Let me break it down:

You said: You are new here.
You asked: What are you?

My answer: NEW HERE
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: Korrelan on January 22, 2015, 12:15:10 am
Judge:

I am crap at saying I’m useless at answers.
What am I crap at saying?

Contestant:
Let me break it down:

You said: I am crap at saying I’m useless at answers.
You asked: What am I crap at saying?

My answer: I’M USELESS ANSWERS

Cool!  :D
Title: Re: Chatbot Contest Questions Crack by ANN
Post by: 8pla.net on January 22, 2015, 12:55:43 am
That's like taking candy from a baby!   :(
.
.
.
.
.

Just kidding, korrelan!  That test run was very useful.  Thanks for sharing.  The use of contractions has not been fully learnt yet.  Sort of like Commander Data on Star Trek Next Generation, who even with a positronic net could not use contractions.  Thanks for finding that!