[ExI] alpha zero

[spike]

From: extropy-chat [mailto:extropy-chat-bounces at lists.extropy.org] On Behalf Of John Clark
ubject: Re: [ExI] alpha zero

 

On Thu, Dec 7, 2017 at 7:57 PM, William Flynn Wallace <foozler83 at gmail.com <mailto:foozler83 at gmail.com> > wrote:

 

​> ​>…I suspect the Singularity is pretty far off.

 

​>…Maybe, maybe not. Even if it won't happen for a thousand years in 999 years it will still seem pretty far out, and when it arrives it will be a big surprise. If it were otherwise it wouldn't be a singularity.​.. John K Clark​

 

 

This whole episode has been swoon out-cold surprise to me.

 

Consider: even if we assume there is something amiss, even if we assume there is some big puzzle piece we aren’t being told… what this Alpha Zero program has already done is so crazy impressive, it is astonishing.  I follow the computer chess championships every round and note that the games are generally dull as a typical afternoon at the Assisted Living center.  But these games aren’t dull at all.  They are full of high-concept, very positional strategic play.

 

But assume for the sake of argument there is some kind of elaborate hoax.  It can’t be they faked Stockfish’s play, because that is verifiable.  Millions of copies of Stockfish (that exact version) are out there, so anyone can put all the settings to the values they give in the paper and verify Stockfish would respond as the game said it did.  The games look very Stockfishish to me.  So if it is a hoax of some sort, they would need to be helping it somehow, presumably with human intervention.  But plenty of the tactical shots wouldn’t be found by a dozen grandmasters, and even if it had been done that way, plenty of people would know about the hoax.

 

Furthermore… as far as I can tell, everyone who signed on to this paper is a real person.

 

https://arxiv.org/pdf/1712.01815.pdf

 

But even if all that, we have good examples there of something beating the bits out of Stockfish, which means… whatever did that may compete for the computer chess world championship next round, and will win if these results are real.  This would occur to DeepMind: that they would be expected to compete and win next round, so even if they perpetrated a hoax somehow, they would do so knowing they were going to get caught, and then no one would believe their company afterwards.

 

So… I conclude with awed astonishment… that these results are probably the real thing.

 

If you had offered a PredictIt bet to me last week that some innovative approach to computer chess would just show up unexpected and slay Stockfish, I would have offered 98 cents for no votes, then lost it all.

 

This whole episode has caused me to look anew with a more open mind on the notion of a surprise singularity.  I always considered that scenario most unlikely, but I considered this DeepMind chess program nearly impossible.  Now I think it must be right.  Shows to go ya.

 

spike

 

 

 

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20171207/ba2bd6fe/attachment.html>