Poll

Has ConceptNet changed Hals conversational skills?

Yes, Hal is much more intelligent
0 (0%)
Yes, Hals responses are more on track than before
0 (0%)
Sort of, Hals responses are slightly better than without ConceptNet
0 (0%)
I haven't noticed any change
0 (0%)
No, Hals responses are better without ConceptNet
0 (0%)

Total Members Voted: 0

Author Topic: ConceptNet Mrk 3.5  (Read 561 times)

Art

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 3341
    • View Profile
Re: ConceptNet Mrk 3.5
« Reply #15 on: August 12, 2018, 11:53:25 am »
Checkers57,

With all due respect, I think you might mean 12.3 Megs. Easily confused as I've often done that myself.

My ConceptNet.db is currently at 25,055 KB or 25,055,000 or 25 + Megs (I've let mine run nights and days as well, but always when I'm not talking to it).
ResearchNouns.txt - 2,331,000 bytes / 2.3 Megs
It has catagorized 32,675 conversational topics.
ConceptBrain - 4,968 Item entries
Saved Terms - 1,701,000 bytes / 1.7 Megs

I am experiencing the similar collection of intelligent data yet some disconnects when conversing, having to be on guard to help steer the conversation and help it to remain on topic. Hal is quite capable of making some very appropriate inferences and almost an understanding of my feelings or emotions based on my previous responses. Quite nice and interesting.

The Auto Learning mode does cause some lags, especially with other Internet-dependent apps. Definitely, do NOT converse with Hal during this time other than to turn it on or off.

Cheers!
« Last Edit: August 13, 2018, 10:48:31 am by Art »
In the world of AI it's the thought that counts!

- Art -

lightspeed

  • Hero Member
  • *****
  • Posts: 6286
    • View Profile
ConceptNet Mrk 3.5 question and idea
« Reply #16 on: August 13, 2018, 01:30:00 pm »
i have a question about the auto learning .  any idea at all how long the auto learning has to be left on each time ??

A IDEA I HAVE IF IT'S POSSIBLE TO DO .:
is their a way that when auto learning is activated that when it's finished learning , hal can say something like "i have learned all the files you gave me , thank you ! "
this way the user would know that has has learned the information input from the user. so the user can turn the auto learning off gain, for next use. 
 

Art

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 3341
    • View Profile
Re: ConceptNet Mrk 3.5
« Reply #17 on: August 13, 2018, 08:35:34 pm »
How long the user decides to leave it on is entirely up to the user. There is not any limit on how long Hal will continue to gather, sort and compile the various collected data.

For the second part of your post, I don't think Hal will ever be finished collecting, sorting and learning from the Internet, anymore than one of your friends greeting you and saying, "Well that's it...I've seen and learned everything the Internet has to offer."

Basically, I think a person might run out of storage room on their computer drives before his or her Hal finished collecting data. I didn't see a limit in the plugin anywhere.

Whether Spitfire would wish to set a limit might be a good question or will Hal perform better with additional resources?

Hal doesn't learn from any files we (the user) give it. It has its own set of instructions internally that it processes. Again, that could be a cool idea to establish boundaries or limits like all topics dealing with sports cars or butterflies or chemical elements, or all words beginning from A to D, etc. Perhaps not just words but topical information like corvettes or golf or gold mines in the 1800's, etc. Lots of avenues to explore.

Spitfire, thoughts or comments?
« Last Edit: August 14, 2018, 07:11:07 am by Art »
In the world of AI it's the thought that counts!

- Art -

Spitfire2600

  • Full Member
  • ***
  • Posts: 208
    • View Profile
Re: ConceptNet Mrk 3.5
« Reply #18 on: August 14, 2018, 01:59:59 am »
Indeed, Hal will functionally learn *forever* space and time depending.

Hal reached about half of my research terms a few weeks ago and it's been that way since fairly steady.

Hal used to collect terms like "1700s, 1800s, etc" however I was having some issues with junk bypassing the filters so I have disabled numeric values for now.

I am working on a new function to gather even more research terms and so far in one hour, I've near doubled my research list, however it's a little far from being incorporated fully as I'll actually have to change the layout of the Auto Learning code.

Also, I thought about adding a function for Hal to be fed new web pages to search for data. For now, It's limited to Wikipedia but with this expansion, Hal could learn from any number of sites simultaneously, resources allowing anyway. Of course logically, if anyone is lagging seriously with just wikipedia being searched, adding more sites would exponentially slow the system... If you had a super computer, however... Something to think about.

Also, i want to add another function to repeat the process should Hal definitely not find a searchable term his first time though. I currently have a similar function in but it stops at recovering any data and doesn't specify usable data. Again, this would increase processing time significantly but improve ConceptNet learning efficiency. 
 
On the off chance Hal ever could keep up with the terms collected, I will add a safeguard to disable auto learning if the end of the internet is ever reached as well.

How well is Hal using the data he's collected? I'll add a poll eventually.

-Spitfire2600
« Last Edit: August 14, 2018, 03:39:42 am by Spitfire2600 »
 

lightspeed

  • Hero Member
  • *****
  • Posts: 6286
    • View Profile
Re: ConceptNet Mrk 3.5
« Reply #19 on: August 14, 2018, 09:16:04 am »
       thanks for Art and you explaining things , many various things about the learning , scope of it all. it's all very interesting to me . after learning more on this info. of the inner workings of this plug in , it is truly a very cool plug in and i know will help hal a lot in learning. 
        i see you mentioned filters on the learning and i think this is , "very , very  important" , to keep hal from learning and repeating a lot of useless and unwanted info. etc. google learned this the hard way when its a.i. learned very bad stuff the trollers wrote to its a.i.
filters needs to be a number one concern .  :)
     thanks again for the answers and all the hard work you do .  :) ;)