It is currently Thu Mar 28, 2024 8:28 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 20 posts ] 
Author Message
Offline
 Post subject: Humans 1 Computers 0
Post #1 Posted: Wed Mar 15, 2017 4:34 pm 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
The following link gives a chess position that humans (including me) can solve instantly but which has so far defeated the world's best computer (at least as regards what it thinks the result should be).

https://www.chess.com/news/will-this-po ... sness-4298

The man who designed the position thinks humans may use "quantum computing" rather than normal computing, and is promoting an institute to examine this idea.

It would never be me, but I suspect someone could design a go position that would similarly defeat AlphaGo but which we humans would "understand" (which I think may be more accurate then "solve").


This post by John Fairbairn was liked by: sorin
Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #2 Posted: Wed Mar 15, 2017 5:24 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
What enables humans to solve that chess position quickly is logic, not quantum computing. That being the case, I suspect that there are a number of programs that could solve it, with appropriate modifications or database. In any event, it does not support the claim that humans have special powers that computers can never attain.

IIUC, Monte Carlo Tree Search go engines can solve any go position, given enough time. (But so can humans, although we usually need more time. ;)) I am not sure about AlphaGo, because it may not do exhaustive search. People who understand it better may correct me here. :)

I, for one, am quite willing to say that AlphaGo and other top go programs make mistakes, and that there are certain aspects of go that humans do better at, mainly because of logic. Suppose, for instance, that a certain group is alive in miai. To a computer program that does not have any idea of miai, but bases its evaluation upon quasi-random probabilities, that group may have a probability of life of only 75%, instead of nearly 100%. I say nearly 100% because of the possibility that it could die in a ko fight. As the program builds its game tree, it will exclude a number of branches in which the group dies, thus increasing its "probability" of life, and furthermore, it may very well play the miai correctly when the time comes. That, as we have seen, is good enough for it to play well. But that does not mean that we could not defeat it with problems that humans can solve by logic.

Currently the main area of go where humans can apply logic is the endgame. I wonder if AlphaGo, given modern tournament time limits, could play Berlekamp and Wolfe's 9 dan stumping problem correctly, or Berlekamp and Kim's $1,000 ko problem.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.


Last edited by Bill Spight on Wed Mar 15, 2017 10:51 pm, edited 1 time in total.

This post by Bill Spight was liked by: dfan
Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #3 Posted: Wed Mar 15, 2017 6:34 pm 
Lives with ko

Posts: 248
Liked others: 23
Was liked: 148
Rank: DGS 2 kyu
Universal go server handle: Polama
I think the key advantage humans have here is that a position like this is astronomically unlikely to appear in a real game. Although an algorithm could develop the insights to solve it, there's really no reason to in all their training. Some sub-ideas apply more generally, but the idea that you'd chain them all together and get the outcome you do just isn't a particularly promising direction to apply to chess in general.

Humans, on the other hand, don't exclusively play chess for their entire existence. We play a number of different games, manipulate shapes in three-dimensions, reason about social situations and are occasionally given riddles and brain teasers. My first reaction was to read out typical chess moves in that position. When that clearly wasn't working I started looking at different angles of the puzzle, making use of strategies I didn't develop playing chess. This sort of solution feels applicable to other real world problems.

The artificial in artificial intelligence is best understood to refer to the environment the "intelligence" develops in, in my opinion. If you took a baby and put it into a chess simulation, feeding them based on performance, they'd be a very weird, very different thing. And even that is closer to any pure chess AI could be to us, because the baby is still preprogrammed with billions of years of biological evolutionary pressure. In a sense, I think this is an interesting application of a common real world idea into the realm of chess where it's utterly foreign. As such, we have no trouble with it, but the chess programs don't have any of the relevant concepts built up to approach it. Interestingly, this also seems to be our weakness playing strong engines. We use logic and experience and such to rule out paths it turns out we shouldn't have, while the computer just reads and reads and reads until it finds the critical sequence we missed.

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #4 Posted: Thu Mar 16, 2017 2:01 am 
Lives in gote

Posts: 420
Liked others: 75
Was liked: 58
Rank: EGF 4k
In the text it says "According to the challenge, it is possible for White to draw or even win".

While a draw seems quite obvious to me I have no idea how white should win that position (well..., if black allows white to play c7 and c8 without taking the pawn and/or moving out the king then it would work...). Any idea, anyone? ;-)

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #5 Posted: Thu Mar 16, 2017 2:28 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
In the text it says "According to the challenge, it is possible for White to draw or even win".


If you look at the original diagram provided by Penrose, it simply says the challenge is for White to draw.

The body text, not by Penrose, adds "or even win" in quotes without attributing the quote. Possibilities include the Dutch author not properly understanding the meaning of "defeat" in the title or, if it is a real quote, that Penrose was just emphasising that it is a legal position.

I don't think it's logic that helps us solve this problem. Logic works from a known base and follows rational steps in sequence. The known base here is chess moves and rational steps testing each move takes quite some time. But humans, while needing some chess knowledge in the background, see the answer here much quicker than that - making a quantum leap to shift the base. Penrose seems to call that "consciousness." In practice, I think humans achieve the leap by saying to themselves "Oh, this is obviously a trick question", a response triggered by the unusual shape and the three bishops rather than the usual "look for a queen sacrifice" response we have to most chess positions. A computer can't yet have such an out-of-body experience.

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #6 Posted: Thu Mar 16, 2017 2:49 am 
Gosei
User avatar

Posts: 2011
Location: Groningen, NL
Liked others: 202
Was liked: 1087
Rank: Dutch 4D
GD Posts: 645
Universal go server handle: herminator
John Fairbairn wrote:
The man who designed the position thinks humans may use "quantum computing" rather than normal computing, and is promoting an institute to examine this idea.


Interesting position, but to attribute human success here to "quantum effect happening in biology" is utter bollocks.

Really, this is just very simple logic: Black's bishops can't attack any white fields, and the rest of his pieces can't move unless white lets them out by moving the pawns, so as long as white keeps the pawns in place and keeps walking around on white fields with his king, he can't be mated. Ergo, he can force a draw.


Last edited by HermanHiddema on Thu Mar 16, 2017 4:03 am, edited 1 time in total.

This post by HermanHiddema was liked by 2 people: Bill Spight, gamesorry
Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #7 Posted: Thu Mar 16, 2017 3:13 am 
Gosei

Posts: 1494
Liked others: 111
Was liked: 315
I don't really like this article and the host of duplicates out there. It felt more like I had succumbed to click-bait when reading it, rather than finding some interesting discovery by the great Roger Penrose. What is there to say about this other than they've exposed a bug in the classical chess engine. If you push this into a MC/UCT chess engine I wonder if it spits out the same poor evaluation? At least on reading the Telegraph I found that I could email in my answer.

P.S. The Telegraph also states that you can win if Black makes a mistake.
P.P.S. https://www.youtube.com/watch?v=lJmzfdF4CZk

_________________
North Lecale

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #8 Posted: Thu Mar 16, 2017 4:56 am 
Lives in gote

Posts: 418
Liked others: 9
Was liked: 83
Rank: kgs 5 kyu
KGS: Pio2001
John Fairbairn wrote:
It would never be me, but I suspect someone could design a go position that would similarly defeat AlphaGo but which we humans would "understand" (which I think may be more accurate then "solve").


What about a position with many intersections where only one player can make a legal move, i.e. a group with many eyes ? Then start a molasses ko. The first one to pass looses the game.
The right move for the one with more legal moves would be to start the molasses ko.

Computers would have to read the sequences until the board is full in order to evaluate the result.

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #9 Posted: Thu Mar 16, 2017 5:23 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Pio2001, that reminds me of a position I once saw with some mutually running ladders with all sorts of ko recaptures going on, it looked rather like a fizzing Conway's game-of-life automaton. All those kos made playing out the ladders require hundreds or thousands of moves which could confuse a computer. Anyone else recognise what I'm talking about and can find a link?

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #10 Posted: Thu Mar 16, 2017 6:09 am 
Gosei
User avatar

Posts: 2011
Location: Groningen, NL
Liked others: 202
Was liked: 1087
Rank: Dutch 4D
GD Posts: 645
Universal go server handle: herminator
Uberdude wrote:
Pio2001, that reminds me of a position I once saw with some mutually running ladders with all sorts of ko recaptures going on, it looked rather like a fizzing Conway's game-of-life automaton. All those kos made playing out the ladders require hundreds or thousands of moves which could confuse a computer. Anyone else recognise what I'm talking about and can find a link?


This one?



This post by HermanHiddema was liked by 3 people: daal, sybob, Uberdude
Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #11 Posted: Thu Mar 16, 2017 6:22 am 
Gosei

Posts: 1494
Liked others: 111
Was liked: 315
Make it stop

_________________
North Lecale

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #12 Posted: Thu Mar 16, 2017 6:35 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Yup Herman, that's it, thanks!

P.S. This kind of fuzzy search is another thing humans are very good at, though computers are catching up.

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #13 Posted: Thu Mar 16, 2017 6:44 am 
Lives in sente

Posts: 727
Liked others: 44
Was liked: 218
GD Posts: 10
I make Leela analyse above sgf file to see if it can come up with correct answer, this is the result

Image

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #14 Posted: Thu Mar 16, 2017 6:49 am 
Oza

Posts: 2180
Location: ʍoquıɐɹ ǝɥʇ ɹǝʌo 'ǝɹǝɥʍǝɯos
Liked others: 237
Was liked: 662
Rank: AGA 5d
GD Posts: 4312
Online playing schedule: Every tenth February 29th from 20:00-20:01 (if time permits)
Whoever invented that had way too much time on their hands!

_________________
Still officially AGA 5d but I play so irregularly these days that I am probably only 3d or 4d over the board (but hopefully still 5d in terms of knowledge, theory and the ability to contribute).

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #15 Posted: Thu Mar 16, 2017 6:59 am 
Lives in gote

Posts: 436
Liked others: 1
Was liked: 38
Rank: KGS 5 kyu
Maybe I am not the strongest player but I can't see how white can win this.

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #16 Posted: Thu Mar 16, 2017 2:15 pm 
Lives in sente
User avatar

Posts: 1206
Liked others: 51
Was liked: 192
Rank: KGS 5d
KGS: Str1fe, Midorisuke
Krama wrote:
Maybe I am not the strongest player but I can't see how white can win this.


Found the computer.


This post by Shaddy was liked by: Drew
Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #17 Posted: Fri Mar 17, 2017 1:31 am 
Lives in gote
User avatar

Posts: 301
Location: Illinois
Liked others: 228
Was liked: 84
Rank: infant
Everyone involved in the Penrose project are likely orders of magnitude smarter than I am, and yet I find this line of thinking:

1. design strange chess problem
2. existing computer chess programs are confused
3. human brains are quantum computers

to be an incredible failure of imagination. Why was the first instinct not to simply tear down the software and see why it's choking? To magic up "because quantum brains" is, I feel, arrogant in the extreme.

Also, @Shaddy, your avatar seems a bit misinformed ;-) Wikipedia: Red Bananas are definitely a (tasty!) thing

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #18 Posted: Fri Mar 17, 2017 8:51 am 
Lives in gote
User avatar

Posts: 388
Location: Riverside CA
Liked others: 246
Was liked: 79
Rank: KGS 7 kyu
KGS: Krill
OGS: Krill
Drew wrote:
Everyone involved in the Penrose project are likely orders of magnitude smarter than I am, and yet I find this line of thinking:

1. design strange chess problem
2. existing computer chess programs are confused
3. human brains are quantum computers

to be an incredible failure of imagination. Why was the first instinct not to simply tear down the software and see why it's choking? To magic up "because quantum brains" is, I feel, arrogant in the extreme.

Also, @Shaddy, your avatar seems a bit misinformed ;-) Wikipedia: Red Bananas are definitely a (tasty!) thing


I think what you're missing is that in the background are more significant indicators (to Penrose) that human brains can solve problems that no mechanical machine can (the move to quantum computing from just this argument, even if successful, is still a bit of a leap). In summary, elsewhere, Penrose has made arguments that the human brain cannot be (and can't be modeled by) a Turing machine or any machine that a Turing machine can model by appealing to facts about what is involved in the proof of Gödel's incompleteness theorems (see here for a brief explanation).

To be clear, this doesn't allow for any direct argument from the chess position to the hypothesis that the human brain is not a mechanical computer, but I guess Penrose considers it to be weakly confirming evidence, in light of what he takes to be a stronger argument (see above), because it shows yet another case where mechanistic computers seem to be limited in comparison to human brains.


This post by Monadology was liked by: joellercoaster
Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #19 Posted: Fri Mar 17, 2017 12:59 pm 
Lives in sente
User avatar

Posts: 1206
Liked others: 51
Was liked: 192
Rank: KGS 5d
KGS: Str1fe, Midorisuke
Drew wrote:
Everyone involved in the Penrose project are likely orders of magnitude smarter than I am, and yet I find this line of thinking:

1. design strange chess problem
2. existing computer chess programs are confused
3. human brains are quantum computers

to be an incredible failure of imagination. Why was the first instinct not to simply tear down the software and see why it's choking? To magic up "because quantum brains" is, I feel, arrogant in the extreme.

Also, @Shaddy, your avatar seems a bit misinformed ;-) Wikipedia: Red Bananas are definitely a (tasty!) thing

Yeah, I've come across red bananas in the supermarket since starting to use this avatar. It might be time for a change.

Top
 Profile  
 
Offline
 Post subject: Re: Humans 1 Computers 0
Post #20 Posted: Fri Mar 17, 2017 2:48 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Monadology wrote:
To be clear, this doesn't allow for any direct argument from the chess position to the hypothesis that the human brain is not a mechanical computer, but I guess Penrose considers it to be weakly confirming evidence, in light of what he takes to be a stronger argument (see above), because it shows yet another case where mechanistic computers seem to be limited in comparison to human brains.


First, all confirming evidence is weak.

Second, this is not evidence of brain vs. computer, because it would not be difficult to write a program that could reason about such positions. Not that that program would be a strong chess player. ;) Computer programs are limited not only by hardware, but by computer programmers. :lol: As we have seen with AlphaGo. Not that other programmers were bad, but it took a new combination of ideas to create AlphaGo.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 20 posts ] 

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: Bing [Bot] and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group