Improving Ultra Hal

[Sentenza]

   The programmers work hard on it, but the brain... the answers are limited.   To improve the brain i have an idea.  Your programmers can create a chat program like MIRC or other and record all conversations then they can put it in a global brain.  Then after, when a user ask a question to Ultra Hal the program try to match a similar conversation in his database and give the appropriate answer. I imagine that after many years analysing conversations your program will really act like a human.

[Art]

Welcome!

Yes after chatting with countless thousands of users Hal will now be able to swear in 40 languages, offer insults never before heard, and generally speak and use grammer that would cause Kipling to roll over in his grave.

The results from such an experiment I fear, would be much less than desirable.

[Sentenza]

(forgive me fo my poor english, i'm french i may look stupid...)

 yes i understand that the grammar used in chat is terrible.  But maybe a chat program with a part of the interface with a dictionnary and a composer... so people are bind to use the words and the expressions you give to them... and people can propose update and can add their own expressions approved by moderator.  The goal is to make human operators to interact and the AI to analyse the conversation.

And also... why you don't include a button or something to make a correction when the AI give bad answer. For example if i say something and Ultra Hal give an erroneus message.  So i can answer to my own question and Ultra Hal learn about it. Instead of programming The Brain... wich appear to be a complex thing for people who are lazy...

The more people will use it, the more Ultra Hal will learn. For now i download the program... i'm alone... and Ultra Hal learn nothing! And i don't have the tools to make correction to teach him how to think... how to relate. A better way is to make the program mimic the process of thinking (step by step).

Finally why there is no community to developp the brain.  Where the people who have more knowledge share freely the changes they make on the brain.  So we go from version 1.0 to version 7.5 or higher... from first genereation to second generation...

There is already a SETI project... why not a Distributed Artificial Intelligence? "Bots" who scan the internet for partial conversations on forums, on websites... who analyse everything and statistically give an answer to a human operator.

[Art]

Your English is quite good no need to apologize.

You pose a lot of thoughtful and interesting points.

The AI brain collection similar to SETI sounds great
but it would be a massive undertaking to say the least.

What I've seen are many different "groups" of people
all over the net, dealing with their individual or
specific areas of interest regarding AI. There isn't
one common "standard" since there is so much diversity
with regard to approach, programming languages,
applications, "standards", purpose and funding to name
but a few.

The best we can hope for at the present, are small groups
of great people like we have here at this open forum
along with some others like www.digitalgirl.co.uk and
several university sites.

Correction of Hal: I think Jerry (onthecuttingedge2005)
recently wrote a correction script that will allow you
to "correct" Hal is you accidentally type an incorrect
entry. Use the Members or Search feature at the top
right of the screen and search for his correction script.

I found the thread for you:
http://www.zabaware.com/forum/topic.asp?TOPIC_ID=2869

Hal started out many years as a very simple help bot and
has evolved to version 6. Each one is better, more
powerful than the previous one. As long as there are AI
enthusiasts and bot fans, Hal will continue to advance.

[freddy888]

On that topic of expanding the brain through a chat program, this can already be done using the AIM chat software... The HalAIM bot can have it's learning level set just as you would in the normal Hal.  Like Art says though you are wide open for any kind of use and abuse.

If any one wanted to do this though, I think probably the best solution would be to set up a Hal AIM bot on a server somewhere and either invite serious developers or a have some kind of membership scheme so you could control access.  Whatever, it is certainly already possible to do something like that.

Jabberwacky is a good example of how the open access to the net can create a lot of varied responses.

[badhair1963]

Why not just create an "invitation only" chat room on ICQ or IRC for the purpose of teaching Hal?  There could be rules regarding the use of profanity and such-like to prevent the program from picking up things which would be inappropriate for younger users to be reading or hearing.

It can't be that difficult to do.  Maybe Mr. M. could set up one which would then be specifically endorsed by Zabaware rather than having loads of different ones popping up all over the place.

[Sentenza]

  After creating a chatroom for serious user who want to teach HAL.
There is another thing (very difficult for programmers)... to create a bot who can search answers for questions a human operator ask.  The most difficult part is to have a direct hit (100/100) the exact response to the question.  For that work i think they need to understand the way, step by step, a human try to find an answer on the internet to a question. Why not to create a topic where you ask people: "Give us a clear description ,using more step as you can, on how you procede to search the internet and find an answer to a question".

With a better understanding of how a human procede to solve a problem, you will be able to create an intelligent bot repeating this steps.

With a search engine integrated to HAL, the program will be able to give clear answer to a human and the number of answer will be greater.

Now imagine that many user (many HAL) connect to a network and interact together, giving their search results to each other...
On this network there is part of brain, updates, search results, etc...