Artificial Superintelligence notes

  • 30 Replies
  • 2993 Views
*

infurl

  • Trusty Member
  • ********
  • Replicant
  • *
  • 606
  • Humans will disappoint you.
    • Home Page
Re: Artificial Superintelligence notes
« Reply #15 on: November 27, 2017, 12:08:28 am »
Chess engines disprove your above word salad. They develop tables that give relevance by doing analysis.

Yet they fail miserably at Go.

Shake your head again I dare you. You've obviously got a screw loose or you wouldn't be so paranoid.

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Artificial Superintelligence notes
« Reply #16 on: November 27, 2017, 12:33:26 am »
Chess engines disprove your above word salad. They develop tables that give relevance by doing analysis.

Yet they fail miserably at Go.

Shake your head again I dare you. You've obviously got a screw loose or you wouldn't be so paranoid.
Oh my, maybe you've been listening to too many deepmind employee dingbats. A mobile phone Go app called Crazy Stone had a 6-dan rating in 2014, but the program development appears to have died over 3 years ago. Go is not popular. Even the popularity of chess makes it difficult to make a living for software developers. At the rate Go development is going, it would take centuries before it got close to how much work has been put into computer chess.

Anyhow, Crazy Stone is not neural networking. It's one of the few Go programs that was taken somewhat seriously, designed for mobile phones. smh In 2014 it had a 6-dan rating, which is very high, and it was climbing. Rating just don't instantly climb. You have to play games to increase your rating. So we don't know how high it's rating is. There are hundreds of chess programs. Most are no where near the rating of the top 5. That clearly demonstrates the obvious, that it takes a special person to design and write the required code to surpass humans on a single desktop pc.

smh

*

korrelan

  • Trusty Member
  • *********
  • Terminator
  • *
  • 972
  • Look into my eyes! WOAH!
    • Google +
Re: Artificial Superintelligence notes
« Reply #17 on: November 27, 2017, 12:39:18 am »
It thunk... therefore it is!

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Artificial Superintelligence notes
« Reply #18 on: November 27, 2017, 12:41:27 am »
https://en.wikipedia.org/wiki/AlphaGo

smh

 :)
Yeah, running on over 1000 cpus. lmao. My ASI will destroy deepmind at go on one single cpu.

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Artificial Superintelligence notes
« Reply #19 on: November 27, 2017, 12:43:43 am »
I just looked it up again. To be specific, deepmind's AlphaGo used 1900 CPUs (god knows how many cores) and 280 GPUs during it's Go matches.

*

korrelan

  • Trusty Member
  • *********
  • Terminator
  • *
  • 972
  • Look into my eyes! WOAH!
    • Google +
Re: Artificial Superintelligence notes
« Reply #20 on: November 27, 2017, 12:51:04 am »
Can we see an example of your ASI?

Here is a link to my research so we can compare ideas.

http://aidreams.co.uk/forum/index.php?topic=10804.0

 :)
It thunk... therefore it is!

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Artificial Superintelligence notes
« Reply #21 on: November 27, 2017, 01:28:45 am »
I had 12 day old full source at github, but since my database was unheard of times faster than mysql and obviously faster than anything anyone can buy, I decided to hold off on uploading it since I'm no expert in law and did not want a company stealing it and copy-writing with some loopholes in the legal system and the power of an army of lawyers.

You know I was just about to show a snapshot of my code in code::blocks since it seems you don't think I'm coding, but why? I'm here to discuss ASI, not to prove I'm coding. ... And besides, can this forum even upload images??? Good grief please tell me it can.

If all goes as planned you'll learn about my ASI in about 2 years .... Huahahahaa . In another thread I've already outlined the basic structure.

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 855
Re: Artificial Superintelligence notes
« Reply #22 on: November 27, 2017, 01:45:21 am »
Here is my theory rough theory of AGI. sorry it is a tough read. Or wait for the better easier one in a few months.

AGI basic structure. Jeff Hawkins influenced:

https://groups.google.com/forum/#!topic/artificial-general-intelligence/UVUZ93Zep6Y 



*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Artificial Superintelligence notes
« Reply #23 on: November 27, 2017, 01:57:10 am »
Crazy Stone go program appears to be written in Java since it was written for google android apps. Anyone who knows anything about performance programming is shaking their head right now.

*

Thierry

  • Nomad
  • ***
  • 60
  • God is a sphere. Press 9 to enter.
Re: Artificial Superintelligence notes
« Reply #24 on: November 27, 2017, 02:09:54 pm »
And besides, can this forum even upload images??? Good grief please tell me it can.


You can proceed this way :
1. Choose a hosting image website (i am in IMGUR.COM). This is usually free.
2. Create an account
3. Download an image.
4. Make it private or not
5. Copy/paste the BBcode link related to this image.

*

Zero

  • Trusty Member
  • ********
  • Replicant
  • *
  • 577
  • Offline
    • Thinkbots are free
Re: Artificial Superintelligence notes
« Reply #25 on: November 27, 2017, 03:17:43 pm »
Here is my theory rough theory of AGI. sorry it is a tough read. Or wait for the better easier one in a few months.

AGI basic structure. Jeff Hawkins influenced:

https://groups.google.com/forum/#!topic/artificial-general-intelligence/UVUZ93Zep6Y

Nice! There are several fuzzy (undefined) high level descriptions, that would probably be hard to turn into real algorithms (for now), but the overall structure seems good.
 O0
Thinkbots are free, as in 'free will'.

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Artificial Superintelligence notes
« Reply #26 on: November 27, 2017, 04:55:49 pm »
Neural networking can be summarized as the worlds slowest interpreted programming language. It's used when an SE is unable or too lazy to do it the correct way. Anyway except NN is the correct way. If we look at chess history we'll see that significant improvements were made every so often. It didn't happen all at once.

In short, NN never results in the fastest code. Great, sometimes you can't accomplish something within a given time frame, so you use NN. It seems that decades of chess programming from massive numbers of SEs has resulted in non-NN code that's awesome. Although I'd say they still have a lot of improvement in the intelligence and self-learning category. Chess engines are approaching 3700 ELO rating. The best NN chess engine that I'm aware of is Giraffe.  As of November 18, 2017, Giraffe is only rated at 2461 at 40/40 minute games 2409 at 40/4 minute games. Over 2 years ago it's elo rating was 2366, perhaps due to faster computer or more optimized code. Giraffe's ELO rating is about 1300 ELO lower than the top non-NN engines. An ELO rating difference of 1300 points is like comparing a caterpillar to a human in thinking power lol.

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Artificial Superintelligence notes
« Reply #27 on: November 27, 2017, 05:06:09 pm »
p.s. Chess programs elo rating generally increases by 50 to 70 points from doubling the computers processing speed. If we use 70 points, a 1300 elo increase requires a computer that's 390,000 times faster. Using 50 points the computer would have to be 67,000,000 times faster. So maybe if google transforms all of texas into a massive CPU lab then Giraffe NN chess engines could have a high elo rating.

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Artificial Superintelligence notes
« Reply #28 on: November 27, 2017, 05:24:14 pm »
p.s.s. Sure I'll admit that the NN chess engine shouldn't be millions of times slower than a non-NN engine. Not even NN is that slow lol. It probably means Giraffe neural network chess engine could use a lot of improvement. It would be interesting when we find out the difference. If I had to make a guesstimate I'd say that non-NN methods have the potential of being roughly a hundred times faster, which would mean a NN chess engine would be rated 3200 to 3300.

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Global Moderator
  • ********************
  • Cleo
  • *
  • 4936
Re: Artificial Superintelligence notes
« Reply #29 on: November 28, 2017, 09:40:15 pm »
If I may ask a few obvious questions...

How do you display or show the operation of your ASI when it is running?

Do you simply type in a question and wait for an answer? Like a chatbot or Alexa?

Does it generate visual responses to programmed data?

Can it show the results from various input methods you assign it?

I don't think we're looking to see your source code per se, as that is rather personal, but rather how you quantify the output of your ASI?

Just some thoughts for you.

Respectfully,
- Art -
In the world of AI, it's the thought that counts!

 


Users Online

35 Guests, 0 Users

Most Online Today: 44. Most Online Ever: 208 (August 27, 2008, 09:36:30 am)

Articles