AI Zone Admin Forum Add your forum

NEWS: Chatbots.org survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

Harumi
 
 
  [ # 16 ]

Jean,
Harumi is working nice ‘till now.
I am trying the english version because of my lack in French. But till now it is ok.

Well. I’ve seem that Harumi has some kind of expression/manifestation of feelings related to fuzzy logics, bout how do I activate it? ( in what situation Harumi uses it?)
Or feelings device is more active on French version?

I was planning to do something similar, but it is only in the ideas field. I am planning to do something in a simple robot with some detectors in it’s skin. I am planing to do a hugable toy for children.
I was planning to use some kind of device like the expression of “feelings” and keeping memories of this “feeling” related to each person related to the toy. Well… Something like in The Sims ( the game).

What do you think?

 

 
  [ # 17 ]

Fatima : It reminds me Tamagotchi, something from 90’s. It’s a good idea but like many good ideas, it’s difficult to make it real. Harumi idea is very simple : speaking with a robot. Making it real is really hard… if not impossible. Anyway, as long as you work a lot and believe in your idea, you can make a dream come true.
About the way Harumi use feeling :
Each stimulus in Sheet1 is linked to 1 or many sentences. The second column is reserved to “feeling”, aka a number positive or negative making global feeling of Harumi going up or down.
There is also a databank in which images and sentences from Harumi are linked to a certain level of mood :
Example : between 0 and 30, her feeling is bad. If you ask her how well she is, she will answer something like “I’m depressed, I’m sad…’. Feeling is a number between 0 and 100, 100 the best feeling.

When speaking to her, if you say a stimulus where the feeling is say 4, then her global feeling grow by this number.
If you use a smiley in a sentence, you move up or down her global feeling. When you teach her a sentence in which she detects a smiley, then she record the number linked to the smiley :
Example :
User : I’m happy smile
Harumi : You are happy?
User : I feel fine
Harumi : All right.

In Sheet 1, you will see :
Column 1   Column 2   Column 4
I’m happy   1           You feel fine.

Now she “understands” that “I’m happy” means a good feeling.

Note that global feeling goes down by a number of 1 or 2 at each launching of Harumi. This way, if the user use Harumi just like a way to launch applications and internet site, her global feeling will go down slowly until she will say to the user how sad she is.

Fuzzy logic has nothing to do with feelings. It’s a way to give a number value by matching user sentence and Stimulus database instead of a simple “true” or “false”.
For example, if you have in the stimulus database :

Column 1               Column 3
pictures AND family -> c:\pictures family
pictures             -> c:\my pictures
pictures AND family AND India -> c:\my pictures\India

If user say : “Please show me family pictures from India”
then Harumi gives
2 points to first entry
1 point to second entry
3 points to third entry
so she will open the c:\my pictures\India repertory

My precedent algorithm was boolean logic so she opened the 3 repertories since “True” was given to each of those 3 stimulus. Now you get how interesting is fuzzy logic.
fuzzy logic works also inside a stimulus :
If you have 50 sentences linked to 1 stimulus, Harumi will pick up the sentences with the highest probabilities of matching with user question.

 

 

 
  [ # 18 ]

Jean,
It is hard but not impossible.
I didn’t know about affactive computing. But a friend sent me a link about MIT works about AI and artificial emotion.
http://www.ai.mit.edu/projects/humanoid-robotics-group/kismet/kismet.html

And to my surprise:
A huggable robot for children http://robotic.media.mit.edu/projects/robots/huggable/overview/overview.html

But I think that the basis to artificial Emotions are in Fuzzy logics. Stimulus and answer is a nice beggining for bigger AI and AE.
Harumi is really amazing.

 

 
  [ # 19 ]

  A new version of Harumi will be release soon, soon meaning this month. 3 improvements :

1) I’m currently working on a new algorithm to analyse sentences based on fuzzy logic. Harumi will be able with this algorithm to pinpoint easily the best sentence in her memory that will match the user one.
Example of what you can expect from this algorithm :
User : my car matricule is 4512 ZZ 99 (then user click on the smiley icon where a student appears instead of just pressing enter. This way, Harumi will put in her memory the whole sentence without asking for more details. Note that it will be a new way to enter sentences)
Harumi : I’ll remember that. (To confirm the sentence has been recorded. In Sheet 1 of memory file, you see at this point in column 1 : “my car matricule is 4512 ZZ 99” and in column 5 : “your car matricule is 4512 ZZ 99”
User : What is my car’s matricule ?
Harumi : your car matricule is 4512 ZZ 99
User : Thank you!
Harumi : you welcome!
User : my wife’s car matricule is 777 ZZ 99 (then click on smiley student icon)
Harumi : All right.
User : Can I test you?
Harumi : Go ahead.
User : I’ve found a car with 4512 ZZ 99 as matricule. Whose car is it?
Harumi : your car matricule is 4512 ZZ 99
User : what about 777 ZZ 99 matricule?
Harumi : your wife’s car matricule is 777 ZZ 99
User : bravo
Harumi : you kidding me?

As you see, no need to enter manually something like “my car AND matricule” in column1. Everything get easier and more natural. Moreover, my first tests confirm something else : this algorithm is far faster than the one currently used.

2) I’m also working on “Taboo” algorithms. It’s a way to prevent Harumi from using twice the same answer. After an answer, Harumi will be able to memorise her last answer in order to unselect it for the next “round”. Note that neurones are functionning a bit this way.

3) Last thing : When Harumi will mention a URL, the URL will appear as an hyperlink clickable with mouse.

 

 
  [ # 20 ]

Today advancements in my personnal version (still not online) :
- Taboo algorithm now working (When Harumi is sollicited twice on the same stimulus, she will never answer twice the same thing if more than 1 answer are available)
- URL clickable in Harumi answers is functionnal (pretty cool : you click on the URL, Mozilla/Internet Explorer opens and get on the site)

- I go on in testing with the new fuzzy logic algorithm. Many bugs have been found but Harumi now “understands” sentences that were before impossible to compute.

More interesting, I had the idea of a more powerfull fuzzy logic algorithm (involving statistics and recursivity…) which is for the moment only on paper, (a simple pre-version of it is tested). This algorithm could allow Harumi to undestand complex sentences or to answer to questions related to a long text, a thing unthinkable for the moment.
The power of this algorithm comes from the fact that Harumi would be able to make the difference between something important or not inside a sentence….

With this kind of algorithm, a sentence like “I think I’ll watch again jurassik park this evening” would be treated differently from now : Harumi would automatically (meaning without any pre-dictionnary help from the user/programmer) consider “jurassik park” as the most important part of the sentence. Note that I said the most and not the only!
Harumi would be able to create a hierarchy of informations, a dynamic one depending on the question asked, the context and precedent chats with the user!

 

 
  [ # 21 ]

Jean!!!

Impressive!

Congratulations!

Almost can’t wait to test the news.

smile

 

 
  [ # 22 ]

New version expected be online this evening or tomorrow. List of improvements :

1) For the first time, fuzzly logic partially applyed to stimulus. You can memorize a complex sentence by clicking on the student-smiley icon and the sentence will be recorded. Then if you ask a question with more than 2 words matching the sentence, there will be strong chances that Harumi will “get” what you mean and answer properly.
I’m working on this function a lot, hoping to improve it in a later version.

2) Calculs : proportionnality and variables
Example :
User : 1€=1.4$
Harumi : I like learning
User : 47€
Harumi : 65.8$
User : x=17
Harumi : Ok
User ! y=3
Harumi : I get it
User : x+y
Harumi : 20

3) Bug correction : the way definitions are treated before matching with stimulus
Bug correction : Internet detection now working properly

4) Internet site appearing in answer window are now clickable

5) Emoticons : if you click on smiley happy or sad, Harumi will propose you something linked to your mood.

 

 
  [ # 23 ]

New version released today (30.07.2011) !

 

 
  [ # 24 ]

Next Harumi version will be released before this week’s end (7th of august).

The fuzzy logic algortihm will be improved :
In the version released yesterday, fuzzy logic is applyed only on the couple stimulus-user request.
In the next version, Harumi will explore all available answers for each stimulus.
Example :

Stimulus                 Answer 1                               Answer 2 ......                      Answer n
paul                     he lives in japan                       he likes playing video games
paul likes japan         paul likes japan

In current version, if you ask :
“Does paul live in Japan?”
Harumi answers “Paul likes japan”

The reason is that 2 points are attributed to stimulus 2 and only 1 on stimulus 1.
By exploring all answers available for stimulus 1 and 2, the answer becomes different :
3 points are attributed to stimulus 1 + Answer 1 and only 2 for stimulus 1.
With this algorithm, Harumi answers “He lives in Japan”.

 

 

 
  [ # 25 ]

Last Harumi version released today (9th august 2011) http://jeffelson.e-monsite.com/

 

 
  [ # 26 ]

Jean-François, I added the URL to your post so that interested parties can head straight there without going back to the first page of this thread to get the link. If that’s an incorrect URL, please let me know, so that I can fix it. smile

 

 
  [ # 27 ]

Hi Jean!

She is having a bug that she was not having in the previous version:

In the previous english version she was able to access the internet links related to the icons.
I thought I could take the links and put in the memoire file, but I am not sure if it will work.
Is it the correct thing to do?
Anyway it would be a problem for a person who wasn’t familiar to the older version.


What more can I say?
Congratulations! I love your work with Harumi. She have great possibilities!

I’ve seen in the site that you want to build something like a self-learning log..
It might be a hard work! But don’t give up!
If it is your dream, go on and take it!

I’m a Harumi’s fan!

 

 
  [ # 28 ]

Hey Jean-François

Thanks for keeping us posted! Don’t forget to upload the directory entry once in a while:
http://www.chatbots.org/chatbot/harumi/

 

 
  [ # 29 ]

Dave Morton : Thank you, the URL is correct

Fatima Pereira : It’s indeed possible to transfer links from one memoire file to another. Thanks for your report smile

Erwin Van Lun : Thanks. The link is still correct. Note that beside upgraded versions of Harumi, I give on my site informations about the development and current/future features with more details than here.

 

 
  [ # 30 ]
Jean-François Battistini - Aug 11, 2011:

Erwin Van Lun : Thanks. The link is still correct. Note that beside upgraded versions of Harumi, I give on my site informations about the development and current/future features with more details than here.

This thread will be read by quite a few people during this period, which is cool and quite interactive, but the directory is far more visited (Chatbots.org has over 1300 visitors a day), so if you’d like to have more traffic, I’d certainly recommend to update your entries and add entries for other languages! Nothing mandatory here, just a suggestion.grin

 

 < 1 2 3 4 >  Last ›
2 of 14
 
  login or register to react