David Silver (Deepmind) inaccuracies

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Eelco de Groot
Posts: 4265
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: David Silver (Deepmind) inaccuracies

Post by Eelco de Groot » Sun Dec 24, 2017 4:03 pm

Rebel wrote:
Eelco de Groot wrote:
Rebel wrote: #2. What does David Silver mean when he says: "this previous world champion"?
I think that he means that Stockfish 8 was the previous TCEC winner, which would be correct, although you can have doubts that TCEC is not a FIDE recognized championship. Houdini now being the TCEC champion. I don't think Mark Lefler would have liked much being Alpha Zero's guinea pig instead of SF, although he would have liked learning from the games.
To me it sounds as brutal commerce, we AZ are now WC :wink:
They're science guys Ed. I think we should consider ourselves honoured that we could feature as true Drosophilas of AI. They've set their sights on other things than sports titles.

I don't blame them, these are exciting times. And if they are frightening to some, scientists rarely let that stop them.

It is like the acient story -or legend/myth?- of the Golem. They say that the name comes from the word for clay, but I also see the root Gol again, as in Golgotha, 'Gol' meaning skull. And is Golem really not related to Gollem, which the above link denies? Tolkien knew a lot about the roots of Christianity. Golem reads a bit like "dead man" to me. But I'm not an expert on Hebrew :)

Clay. Live may have arisen from clay.

Golem, Frankenstein's monster. Or the Greek myth of Pygmalion. Why is this all so old in man's memory or fantasy. At a time when nobody had ever seen a machine? At least not that we know of. "Wheels within wheels" it says somewhere in the bible. Count the number of metallic robots in Greek mythology.
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

mjlef
Posts: 1454
Joined: Thu Mar 30, 2006 12:08 pm
Contact:

Re: David Silver (Deepmind) inaccuracies

Post by mjlef » Mon Dec 25, 2017 5:23 am

Rebel wrote:David Silver (Deepmind) inaccuracies (1½ minute snippet).

https://www.youtube.com/watch?v=3kEMwF0 ... e=youtu.be

#1. TCEC technically is not the computer chess world championship. Stockfish isn't the ICGA world champion, Komodo is. Mark will not be amused.

#2. What does David Silver mean when he says: "this previous world champion"?
People can run tournaments however they want, but you are right that the World Computer Chess Championship is run by ICGA and not TCEC. But both are great tournament. The two tournament differ in a bunch of ways.

TCEC plays all programs on the same machine, eliminating hardware differences. Of course the hardware and settings might not be ideal for specific program.s Program are not allowed to have opening books. And programs cannot be adjusted between games (such as setting different Contempt values for different opponents). It is more or less a uniform platform test. One benefit is they play a lot more games than could fit into the roughly 5 days of the WCCC.

WCCC allows any hardware, so teams try to get access to the best machines they can. Plus Opening books, Plus allowing program changes of settings or even recompiles for each round. Opening books are fine tuned for months and changed on the fly between games depending on the opponent. Settings can be matched to a specific opponent.

Human chess is closer to WCCC than to TCEC. But WCCC has a lot fewer rounds, which means luck plays more of a factor. Neither tournament is likely to get past the error margins in the games played, meaning the winner might not be the actual strongest program. In an ideal world, we could play games on TCEC-like hardware for a year and determine who is likely best at these long time controls. But the program change very quickly, so such results are outdated before the games can be completed.

bhamadicharef
Posts: 31
Joined: Fri Nov 25, 2016 9:14 am
Location: Singapore
Contact:

Re: David Silver (Deepmind) inaccuracies

Post by bhamadicharef » Wed Dec 27, 2017 3:36 pm

The full video is at https://youtu.be/A3ekFcZ3KNw

I like this new approach, as much as I like SF knowledge embedded into
it source code. I think deep learning has a lot to show us, its true
knowledge of chess encapsulated into its weights will also need some
novel visualization methods to show us more than 3x3 patterns !
Brahim HAMADICHAREF
Singapore

pilgrimdan
Posts: 405
Joined: Sat Jul 02, 2011 8:49 pm

Re: David Silver (Deepmind) inaccuracies

Post by pilgrimdan » Wed Dec 27, 2017 4:40 pm

bhamadicharef wrote:The full video is at https://youtu.be/A3ekFcZ3KNw

I like this new approach, as much as I like SF knowledge embedded into
it source code. I think deep learning has a lot to show us, its true
knowledge of chess encapsulated into its weights will also need some
novel visualization methods to show us more than 3x3 patterns !
the chess stuff starts at approx. 23 - 24 minutes ...

bhamadicharef
Posts: 31
Joined: Fri Nov 25, 2016 9:14 am
Location: Singapore
Contact:

Re: David Silver (Deepmind) inaccuracies

Post by bhamadicharef » Fri Dec 29, 2017 3:26 pm

Cardoso wrote:Maybe in the future someone will invent a NN system with the capability of explaining in details it's decisions, we will have to wait.
In the past there were techniques to extract rules from neural networks, this was typically used in biomedical, to assess the knoeledge from nnets by doctors, but the nnets were much simpler than the current ones used by alphazero for example. graphcore [ https://www.graphcore.ai ] presented some visualisation of resnets and we can only hope they will detail how they do in the future. I look forward to see how new AI systems like Wave Computing [ https://wavecomp.ai ] will perform on resnets to be developed for chess playing !
Brahim HAMADICHAREF
Singapore

pilgrimdan
Posts: 405
Joined: Sat Jul 02, 2011 8:49 pm

Re: David Silver (Deepmind) inaccuracies

Post by pilgrimdan » Fri Dec 29, 2017 6:03 pm

Cardoso wrote:Maybe in the future someone will invent a NN system with the capability of explaining in details it's decisions, we will have to wait.
I would imagine ... when someone does ... the 'explanaton' would look like chaos and/or randomness ...

if that is so ... what does that say about intelligence ... human or other-wise ...

rules give rise to patterns ... fuzzy or other-wise ...

much discrete-ness is a hard nut to crack...

but when we embrace approximation...

that is when a break through occurs...

I don't know how much memory is involve in 5,000 tpu's

a lot I suppose...

it is a kick in the head...

Post Reply