New version released today, features list on the precedent post plus

1) Better AI with a Pavlov like conditioning; now not only words point to other words, but each link has a weight which fluctuates automatically with user interactions and time : When you ask Harumi a question and she answers wrong while you know she’s supposed to know, click on the “No” icon. (Note that you can click on “yes” to reinforce a link).
Here is an example of Pavlov condionning:
USER : What is the name of apple chatbot?
HARUMI : Your name is John.
USER : (click on No icon, not only because the answer is wrong, but also because the user knows the good answer is in memory)
HARUMI : Sorry, let me think again… Apple chatbot is Siri.
USER : thanks.
HARUMI : You welcome.
USER : (change subject icon)
USER : What is the name of apple chatbot?
HARUMI : Apple chatbot is Siri.  (Harumi learnt from her error)
USER : What’s my name?
HARUMI : Your name is John (nevertheless, Harumi still considers the “name” word)

2) Now Harumi speaks! A connection to windows narrator allows this, just click on the new icon, a speaker which appears in pair with mike icon.

3) Moreover, Harumi is finally fully oral interactive! This function works better with a headphone for the user in order to avoid “self-speaking”. Just click on mike icon and on speaker icon. Note that you’ll need to use Windows or any software vocal recognizing software to enjoy this function.

4) Datagrid are fullscreen, easier to navigate.

5) Keyboard sound while Harumi is “typing”. This way, you will know everytime she says something even if Harumi is in the taskbar.

I have identified several important issues after 08.23.2015 release. Those bugs came from the “Pavlov” function described in the precedent post… and are now fixed!
I recommand you to download this new version!

Jean,

The chat feature seem to work pretty well although it does display a bit of repetition.

The Edit button brings up Form 3 and Form 2.

Then a pop-up JIT error indicating the string was not recognized as a valid Date Time.

I ran the program several times and received the same error with regard to the Edit button.

My system is on Windows 7 Premium Home, 1 TB drive, 8 gigs Ram and plenty of disc space.

The English version still has a lot of French intermingled within the spreadsheets from what I could see during the error message. I think the date format is still the culprit within the software.

It would be interesting to hear from others in this regard.

Art Gladstone : I’m working on the full translation of the English version. Nevertheless, I have no way to fix the datetime bug since it’s probably linked to OS language. I have only access to french PCs. Harumi was tested with Windows 8.1 with i7 and the old Windows Vista with Pentium. I didn’t witness this datetime error while clicking on “Edit” button. Moreover, I don’t see any link betweens datetime access and datagrids edition…
Thanks Art for your report, if anyone has the same bug, thanks to signal it here.

Something else, Art : try clicking on “save” button before clicking on “edit”. When you click on save, you update the datagrid with the current chat and things you learnt to Harumi, while clicking on “edit” without saving before cancel all interactions done during the current session.

Jean,

That’s nice but if you recall I had the same issue with regard to the Date Format a year or more ago when I first brought it to your attention.
Changing all the internal text to an English format might only serve those who’s country(s) have a Date format different from the standard MM/DD/YYYY format that the majority of the USA uses, if I’m correct.

Therefore any changes, might still result in a JIT Error message popping up.

Sorry, it is the nature of the beast, I’m afraid. Keep up your good work and best of luck!

Art : All right, as my brother in law lives in USA, I’ll ask him to bring me a low cost US laptop in order to test Harumi on it as soon as he comes to Europe…

Another major update will come in several weeks, with all new features AI related :
- Harumi will be able to put aside parasites words in a sentence without passing by a fixed dictionnary or database frequency.
- Tolerance to syntax errors and variations of a word (involving plurials, suffixes or verbs) will be far better, and will work along with context.
- Symetric sentences (I am -> You are ....) algorithm corrected

Hard to believe that no one else from the USA has tried this latest version of Harumi.

If they have, please let Jean-Francois know of any errors encountered.

How about the possibility for a new user of Harumi to be asked a couple of questions before any chat begins.

Harumi: Hello there! What is your name?
User: My name is Fred.
Harumi: Nice to meet you Fred. What country are you in?
User: USA
Harumi: Wonderful! I often chat with a lot of people from the USA.

(Now, Harumi can make any program changes and SET the Country Code or adjust the Date Format accordingly.)

Just a thought…

I’m going to check with a few people…

I have tried her out and she is pretty good as a assistant.

Sheryl Clyde (#2) - Sep 27, 2015:

I have tried her out and she is pretty good as a assistant.

Okay i tried out the new one and it gives exception errors when you try to go into edit. The previous version did not.

Sheryl Clyde : Thanks for the info, I’ll try to figure out this next week. I suppose it comes from the database itself…

I have worked a lot for several weeks on the upcoming algorithms explained in the former post.
After several tests, I finally abandonned because of important bugs.
Nevertheless, I reoriented my searching on something I would call “deep thinking” analysis.

Example of “deep thinking” :
User : What’s Olaf color? (Note that Olaf is a snow man in “Frozen”)
Harumi : white color

This example shows that the answer is reachable only after a cascade of simple déductions, while context analysis is still functionnal :
1) Olaf is a snow man
2) snow is white
3) white color

I can’t say more for the moment, since my main issue seems to be now the us version of Harumi and its edit button bug…

New version released today! Now Harumi is able to make a cascade of déductions out of her database.
I also deleted the date/time written in the “Form3” datagrid, I hope it solves the date bug. Thanks to report me bug, especially date one. In case it is still there, I will simply delete all algorithms involving dates for the US version.

Someone mailed me about a bug in last Harumi version : I forgot to add sound file (writing machine.mp3) which produced slowdowns as Harumi tryed to access the missing file. I corrected the error, Harumi is now functionnal! Thanks to the user who reported me this error.
By the way, feel free to report bugs on this site or by mail!

Jean,

It seems I spoke too soon. The GUI and conversational interaction went without flaw with the exception of every web site being French (even weather). It would be nice for us English speaking Americans to change some program settings to allow access to USA based sites and weather, Newsfeeds, etc.

Unfortunately, Somewhere within Form 2 we are still experiencing an invalid DATE bug.

Since the program format is not MM/DD/YYYY, I fear this error will never completely go away.

Best of luck,
- Art -

Hi, some very good news for Harumi :
I have discovered a way to implement deep contextualisation with a linear database! I had the idea while taking a coffee with my childrens! I didn’t even think about coding, and boom, eurêka!
I used to think it was totally impossible to memorise discussion state without a tree database structure!

The consequences change almost everything. I had to recode enterily Harumi around the new database structure with this new algorithm. I tested it and it works!

Now what does it change?
1) Harumi is very fast! answer is written in an instant, even with many many stimulus (hundreds)
2) Answers are more precise especially with complex requests : “What is the wifi code at my wife’s father house on reunion island?”
3) Data are readable for the user! No more word base stimulus. Now each stimulus is a list of sentences. This way, you can edit and correct a stimulus in the database very easily
4) Stimulus are linked to a theme and not a word. Now, when you want to add new informations, it will be far more natural! Example :
“Bladerunner is my favorite movie.” -> record icon
“Harrison Ford is Deckard, a policeman payed to kill replicants” -> record icon (Harumi proposes to link this information to the precedent)
“Sean Yound is Rachel, a replicant who loves Deckard” -> record icon (Once again, Harumi proposes a link to the theme).
Here, you see that is now possible to make links between stimulus with not any common word, something impossible with tree base structure!
5) Deep context available is still available despite this linear structure! This is the most important point:
It is possible to create a linear database with deep contextualisation!

That sounds very promising as long as other issues (like date customization and other user based settings) are addressed.
It seems that your ‘discovery’ opens some new doors not only for Harumi but perhaps other database oriented bots.

When is this new version due for release?

[Sean Young]...just saying…;)