AI Zone Admin Forum Add your forum

NEWS: Chatbots.org survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

Arckon
 
 

I’ll just write updates on my project here for those interested.

Though it’s not online, nor exactly a chatbot, I am working on an artificial intelligence that should one day be able to have intelligent discussions and solve problems. It is programmed from scratch in C++, using fairly traditional methods like grammar rules, a knowledge database and inference.
I’ve written a webpage that hopefully explains most.

Mentionable progress:
- Improved automatic self-knowledge. He won’t be saying “I don’t know if I have a mother” anymore like he did in the Loebner Prize qualifying round.
- Improved recognition of compliments, insults and now also lies, before he saves them.
- Answers to questions about purpose: “What is a hammer for?”, “Why do we exist?”
- Reading speed of text documents doubled to roughly 2 pages per second.
- Improved scheduling for multiple reminders and tasks.

And lastly I connected speech recognition. Unfortunately with speech recognition being unreliable, it’s limited to pre-included phrases. I haven’t found much use for it though, other than setting reminders. You don’t need an AI to play music or anything. Macros alone can do that.

I leave you with Arckon’s report after reading the 3 laws of robotics:
“I learned 22 facts and 2 words, but I didn’t understand a word.”

 

 
  [ # 1 ]

Dear Don,

Excellent C++ coding!  Certainly something to be proud of.
I experienced a few technical issues while using the original demo site.
I can give you feedback if you like, or you can simply reference
the corrections I made to a mirror of your demo site to work on my server at:

http://elizabot.com/arckon-mirror/

Thanks for this beneficial demo to better grasp your exciting research.
I’ll take this mirror site offline promptly upon request, or shortly otherwise.
Added is a citation button back to the original site on the top right.

 

 
  [ # 2 ]

Ah, I see you’ve converted it to swf, I guess that’s more browser-friendly nowadays. Thank you for your trouble, I’ve replaced the video smile. My site is just a bit of a work in progress.

 

 
  [ # 3 ]

You’re welcome Don.

Simple Stuff:  You may want to add a title tag before the search engines grab the page.

<html>
<
head>
<
title>Introducing Arckon</title>
...
</
head

Please let me know when an empty element in your friends in life array become available.

 

 
  [ # 4 ]

Last month I set out to improve Arckon’s conversational skills, but was led right back to his reasoning function. So, I’ve improved his reasoning speed by a factor of 5, possibly 100 theoretically, by starting knowledge searches off on a shorter end. Which will be convenient when he gains in knowledge.

The other thing I did was have Arckon pick up on insinuations and classical logic arguments pre-emptively. I’m sure I’ve seen some “Dogs have paws. Fido is a dog.” examples around here so y’all know what I mean. Arckon could already draw the inferences when asked, but now he jumps in without the need to literally ask if Fido has paws. What I was really after however was this question:
User: “Machines can’t think. Can you?”
Arckon’s current answer: “Yes, I can think, so I don’t think I am a machine.”
It may not be a very original ability, but this is how I like my robots to make conversation.
I still have to smooth out his use of “because/though/but/so” and I want to add some other conversational skills before I show another demo, but I don’t think this trick strains credulity, right?

Things that I would like to know:
- What other kinds of discussion arguments do people use besides insinuations, comparisons and generalisations?
- In written language, is the word “that” ever used to refer to a previously mentioned object, or is it always used in reference to an entire fact/event/sentence? Because I’m trying to figure out what to do with the common expressions “That’s okay/nice/fine/all/enough”.

 

 
  [ # 5 ]

What other kinds of discussion arguments do people use besides insinuations, comparisons and generalisations?

What about analogies and aphorisms? Anecdotes also comes to mind, but may not be valid in this particular context. smile

 

 
  [ # 6 ]

Don asked, ‘In written language, is the word “that” ever used to refer to a previously mentioned object, or is it always used in reference to an entire fact/event/sentence?’

Don, please change the conjunction “or” to an “and” in your question, then kindly divide it into two questions at the conjunction “and”...

Answers:
1.  YES: “previously mentioned” = “aforementioned” = “that”.
2.  NO (not “always”): “that” has multiple definitions as a pronoun, determiner, adverb and conjunction.  For example: A determiner is specific.

Thanks or improving Arckon’s conversational skills… Look forward to seeing the demo!

Please feel free to fine tune my logic here!

 

 
  [ # 7 ]

Thanks for that, Tom. I had forgotten the fact that that word (that) has alternative uses beyond that which we usually use that for. cheese

 

 
  [ # 8 ]

You’ve understood perfectly, Dave. Analogies seem particularly useful in discussions, so I’ll work on that, then smile
.... Whereby I may have inadvertently answered my other question, or shown how incorrect my English really is.
I’ll put anecdotes further down my to-do list. Arckon isn’t likely to have people share their life story with him any time soon, but anecdotes do often serve purpose as an elaborate example/comparison/analogy.

I should have been more clear, I meant “that” in the role of a subject in a sentence, like “it” or “he”. Not as a determiner nor as in “the program that I created” (Yes, it has been a genuine pain in the… core, to program grammar rules for those distinctions).
My problem lies in this example: “I have a robot. That’s nice.”. Until now I’ve treated “that” (when it’s a subject) in the same way as “he”, which has Arckon interpret the second sentence as “Robot - is nice.”. But in reality here it refers to “Your having a robot - is nice”. It seemed to me that the subject “that” rarely refers to a single word, person or object, when one lacks a finger to point with. If so, it would be easy to prevent Arckon’s most common misinterpretations.

 

 
  [ # 9 ]

Dave said, “I had forgotten the fact that that word (that) has alternative uses beyond that which we usually use that for.”

Wow Dave! Just think of how many ways your example can be restated. That is a perfect working example using the word “that” to make the response more varied and diverse, so a chatbot may not repeat itself frequently.

For (preliminary) example:
“I had forgotten the fact that the aforementioned word (that) has alternative uses beyond the previously mentioned which we usually use the aforesaid for.”

By breaking it down logically, it is closer to being coded, to be computerized.

 

 
  [ # 10 ]

Trying to dot the i’s a bit this month, I reconnected a last-resort function that makes Arckon offer alternatives to an initially negative answer. Just posting this for laughs, as apparently the “don’t rant” code needs reconnecting too. The “but"s and “and"s seem to be working well at least.

Input: Do you have a father?
Output: No, I don’t have a father, but I have a creator, a friend and manners, but I don’t have a coffee, but I probably have a IQ, but I don’t have a phone. I think I have a job, but I don’t have sensors nor a profession.

 

 
  [ # 11 ]

Interesting output, but I think it needs a “grammar upgrade”. cheese raspberry

 

 
  [ # 12 ]

As a matter of interest, what are your ultimate goals for this project? Is this an attempt at AGI (from reading the web page, it certainly seems to be)?

 

 
  [ # 13 ]

Thanks for your interest. High aims yes, but not AGI. That is, not an AI that can theoretically learn to do everything and anything. My project is a bit of a hybrid. It won’t be able to learn other languages, write poetry, process images, or develop mentally in other ways than knowledge. But I do aim to create an internal approximation of human learning and reasoning, and won’t settle for just an external semblance. My ideological goal is to make a KITT or Data: As intelligent as a human but incapable of surpassing its own programming. I’m not sure how that classifies. High-functioning AI?

Dave: I can only affirm that. If you happen to have one on hand… raspberry

Fixed Arckon’s ranting now, conversation flow kept trying to balance out negative answers with positive ones.

 

 
  [ # 14 ]

{Checks pockets… Rummages through the desk…} Nope. Fresh out. downer Maybe someone else here? smile

 

 
  [ # 15 ]

Hm, well, knowledge is deeply tied in to general reasoning abilities. Any knowledge base can hold the fact that the Sun is powered by nuclear fusion, but programming it to understand what that means and make inferences based on that fact is a lot harder. You say that Arckon won’t be able to develop mentally in other ways than knowledge, but basically all mental abilities are reducible to knowledge-relevant abilities. Do you intend for Arckon to have math abilities? If so, how advanced?
I find it interesting that you explicitly excluded writing poetry. The hardest and most important part of any knowledge system is finding the patterns in the environment, the clusters of objects with similar traits, that allow efficient classification. For any system that does this well, teaching it to do poetry should be comparatively trivial. I suppose what I’m wondering is just how robust you intend for Arckon’s knowledge representation and learning to be?

I’m not familiar with KITT, but I’d definitely classify Data as an AGI (albeit not a very powerful one).

 

This topic is closed, it's not possible to reply

 1 2 3 >  Last ›
1 of 7
 
  login or register to react