Google's Deepmind AlphaZero destroys Stockfish 8 at chess

  • 29 Replies
  • 2678 Views
*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« on: February 28, 2018, 06:01:18 pm »
This impressed me since Deepmind is neural networking. After digging into the details it turns out that AlphaZero ran on 4 of googles TPUs during the chess match. Wikipedia states how fast a 2nd generation TPU run at, but it's unclear to me if one TPU runs at 45 TFLOPS or 180 TFLOPS. After doing the math it doesn't seem to make much difference because the chess ELO rating system is not linear. Google doesn’t give us much details on the PC that ran Stockfish except that they only gave it 1GB of RAM cache, which is causing a lot of complaints in the chess community because that’s ridiculously low. My i5 core's 0.01087 TFLOPS is so-so. High performance PCs are about 0.050 TFLOPS. There are expensive PCs with a lot higher TFLOPS, but consider Google only gave Stockfish 1GB let’s use 0.060 TFLOPS. So 180 TFLOPS * 4 / 0.060 TFLOPS = 12,000. That means Deepmind AlphaZero was equivalent to 12,000 high end PCs.

Computer chess programs typically gain about 60 ELO points when doubling the computing power. If we double it 13.6 times we get roughly 12,000, which means AlphaZero has an advantage of about 816 ELO points. However, if we assume that one TPU is only 45 TFLOPS, then it comes to 756 ELO points. Not much difference.

The ELO rating system states that 400 ELO points lower means you’ll win 10% of the time. AlphaZero won 290 and lost 24, close to 10%. So AlphaZero’s rating is about 400 ELO points higher, but it has roughly 800 ELO point gain in computing power.

End result, this shows the Google DeepMind AlphaZero neural networking program has roughly 400 ELO points lower than Stockfish 8. That’s a huge difference, which means Stockfish 8 would win about 90% of the time. BTW the latest Stockfish chess program is version 9. Also lets not forget that google only gave Stockfish 8 1GB of RAM cache. So who knows what the real difference would be.

This is all in agreement with my assessment with neural networking, that in the end it will just be slower. I think a good comparison is to think in terms of programming languages. There are flexible languages such as Java that can run on any cpu. And there are easier to program languages. In both cases the end result is that it’s slower. I’m convinced that using the chess algorithm method such as Tree Search will eventually be the best and fastest method. Neural networking is the easy method. Google can use their army of computers, Google brain, to solve the neural network, which is essentially equivalent to writing the source code. It writes the source code for you, but it’s not written in an optimized computer language such as C/C++. It’s written in an highly interpretive language. In Java there’s what they call bytecode. It’s not assembly, or C, or Basic. It’s bytes of code that represents an instruction. An instruction that is not related to any cpu. It’s generic, which is why the same bytecode can run on any cpu. A neural network is even farther away from being optimized computer code. The payoff is that anyone can do machine learning, but the downfall is that it’s going to be a lot slower.

I push for AI developers to work on the Tree Search method because in the end it will make a big difference. IMO that difference could hundreds of times faster, which could make all the difference. Imagine a robot that’s for sale that could be a maid at someone’s house, or being a therapist, a psychologist, etc. Then imagine that robot being so slow that it would take it 300 times longer to figure things out, which would probably mean it would not be so useful.

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 855
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #1 on: February 28, 2018, 07:55:56 pm »
 On a sub scale i will agree with a search tree search. I more agree with many program sub routines working it in parallel.
 But high up. And all working together. The collective result is not a tree search. It is a sequence selection instead.

 Let me explain.
In the next step in a computer program will come to a four way tree. Then it will select the a direction.
 In a AGI system, in most cases, the AGI selects the most rewarding first. Straight out select that as goal. But if for some reason
it cannot achieve it, it select the next best fork. AGI uses a sequence list with four entries. Not a parallel decision tree.

 A sequential list like a sequential link list. Which rNN do quit well.

Karl Friston - 2016 CCN Workshop: Predictive Coding: 

https://www.youtube.com/watch?v=b1hEc6vay_k&feature=youtu.be&t=255   

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #2 on: February 28, 2018, 08:14:18 pm »
Well Google has another advantage that it's TPU is custom designed specifically for its AI. Can you imagine if the owner of Stockfish could do the same? Stockfish uses Tree Search, which loves parallel processing. Give Stockfish a million cores and it will use all of them very efficiently. :)

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 855
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #3 on: March 01, 2018, 12:50:40 am »
 Tree search is a brute force method.

*

ranch vermin

  • Not much time left.
  • Terminator
  • *********
  • 958
  • Its nearly time!
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #4 on: March 01, 2018, 03:03:52 am »
If you code badly enough, It might not even matter how many cores you have, optimization is very important - and you could see a less poewrful computer win, I imagine custom logic gates would also be a slightly different mix of things than generic universal logic that computers use.  Comparing things is slightly more detailed than just the hz and amount of cores.

Also sometimes the programs could be symantically different,  and the winning symantics will win dispite the worthless power wasted on the other side.
« Last Edit: March 01, 2018, 03:41:19 am by ranch vermin »

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #5 on: March 01, 2018, 04:40:11 am »
Chess programmers are always optimizing stuff. There are countless types of tree search. Some spend more cpu time on making smarter pruning and therefore don't search as deep while other engines spend practically no time and just go deep.

Does anyone know much about Deepmind NN? I read it has problems / limitations so far. I get the impression it can't be like a biological brain, like a human, or even an animal brain. Using the flexible tree search method I can easily see the AI becoming human like.

*

ranch vermin

  • Not much time left.
  • Terminator
  • *********
  • 958
  • Its nearly time!
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #6 on: March 01, 2018, 06:23:03 am »
I doubt its that powerful, im more of the opinion that this ai job isnt as complicated as people make out.   So it does small things well,  but I wouldnt trust it by itself to do anything dangerous.

Reinforcement learning if you get it up and running can do some pretty amazing yet simple things.   Im trying hard to get mine up and running, but i have motivation problems and my head is quite ill, but im confident ill get similar results myself when its finally working.

<deleted needless selfish ranting,  sorry :)>
« Last Edit: March 01, 2018, 10:17:54 am by ranch vermin »

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #7 on: March 01, 2018, 03:13:00 pm »
How would deepmind train it's AI to evolve into a brain that's at least comparable to an insect? Does it seem possible with their present TPU super computers? Now that would be great. Put it in an environment. Give it some limbs and eyes so it can learn to walk around. Put other creatures in the sim. I can hardly wait. They need to stop simulating games. Geez lol. It might take ages to do that on a normal PC. Maybe if you made the sim as simple as possible?

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #8 on: March 01, 2018, 03:24:20 pm »
Oh yeah I agree that it could be dangerous to let powerful AI unleashed. It could be very dangerous for humans, probably because humans would try to destroy the Synths and therefore they would have no choice but to attack back. Although in all fairness the outcome would be beneficial for intelligent life. Come on. We can't expect to be on the top forever lol. Who knows. Maybe super intelligence will actually be compassionate. Maybe the Synth Gods will nudge us along on our evolutionary path and be nice to us. After all it's a pretty big Universe. :)

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 855
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #9 on: March 01, 2018, 04:00:26 pm »

How Does DeepMind's AlphaGo Zero Work?: 


   



*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #10 on: March 01, 2018, 04:28:08 pm »
It must be nice to have access to mainframe supercomputers. :)  I didn't know that AlphaZero used a monte carlo tree search. Very cool! They're getting smart at deepmind. ;) Enough with the games already, deepmind. Time to create life!

*

ranch vermin

  • Not much time left.
  • Terminator
  • *********
  • 958
  • Its nearly time!
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #11 on: March 01, 2018, 05:07:40 pm »
It seems silly just wanting to make a little bug walk around the place by itself... But if you do it right its diabolically useful.

Imagine if you could suggest things to ants to do your bidding what you could get up to,  theres definitely going to be theoretical laws made solid when it gets up and running.

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 855
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #12 on: March 01, 2018, 05:10:58 pm »
 By all mean. Deep mind can create what ever. I am sure they can create something that will make them money. I am pretty
sure deep mind is using other peoples work.
 But i still consider there AGI models fare behind mine. And my model is more saintly.   
 With deep mind ethics i fear worst.
 To create in the image of ones self? If i was a thief and gold digger and created thousands. Yes i would have fears for
humanities future.




*

ranch vermin

  • Not much time left.
  • Terminator
  • *********
  • 958
  • Its nearly time!
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #13 on: March 01, 2018, 05:40:38 pm »
Yeh Keghn,  all the scabs go in a big squabble over a poor guys working theory. like seagulls for a discarded chippie its not stylish at all,   then they use it for useless ends.   thats the reality here, SHIT HAPPENS.  is what is going to happen to some degree, it always does.

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Google's Deepmind AlphaZero destroys Stockfish 8 at chess
« Reply #14 on: March 01, 2018, 11:26:41 pm »
I think DeepMind should create the most biologically advanced creature possible. If that's a ladybug, then go for it. That would stun the world. If it's a human, then go for it!

Keghn, have you thought about contacting companies about your AGI? Just make sure you own at least 51% of the company. A company like facebook, Microsoft, Apple, etc. could easily make you the owner of a competing company like DeepMind  giving you access to supercomputers that are tens of thousands of time faster than a high end desk PC. Some supercomputers are hundreds of thousands of times faster. :)

 


FAQ on the newly established MIT Stephen A. Schwarzman College of Computing
by Tyler (Robotics News)
October 17, 2018, 12:00:10 pm
MIT reshapes itself to shape the future
by Tyler (Robotics News)
October 16, 2018, 12:01:15 pm
Tag... you're it...
by Art (AI News )
October 16, 2018, 02:55:01 am
In need of a psychological coach?
by ruebot (AI News )
October 13, 2018, 01:33:37 pm
Leela Chess Zero
by Art (AI News )
October 08, 2018, 03:30:42 pm
Alpha GO Zero
by Art (AI News )
October 08, 2018, 02:48:32 pm
Detecting fake news at its source
by Tyler (Robotics News)
October 06, 2018, 12:00:55 pm
Phase Change Memory
by Art (AI News )
October 05, 2018, 01:21:59 pm

Users Online

75 Guests, 1 User
Users active in past 15 minutes:
ivan.moony
[Trusty Member]

Most Online Today: 163. Most Online Ever: 208 (August 27, 2008, 09:36:30 am)

Articles