John,

I say this only in the best of spirits, because your post does NOT need to be moved, nor anything else need be changed about it.  It’s fine.  However, in my opinion,  you should make a new thread about your border collie.  I would be delighted to read all about it.  I think anyone who enjoys chatbots would really enjoy reading about the most intelligent dog breed. 

I think you know, John, as an owner, but for the sake of anyone else, border collies are the most intelligent breed of dogs.  Even smarter than poodles.  It is almost as if border collies understand basic human language and grammar. With training, you can do more with them than other dogs, like say, “Put your nose on the red ball” or “Put your paw on the blue bucket” and it will actually do it, and many different other similar commands.  It is like they understand your actual words because they can remember a considerable amount of different commands using different words, like “Put your paw on the red ball.”  Some border collies can remember many dozens of separate items.  Then when you show them a picture, they can run and fetch that item from another room out of sight, and bring it to you, just by looking at a picture of it in the first place.

Are there any drawbacks?  Sounds too good to be true… Why doesn’t everyone get a border collie if they are so intelligent?  Well, they really are so intelligent.  That part is totally true.  However, they are “outside” dogs.  They need to be given jobs to do outside, like hurding sheeps, or they can become really stressed out.  That is just how they are, they really need to run off that natural stress level outside, which makes them different than most domestic dogs of lower intelligence.  So maybe a poodle is the most intelligent inside dog?

You’re very correct, Pla. They need a job. Josie’s job is FRISBEE. That’s the f-word in our house. You can’t say it. lol. Her second job is being my sidekick.

If she doesn’t get enough excercise (and mental work is as good as physical work for her) she gets stressed and her behavior becomes an issue. I was out in the middle of January, -20 degree weather, throwing frisbees. My neighbors think I am nuts. They may be right, but the dog needed that job.

If i could make my robot half as intelligent as my dog…

Don <& others>

Would you have your bot conclude its searching once its “limits” of color, size, shape, number, texture, etc. had been obtained? I’ve seen many definitions that have sprawled through several paragraphs with descriptions of a subject, often detailing multiple uses for the subject in question. Contrarywise, I’ve seen some bots only provide a blurb or snippet, leaving more to be desired.

How and at what point does one draw the line? Too soon and your bot might run the risk of neglecting some potentially important information. Too lengthy and your bot could become “bogged down” with useless or trivial information.
Finding that “happy medium”, the gist of the matter, is the way to fine tuning the bot’s knowledgebase, IMHO.

Finding the way to accomplish this is, in itself, another matter.

Thoughts?

John,

I thought it might be amusing and an interesting exercise to enable Simplex to respond to your question! That conversation follows:

Are you curious

  I could not agree with you more! it is for others to say.
what I do know is ...
that
when I don’t have more important demands I attempt to discover new links between words.

can you think
  I understand that ... well I believe I can!
Having said that ... I should try to be modest ... let others decide!
I understand that there is a great deal of boring data entry work
and
underlying code which will enable me to
priorities what to think about
when I’m running that has yet to be done

Now I’ve heard it all anyway break a leg
  Thank you. I believe one creates one’s own luck!

So you can undersatand idioms!
  I believe that ... yes ... I can!
Having said that ...
there are so many
to learn

I’m off to the pig and whistle
  Remember the morning follows the night before! Don’t do anything I would!

Jim.

I was hoping you’d chime in on this topic, Jim.  Interesting response from Simplex.

I especially like “when I don’t have more important demands I attempt to discover new links between words”, because I thought that when my robot was least active (ie: bored) it would go try to answer some questions on it’s own…curiosity.

Art Gladstone - Apr 25, 2014:

Don <& others>

Would you have your bot conclude its searching once its “limits” of color, size, shape, number, texture, etc. had been obtained? I’ve seen many definitions that have sprawled through several paragraphs with descriptions of a subject, often detailing multiple uses for the subject in question. Contrarywise, I’ve seen some bots only provide a blurb or snippet, leaving more to be desired.

How and at what point does one draw the line? Too soon and your bot might run the risk of neglecting some potentially important information. Too lengthy and your bot could become “bogged down” with useless or trivial information.
Finding that “happy medium”, the gist of the matter, is the way to fine tuning the bot’s knowledgebase, IMHO.

Finding the way to accomplish this is, in itself, another matter.

Thoughts?

In conversation, my program stops asking after 4 basic questions or sooner. Just the answer to “What is a dog? A pet.” can provide enough indirect information. Presumably humans start by explaining the most important aspects first. There is also an order of importance to different kinds of facts, which is why I’d like to know by what traits humans and children primarily classify objects.
When reading documents, my program just learns all there is to learn including trivial information. When the program is asked to give a description of a subject it examines all information but limits its answers to what is most relevant to current conversation. So I only draw a line at the output. Granted, it’s not heuristic.

As the others are describing, curiosity should only come into play when no more immediate concerns demand attention. Otherwise it’s often inappropriate.

Thanks Don!

@Jim, Same question applies. How does your bot determine that is has enough information / data to supply an answer?
How does it determine that it has enough information to satifsy the inquiry?

Thanks

Art/Don,

At present this aspect is something which I have under review.

The simplistic answer to “How does your bot determine that is has enough information / data to supply an answer?” is indicated by the fact that it either has an answer or it dosn’t.

If the system contains a relevant field pertaining to the question asked, it is usually a simple matter to come back with an answer.

The problem arises when no such field has been provided for. In those cases, I try to limit what needs be searched to come up with the answer.

There are many techniques that I employ and have employed to achieve this.

This subject is being tackled empirically with continuous research in the hope of coming up with the best general approach. However, it should be borne in mind that there are many ways to skin a cat.

If you have a specific client in mind who has specific needs then the best approach is not likely to be the general approach.

In such cases, I suspect a tailored approach may well be needed.

In addition to the above, the importance of context should not be overlooked. eg Should someone ask for your name in a bank then the probability is they want your full name whilst in a bar your first name will suffice.

With the advent of cheap storage and the liklihood that this trend will continue into the futre, I personally do not feel that
you need over concern yourself with storing too much info. What should be of of concern is to decrease the the probability of becoming bogged down by introducing better methods to enable you to limit what has to be searched in response to any request.

Jim.

That basically evaded my question then you covered it by piling on more verbage…

“This subject is being tackled empirically with continuous research in the hope of coming up with the best general approach.”

What?

How is your bot’s info acquired, selected or tailored for suitable use, then presented?

The fact that it either comes back with an answer or it doesn’t, makes me wonder, as the vast majority of bots will always return an answer mind you, not always the best answer perhaps, but still, an answer none-the-less.

How should you handle your bot’s responses based on the data it has found? How much is too much and how does one shorten it and still keep the gist of the response? Again, how much do you feel is too much and how little is too little?

We could certainly get into filters when / if your bot wanders the net grabbing sometimes questionable information including slangs, foul or generally misinformed statements, but I assume this will be subject for another discussion.

I would like to learn different methods as I’m sure lots of others here would as well. I don’t think any one person has this whole bot thing completely nailed down just yet!!

Not all the people here are at the same evolutionary level of some of the more acclaimed individuals who haunt these halls but we all try to learn and experiment with new methods and ideas.

Share what you’re able and we’ll all benefit! Thanks for your time and I’m really not busting on you!

I see I’m not the only one who noticed that. I was going to reply with a comment about semantically null statements, but decided against it. Maybe next time.

Art,

Thank you for your response.

I am confident that in your heart of heart you understand perfectly why I was obliged to respond in the manner I did.

Possibly, we have all devoted years in developing our AI systems/“chatbot” and so I must quickly learn to be more sensitive especially as I do not know that my approach is right. Every fibre of my being is telling me it is correct, but I just don’t know! Contrary to my nature I must learn to walk on egg shells!

A lifetime of PRIVATE research has led me to my conclusions which include in my view that the traditional chatbot approach is not the way forward. I may well be completely wrong. I just don’t know.

Whilst it is reasonable for me to point the forum in what I perceive to be the right direction (which may be wrong) it would be foolish of me to place my tenuous conclusions in the public domain.

In response to your question “How is your bot’s info acquired, selected or tailored for suitable use, then presented?”
and subsequent question I can only reply, context, context, context and context. There is no magic bullet.

I tried to hint at this by saying in my original reply “the importance of context should not be overlooked. eg Should someone ask for your name in a bank then the probability is they want your full name whilst in a bar your first name will suffice.”

In my view, it is imperative that any AI system/“chatbot” must devote what may appear to be inordinate computing power to keeping track of context!

Also look for elephants!!!

Jim.

I have the impression that Jim just doesn’t want to divulge much on how his bot works, understandably while he believes it is fairly unique and hard-earned.

How is your bot’s info acquired, selected or tailored for suitable use, then presented?

And that, is a lot to ask
My method for curiosity was purely towards an AI that runs on facts, the more, the better. If one has a chatbot telephone app however, one has to somehow filter all non-essential information because you wouldn’t have the luxury of large storage space. And then the chatbot would simply not have a suitable answer to trivial questions, it would be limited. If instead of a knowledge database the chatbot stores only sentences, again you’d need a different approach, e.g. when pillaging Wikipedia, only grab the first two sentences that typically contain the summary. If you’re getting your information live online, you might, like Watson, include in the answer only the parts of the text that statistically match keywords of the question. If you want to shorten grabbed sentences, you could leave out adjectives and other non-essential words. etc, etc. I could go on all day.
The question shouldn’t be how one of us handles things, but how one might handle your specific case and purpose, which haven’t been stated.

It used to be that my program too would either answer the question or say “I don’t know”, and that would be that. Curiosity didn’t come into play until I tried to get more conversation going, in which curiosity is socially appreciated, but optional.

Mr Curran, your name is rather prominent in AI research, are you the James Curran associated with Stephen Clark?

http://sydney.edu.au/engineering/it/~james/pubs/
The reason I am asking is that your usage of the name SIMPLEX seems to be based on a deeper understanding of some of the proposed learning models than your posting seems to allude to.
http://link.springer.com/chapter/10.1007/978-3-642-29347-4_13#page-1
Have you given any additional thought to publishing some screen shots or better yet a demo video?

Vincent L Gilbert

@ Don,
Yes, I got that impression as well. It is his prerogative to not divulge his methods but some useful hints might serve to strengthen his position.

@Vincent,
Nice find! We’ll wait to see what unfolds. <Will the real James Curran please stand up> - ala TV’s What’s my Line?

Well Mr Curran, at the very least you have managed to generate a long running thread!

<Will the real James Curran please stand up>

And on that note, I have to point out that Art and Don’s posts highlight another curious aspect of Mr Curran’s postings. Though he should be a native English speaker, his use of colorful adjectives do not reflect that of a typical native English speaker.  that plus his “deflection as a response” responses makes it decidedly…

Bot like

VLG