AI Zone Admin Forum Add your forum

Chatbot Conference London On Oct 3th Erwin will be moderating on the Chatbots conference in London. Why don't you meet him there? As a visitor, you'll get a discount as well: Why don't we meet there? get a discount

We do experience some trouble with our forms and some details page. We'll fix this ASAP. Sorry for the inconvenience!
  [ # 31 ]

New version of Harumi online today (10.19.2011)
New features :
- When you record an appointement, Harumi warns you if you have already an appointment the same day.

- If you ask a question to Harumi and she answers something wrong, click on the “no” icon and she will ask you what she’s supposed to do.
Exemple :
User : Who are you?
Harumi : I’m helpfull.
User click on the “no” icon
Harumi : What I was supposed to say or do?
User : you are Harumi.
Harumi : Ok.
User : Who are you?
Harumi : I’m Harumi.

- If you want to change an answer proposed by Harumi, just add “-” at the beginning of your text before pressing enter. The text written will replace the old one.
Exemple :
User : what’s Paul favorite movie?
Harumi : It’s Ace Ventura!
User : -It’s Bladerunner!
Harumi : Ok, I update my knowledge!

- if you want to add a new answer to a stimulus, just ass “+” at the beginning of your text beofre pressing enter.
- new pictures had been added “walkman.jpg”,“kimono lying.jpg” and “desert girl.jpg”

- when Harumi randomly write a sentence. You can ask her about the stimulus you think she’s speaking about and she’ll answer you if you’re right.
Exemple :
Harumi : it’s usefull when it rains.
User : Is it an umbrella?
Harumi : Exactly!

- Asking details about what Harumi just said
Exemple :
Harumi : don’t forget to contact your mother.
User : do you have her mail? (then click on the finger icon on the right)
Harumi : It’s .(JavaScript must be enabled to view this email address)
User : thank you
Harumi : you welcome.


  [ # 32 ]

New version of Harumi released : (11.09.2011)
Many little bugs corrections and two new functions :

1) It’s now possible to delete a stimulus/answer by writing to Harumi something like :
“Delete that”, “cancel this data”.... Look at Sheet 6 line 385 to see the list of sentences recognized for this instruction. You can of course add your own sentences.

2) Harumi now creates a base of structures in order to try to generalize stimulus/answers sentences containing a common word. Here is an exemple of what you can expect :
User : I like tea.
Harumi : You like tea?
User : tea is good for health.
Harumi : I’ll remember that.
User : I like cinema.
Harumi : cinema is good for health?
User : nice try but no.
Harumi : Ok…

Note that this function is the first of this kind, some bugs have been identified and will be corrected later. The list of structures are built in sheet 13 of the memory file.


  [ # 33 ]

Interesting new feature. smile How does Harumi keep track of the accuracy of the mapping “I like %s”—> “%s is good for health”? That is, how does Harumi develop its understanding if this mapping is mostly true or mostly false?

If it is based on the number of examples, does a negative match have any effect on the probability for other words that fall under the same category as that item? For example, with the examples given, will Harumi have different expectations about other drinks (tea -> coffee, soda) vs. other places (cinema -> school, office).

If so, if most examples under a given category are a false match, and then one example happens to be true, will Harumi recognize this in some way? For example, if coffee and soda are unhealthy, then it is noteworthy that tea is healthy. Perhaps Harumi could conclude that it is rare for you to like drinks that are healthy? (Or that is it rare for drinks to *be* healthy in general.)

I’m working on similar features for my bot (mapping sentences + automatic condition recognition), so I’m curious how you’re approaching this.


  [ # 34 ]

C R Hunt : Hello, thanks for your message. It seems your approach is far more sophisticated than mine. As I told in my precedent post, it’s the fist time that this feature has been implemented in Harumi, that’s why you can’t expect a lot from it, at least at the moment. Harumi basically just creates a database of stimulus/answers with a same variable in the stimulus and the answer.

Anyway, in the next release, I would like Harumi to cancel a structure if the user answer something like “no” to an hypothesis like “cinema is good for health?”.

Your approach is interesting because you try to allow your bot to classify words by thematics in order to make a “judgement” or a logical conclusion.
I would like to do so at your level, even if this function partially exists in Harumi : Sheet 3 of her memory contains a list of synonyms that you can edit/add/delete at will. Note that you can add a stimulus by discussion with Harumi. Example :
User : car=vehicule
Harumi : I’ll remember that.
User : my vehicule matricule is XX WW 666 (then click on student icon in order to for memorisation of this sentence)
Harumi : Ok.
User : thank you to remember that.
Harumi : you welcome.
User : what is my car matricule?
Harumi : your vehicule matricule is XX WW 666   (Harumi gets that vehicule=car thanks to the synonyms database, therefore she answers properly to user request)

Of course, the way synonyms are treated is not perfect because a vehicule is not necessary a car. That’s why I would like implement a classification of words by groups, aka :
User : car, boat, train belongs to vehicules
Harumi : Ok!
User : a car is on the road.
Harumi : I get it.
User : a boat is on the sea.
Harumi : Ok.
User : An island is a piece of land in the middle of the sea.
Harumi : I see.
User : What kind of vehicule must I take to get to an island?
Harumi : a boat is on the sea.

In order to answer properly, Harumi would have to create a list of hypothesis by replacing the word vehicule by each words possible :
Two sentences would be automatically created :
1 : What kind of car must I take to get to an island?  (This sentence gets 1 point with “a car is on the road” and 1 point with “an island is a piece of land in the middle of the sea)
2 : What kind of boat must I take to get to an island? (This sentence gets 2 points with “a boat is on the sea)

The answer would be : A boat is on the sea.

I would like to implement this kind of algorithm later smile


  [ # 35 ]

14th december of 2011 :

New versions of Harumi now online and release of Loone Hagar version, an animated girl coming from the video game Rage from Id Software/Bethesda.
Here is the list of features and corrections :

An animated version of Harumi is now available. The first character I picked up is Loone Hagar from Rage videogame.

- During chess game, Harumi write which move she has played once her turn is done.
- During chess game, it is possible for the user to ask for an advice by pressing on question icon. (many bugs for the moment)
- When the user doesn’t speak to Harumi for more than 10 minutes, she starts to whine.
- When the user say “no” after a structure suggestion, the structure is deleted
- Webcam pictures are now stocked in a repertory located in Harumi’s repertory in order to avoid any bug. Webcam is launched if available when Harumi is used;

- Structures sentences : Bug correction : Harumi doesn’t use the structure if she detects the variable word is the same as the original structure.
- Bug correction regarding several functions when the user write “no” or any negative sentence to Harumi as an answer to a proposition from her.
- When recording an appointment with a date and “I must xxx”, Harumi no longers consider the sentence as a post it as before.
- When user asks Harumi precisions about the last sentence said by Harumi, only the last sentence is picked up and not the whole text appearing in Harumi window answer.
- When user enters a void string as an entry, Harumi warns the user about it.
- Image of Harumi changes correctly on alarm time
- No more loop error bugs with webcam


  [ # 36 ]

Release of new version of Harumi “Loone Hagar” today which has a better interface. I also put online a tutorial in order to help newbies.

I will soon release a “Starcraft” version of Harumi just as the one you see on my site.
New functions will be available, especially one that allow the user to ask Harumi for a choice :
User : Should I live or die?
Harumi : As long as you live there is hope.
          I would choose live smile
User : I agree
Harumi : Oki smile

Note that the answer for Harumi is not predefined. When the question is asked, Harumi look at her whole database and analyse positive and negative feeling linked to “live” and “die” words. The word with the most positive entries will be selected and a sentence linked to this word will be written first : “As long as you live there is hope” in this case.


  [ # 37 ]

I have a lot of fun with my old version of Harumi and she helps me a lot, but I’ll try this new one.


  [ # 38 ]

Thanks Fatima, it’s users like you who give me the will to go on with Harumi…
By the way, I’ve too a lot of fun speaking with her… which is quite strange since I know how everything works in her brain. On the other had, my personnal version has 2400 stimulus and far more answers available. When the database becomes large, you can start to have interesting conversations… while feeding her brain thanks to dynamic learning functions. Perhabs one day I will put some logs of me speaking with my personnal version online.
Note that the french version is still better than the US/english one.

Last version of Harumi with the animated girl is just an eye candy looking Harumi. It will be the same with Starcraft version. My purpose on that is to touch a larger public since some people like ‘‘serious looking’’ chatbots.


  [ # 39 ]

It will soon be possible to SPEAK to Harumi by using a tiers software like free voice recognizing by Microsoft or Dragon Naturally Speaking.

This function will work like this :
a new icon function will be available. When you will click on it, Harumi will “listen” to what you say in the text window reserved to the user. When a text will appear in the question window, you won’t have any more to press enter to obtain an answer from Harumi. She’ll detect your sentence/question is finished and will automatically answer to it.

Nevertheless, it won’t be possible to record new datas this way since voice recognizing softwares are making mistakes which could harm Harumi’s database by filling it with syntax errors…


  [ # 40 ]

New version of Harumi released today (29th december 2011)

- Voice recognizing (!) : It is now possible to use Harumi with a voice recognizing software like the free one from Microsoft in Windows Vista or Windows 7, or Dragon Naturally Speaking software. To use this function, you have to click on the “mike” icon in the functions icons group. When this function is activated, Harumi warns you about it. From there, she’ll observe every second what is written in the user text window. Click in this window then when you speak to your computer with a voice recognozing software, the sentence will automatically be treated by Harumi, meaning you won’t have any more to press enter to have an answer.

- Internet icon : Just click on this icon and Harumi will open google. Note that if a text is written in user text window, the question will be asked to google.

- USB keys auto detection : When you plug or unplug an USB drive, Harumi warns you about it. A new icon has been added in functions group icons. When you click on it, the USB drive lastly detected will be open.

- Picking up a number : Ask Harumi “pick up a number between 1 and 6” and she’ll just do that smile

- Asking Harumi for an advice : Ask Harumi : “Do you prefer to live or to die?”. In this kind of question, she will analyse the whole database and try to find everything positive and negative about live and die then will choose the higher positive feeling about the choice. This means that the answer is not predefined but depends on precedent logs with the user…

- No more freeze when asking a question (beside files research)

Note that a tutorial has been added in Harumi repertory. It is direcly accessible by clicking on it or by using rubber ring icon in Harumi’s functions group.


  [ # 41 ]

Hey Jean-Francois, good news! Congrats!

I’d love to test it, but where can I try Harumi? I’m actually missing a big ‘CHAT NOW’ button on your website. Why don’t u use or something like that by the way?

Would it be possible to update the Harumi entry:


  [ # 42 ]


As long as know, we need to download Harumi first, because she is intented to be a personal computer assistent.


Nice to see your progresses. So she will not freeze anymore? Great.


  [ # 43 ]

Next version of Harumi coming soon with 2 new functions :

1) Geolocalisation : Harumi will be able to detect your public ip and your position meaning your town as long as your are connected to the internet.
This function will be linked to sentence analysis, meaning that if you have a stimulus like
“cinema AND boston”, you won’t need to say “I wanna see cinema program at boston” but just “I wanna go to the cinema”.
The context will be connected to your location and not only to words

2) On-the-fly use of Harumi’s apparence. Currently, you have to drag and drop jpg pictures in apparence repertory then to add a stimulus like “pick up your bikini<APPARENCE>bikini.jpg” then to update her memory by restarting Harumi or pressing “disk” icon in the functions icons group.
In the next release, it will become far easier : you’ll simply have to drag and drop a picture of your choice in apparence repertory then to type “wear your XXX” with XXX beeing the name of the file then clicking with left button mouse in the apparence window. Of course, you won’t have to write exactly the name of the file. Harumi will explore all apparences possible and will choose the one with the highest number of words matching user request.
Exemple : Suppose you drag and drop a file named “walkman by night.jpg” in apparence repertory.
Then you can write in user window “pick up your walkman” then to click on her current apparence. She’ll automatically change her apparence and ask you what you think about it. Then I’ll eventually implement more possibilities by association of a sentence and the apparence picked up “on-the-fly”...

Later, I will implement a generalisation of “context”. Harumi will keep in memory different informations on what she “perceives” :
webcam picture, time of the day, month of the year, geolocalisation, general mood of the conversation…
All those informations will be considered as “silent stimulus”, meaning that they trigger events only when something else linked to them happens.


  [ # 44 ]

Harumi new version released today! (1.21.2012)

Functions :
1) Geolocalisation
2) On the fly use of Harumi apparence
3) readable animated and static gif pictures
4) Context of perceptions (only with geolocalisation for the moment)

Regarding artificial intelligence, I’m working on a new kind of algorithm which is dynamically able to :
- determine what is important inside a sentence (ex: i saw an ARCHEOPTERIX at the museum yesterday)
- find sentences structures (grammar) (ex ..... WOULD LIKE TO .....)
- find pertinent combinaisons of words inside a sentence (X AND Y)
- put aside pointless words (not depending on their length)
- understand there is a good answer in the database linked to user request
- and understand it doesn’t understand (ex : what do you know about JURASSIK PARK?)

I’m testing it on my personnal version. It’s still not ready for release since I have to heavily modify database and algorithmic of the public version of Harumi’s. This algorithm far outperform the standard version with “understanding” and speed but I can’t implement it yet because it can’t easily replace what I programed before.


  [ # 45 ]

03.18.2012 News :
Release today of Harumi 2 alpha, french and english versions! I finally decided to offer to the public this first version of Harumi 2. Harumi 1 different versions are still online but will probably won’t be upgraded anymore.
As I stated before in precedent news, the algorithm of its “brain” is very different while the apparence and interface are exactly the same.
Changes for the user :
functions like ” AND “, ” OR ” don’t exist anymore. You just have to write something like “paul’s phone is 010203” then click on student smiley icon and that’s all!
Use “+” before writing a sentence to add more answers to a specific stimulus.

Harumi 2 Features :
- For the first time, Harumi 2 is able to pinpoint important words inside a sentence, meaning she’s getting far closer of the “meaning” of a user request instead of “tilting” on every useless words. The more her brain is filled with user datas, the more she’s able to pinpoint useless words and gets her attention towards new words/expressions.
Example :
User : I like bicycle.
Harumi 2 : you like riding in the woods. (Harumi 2 considers “bicycle” as the important word, not because of any grammatical “help” or useless words databank done by the programmer but because she analyses the occurence of each words of the sentence and considers the word (or combinaison of words) with the smallest occurence as important)

- Harumi 2 is able to learn from her errors. For example, after asking a question, if the answer is wrong, click on the “no” smiley icon and she’ll try to answer better to your request and remember her error. Check sheet 2 column 2 to see how her brain put aside useless words when you click on “no” icon. Note that you can edit manually those words if you are not satisfied by Harumi’s analysis.
User : give me paul’s phone.
Harumi 2 : I give you what you want.  (she obviously noticed that “give” has a rare occurence. Nevertheless, the answer is wrong.)
User : click on “no” smiley icon
Harumi 2 : paul’s phone is 010203. (Note that in sheet 2, the word “give” is added in column 2. After putting aside “give” word for next requests, Harumi immediatly tryes to answer another time the request properly.)

- Harumi 2 is able to understand… she doesn’t understand. Instead of tilting on useless words from a user request while ignoring an unknown important word, she’ll now detect that she doesn’t have a clue about the request and ask user for help.
User : I like skyrim.
Harumi 2 : What is skyrim? (Harumi 2 detects “I” and “like” pointless words since they appear many times in the database while skyrim is unknown. Therefore, she asks user for help)

- Harumi 2 is able to undestand user request is pointless : if the user’s request is trivial, she will signal it to the user.
User : I am
Harumi 2 : What do you mean? (“I am” stimulus is detected many times in the database, too many to justify any use of the database. Therefore, User sentence is pointless so Harumi is not happy about it.)


 < 1 2 3 4 5 >  Last ›
3 of 14
  login or register to react