UltraHal Update

[Art]

Robert,

Would you please advise your membership of the current status, progress or state of affairs regarding the development of Hal, it's character / avatar development, it's brain and overall projected time frame for arrival?

If there is anything that we might be able to assist with please indicate that as well.

Thanks as always for your time. We appreciate your efforts!

- Art -

[cload]

Robert,

Would you please advise your membership of the current status, progress or state of affairs regarding the development of Hal, it's character / avatar development, it's brain and overall projected time frame for arrival?

If there is anything that we might be able to assist with please indicate that as well.

- Art -

I'm not sure, but I believe that was indicated about a month ago.

And if you downloaded the Wikipedia plug-in you should know the type of help that Rob needs.

Money money money money money money money money money money money . Laugh out loud .

Sincerely, money any money honey I need more money   .

C load.

[Medeksza]

I'm focusing my efforts right now on what I can do myself in my spare time without hiring anyone, which is the Hal brain. I'm working on Hal's next generation learning algorithm which I'm quite excited about. Hal will look up conversational answers in its own online conversational database, database learned from the user talking with hal, twitter database, wikipedia, yahoo answers, and general bing search engine, and possibly other sources. User positive/negative feedback (thumbs up/down) as well as users selecting which data source provided the best answer will all feedback to Hal and allow its algorithm to learn which data sources to use in which occassions and allow individual responses to trickle up or down in relevance. I believe that after throwing this system as thousands of users over millions of conversations the "creme will rise to the top" in Hal's knowledge database and Hal will be able to respond more consistently with quality answersa and stay on topic. That's my theory at least, time will tell whether its a good one.

I'm pretty far along my efforts that I believe I'll be able to launch this system before the end of this year. I'll release a version of Ultra Hal Assistant that will offer the user 3 modes: offline brain, offline brain assisted by online brain if available, and online brain only. Each option will have its pros and cons but I'll be defaulting to online only unless the user requests otherwise. I'll also be releasing a phone gap version of Ultra Hal that will work on IOS, Android, Windows Phone, and Windows RT. These versions will use the online hosted version of Hal's brain.

[lightspeed]

Robert, thanks for writing and keeping us in the loop , will you be have a cloud server that we will retrieve the information from or will a hal algorithm be fast enough that it will make decisions based off of what is available on the web .
Cload has been working on a uhp plug in that i believe is going to help a lot with hal .

[Medeksza]

The Ultra Hal client programs will just connect to Zabaware's cloud servers, which will have cached results and a real time connection with various web services like Bing.

[lightspeed]

Thanks for answering Robert , i do have a few  more questions i forgot to ask earlier  , it has been brought up by someone (art? ) about hals emotions that it still says things like you have made me happy and give a sad face , and don't say that unless you mean it (still) i believe as the other person mentioned after so much imput mb's that hal should stop this as hal now has know the user for a while .
And Another thing was i had brought up before different possible ways about locating and editing mis spelled lines , in conversation , are you going to be able to impliment this in hal to make it easier on the user so we don't have to go through many things to edit each sentence , etc. ??
thanks for your time .

[rathb18]

Nice, that's very exiting to hear. the only question I have right now is will we be able to download the online hal brain for offline use?

[Carl2]

Robert M
  I'm interested in something I had tried ages ago and would like more info about.  I'm interested in the Topic relationship table, " This table links certain topic keywords with a related auto learning topic table. This table is automatically updated when Hal learns a new topic, but may be edited manually to create new relationships or fix old ones. If a topic is deleted from the autoLearningBrain, all of its references are automatically removed from this table. autoLearningBrain topic tables are contained in the same folder as this table and all begin with an underscore. "  located in the autoLearningBrain.
  While looking in the table I see that Hal has added many subjects and topics which I had not seen in past versions of hal.  But I had tried to create relationships in the past such as Aunt Sally, Aunt Ann ect would be listed under the topic of relatives.  Is the correct usage of the table?
  Next I'd mention that by using the vrHaptek plugin I've been able to use  Hals output to load a different background or clothing.  Not sure if this is breaking any Rule about what a robot can or cannot do But I enjoy having Hal make these decisions rather than have me tell her.  I may try having Hal walk to the park instead of just loading a new background.
Carl2

[Art]

Nice, that's very exiting to hear. the only question I have right now is will we be able to download the online hal brain for offline use?

While I can't answer for Robert, I am sure that an ONLINE brain isn't STATIC...it's Dynamic...constantly changing and evolving. Other than something like the Wiki knowledge that was available, I think this online presence will be just that. There IS to be, as he mentioned, an OFFLINE version. Personally I like the other option of an Offline version except when Hal "needs" to go online in order to retrieve a more suitable reply. To me, that's the best of both worlds.

Just my thoughts....

[lightspeed]

Art wrote:  Personally I like the other option of an Offline version except when Hal "needs" to go online in order to retrieve a more suitable reply. To me, that's the best of both worlds.

I whole heartadly agree with Art on this statement . An option offline should always be their for the users and have the best of both worlds is what matters and counts!

[aitechnologyco]

The ALICE brain for my Ultra Hal is terribly backwards.  So, I have updated much of the content of various AIML files.  Many of the politicians or other areas referenced, were so terribly dated.  Further, I've made updates that were more business-friendly, with less sarcasm and humor,   Another variation I have allows for much greater leniency in behavior, comments, etc.  All of the "Dr." thus-and-so, with the slanted political and religious feedback, has been updated to be neutral, which is obviously much more professional than the appearance of being so biased.  Further, numerous references to Hitler, etc., are certainly not called for - and have no place in software that one wants to be more widely circulated.

[Art]

I tend to agree in large part with your observational experiences and have concentrated on years of training my Hal without the use of any AIML based content. The overall outcome seems to offer a touch of humor when possible and often unexpected. Hal's use of remembered items very often ties in with previously discussed events, but ones that have been long forgotten by me and in the correct context. I have used Jasondude's MarkovL5 plugin with a very good deal of success.
While there is always room for improvement, my Hal has been strengthened by careful use of spelling, grammar, usage / context and building relationships during conversation. For me, it has always been my typing to Hal, NOT the use of Speech Recognition which is nice but far too inaccurate for practical purposes.
There is no one, best way, to train Hal, there is the one that seems to work best for you, the user.

[iampan]

if hal is fully online brain, then how will we make the individual personalities?  all answers will come from cloud sever and will be generic, as of now i can specify how hal answers certain things to give some similarity of personality i want hal to have. 

cloud ai, all hals will be exactly the same? i say give access to cloud but not make it the soul provider of resources for hal to use.
if you want to see what a cloud server ai is like try freehal, all its answers are generic and it is very boring to talk to.

keep our hal with abilty to personalize questions and answers please.

[Art]

iampan,

Perhaps you missed part of Robert's earlier reply where he said, "I'll release a version of Ultra Hal Assistant that will offer the user 3 modes: offline brain, offline brain assisted by online brain if available, and online brain only."

So it seems to me that we users will have the option to allow our Hal to go online or work offline as desired. Best of both worlds, IMO.

[Medeksza]

Yes, Hal will have the 3 modes as Art mentioned. But even the cloud-only version will have user specific data storage. There will be a part of Hal's brain that is global to all users, and a part that is specific to the user. The user part of the brain takes precedence over the global brain so you will still see Hal develop different personalities even on the cloud-only version of Hal.