ISF Logo   IS Forum
Forum Index Register Members List Events Mark Forums Read Help

Go Back   International Skeptics Forum » General Topics » Science, Mathematics, Medicine, and Technology
 


Welcome to the International Skeptics Forum, where we discuss skepticism, critical thinking, the paranormal and science in a friendly but lively way. You are currently viewing the forum as a guest, which means you are missing out on discussing matters that are of interest to you. Please consider registering so you can gain full use of the forum features and interact with other Members. Registration is simple, fast and free! Click here to register today.
Reply
Old 17th December 2017, 12:51 AM   #41
Roboramma
Penultimate Amazing
 
Roboramma's Avatar
 
Join Date: Feb 2005
Location: Shanghai
Posts: 11,394
Originally Posted by Peregrinus View Post
Possibly true in the abstract, at least for chess, but a vast number of possible moves would not be made even by a beginner, and a similarly huge number would not be made by a modest player; a chess master would only ever consider a very few moves but spend more time considering the combinations those few moves might lead to.
But that's sort of the point: you have to have some means of differentiating between good moves and bad moves other than just looking at all possible moves. So it's not the finite number of possible moves that makes it possible for a computer, or a person, to play good chess, because neither actually takes advantage of the fact that the number of possible moves is finite. Both determine which is the best move to make at any particular configuration of the board in some way other than going through all possible games and choosing the best one.
__________________
"... when people thought the Earth was flat, they were wrong. When people thought the Earth was spherical they were wrong. But if you think that thinking the Earth is spherical is just as wrong as thinking the Earth is flat, then your view is wronger than both of them put together."
Isaac Asimov
Roboramma is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 17th December 2017, 02:47 AM   #42
lionking
In the Peanut Gallery
 
lionking's Avatar
 
Join Date: Jan 2007
Location: Melbourne
Posts: 42,607
Originally Posted by lauwersw View Post
It seems the comparison of the AI system against the "classic" chess computers wasn't entirely fair: http://www.theregister.co.uk/2017/12...ero_ai_unfair/
I don't see the unfairness. Each "player" was given a minute to move. Stockfish can analyse 70 million positions a minute, Alpha Zero a mere 80k. Stockfish failed to win a single game.

I play chess, not at any great standard, but I understand basic principals. Like protecting your king, and striving for material advantages. AO broke them all. Using the king as forward defence and even attacking. Sacrificing pieces to ensure stockfish had some of its pieces paralysed (see game 10).

This is indeed an AI breakthrough.
__________________
A fanatic is one who can't change his mind and won't change the subject.

Sir Winston Churchill
lionking is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 17th December 2017, 05:40 AM   #43
Fast Eddie B
Philosopher
 
Fast Eddie B's Avatar
 
Join Date: Sep 2010
Location: Mineral Bluff, GA
Posts: 5,324
Originally Posted by dasmiller View Post
I hadn't heard it, but it sounds plausible...

So, I don't have a source for it, but I'd say it's a plausible statement. Unless I made a gross math error, which is entirely possible.
Wow! Thanks for taking the time to work through that. I’ll provide a link shortly to show this has been worked through before.

Quote:
I think the jury's still out as to whether this would have been a better use for Humanity's time.
Grin. At least it might have avoided a lot of wars.


Edited to add: here’s that link: https://en.m.wikipedia.org/wiki/Shannon_number
__________________
“I do solemnly swear (or affirm) that...I will to the best of my ability, preserve, protect and defend the Constitution of the United States.” - President Donald J. Trump, January 20, 2017.
"And it's, frankly, disgusting the way the press is able to write whatever they want to write. And people should look into it." - President Donald J. Trump, October 11, 2017.

Last edited by Fast Eddie B; 17th December 2017 at 05:43 AM.
Fast Eddie B is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 17th December 2017, 07:51 AM   #44
Peregrinus
Graduate Poster
 
Join Date: Feb 2015
Posts: 1,213
Originally Posted by lionking View Post
Using the king as forward defence and even attacking. Sacrificing pieces to ensure stockfish had some of its pieces paralysed (see game 10).

This is indeed an AI breakthrough.
The king can be a valuable, even potent force in some end games; strategic sacrifices can indeed paralyze (or at least severely hamper) one's opponent - see some of Fischer's surprise wins against Spassky.
Peregrinus is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 18th December 2017, 12:32 PM   #45
P.J. Denyer
Illuminator
 
Join Date: Aug 2008
Posts: 4,738
Originally Posted by Arisia View Post
IBM's Watson has gone beyond beating humans on Jeopardy and been put to some practical uses for years at this point.
Thank you! I was trying to remember what the big success for an AI that I was reading about a while back was (not to put down Chess, but ultimately it comes down to number crunching), I was reading that Watson is being used to replace managment positions.
__________________
"I know my brain cannot tell me what to think." - Scorpion
P.J. Denyer is online now   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 18th December 2017, 12:49 PM   #46
Hellbound
Merchant of Doom
 
Hellbound's Avatar
 
Join Date: Sep 2002
Location: Not in Hell, but I can see it from here on a clear day...
Posts: 12,839
Originally Posted by P.J. Denyer View Post
Thank you! I was trying to remember what the big success for an AI that I was reading about a while back was (not to put down Chess, but ultimately it comes down to number crunching), I was reading that Watson is being used to replace managment positions.
I've worked with IBM. I'm pretty sure that's not as intellectually demanding as chess.

Hellbound is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 18th December 2017, 01:05 PM   #47
MikeG
Now. Do it now.
 
MikeG's Avatar
 
Join Date: Sep 2012
Location: UK
Posts: 24,376
A thread on artificial intelligence, not started by PGJ! It's a miracle!
__________________
"The Conservatives want to keep wogs out and march boldly back to the 1950s when Britain still had an Empire and blacks, women, poofs and Irish knew their place." The Don That's what we've sunk to here.
MikeG is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 18th December 2017, 01:12 PM   #48
Fast Eddie B
Philosopher
 
Fast Eddie B's Avatar
 
Join Date: Sep 2010
Location: Mineral Bluff, GA
Posts: 5,324
Originally Posted by P.J. Denyer View Post
...(not to put down Chess, but ultimately it comes down to number crunching)...
Is it not possible that at its core, all intelligence comes down to number crunching?
__________________
“I do solemnly swear (or affirm) that...I will to the best of my ability, preserve, protect and defend the Constitution of the United States.” - President Donald J. Trump, January 20, 2017.
"And it's, frankly, disgusting the way the press is able to write whatever they want to write. And people should look into it." - President Donald J. Trump, October 11, 2017.
Fast Eddie B is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 18th December 2017, 09:57 PM   #49
Puppycow
Penultimate Amazing
 
Puppycow's Avatar
 
Join Date: Jan 2003
Posts: 22,830
Originally Posted by barehl View Post
I'm still waiting to see if any of the current chess playing programs are actually better than human players. We may never find out.
I think they are so far ahead that it wouldn't be fair.

I would like to see Magnus Carlsen, the current best human player vs. a recent program, but I really doubt the human would win. (under whatever rules would be most to his advantage)

http://www.bbc.com/future/story/2015...cant-be-beaten

Quote:
In fact, chess programs today are so strong that even the best human player in the world (currently the 24-year-old Norwegian Magnus Carlsen) wouldn’t have a hope of winning a tournament match.
Quote:
“I have a living room full of computers constantly playing games trying to prove that an idea is better than what we already have,” says Lefler.

Overnight, Lefler’s six computers play through over 14,000 games each during an eight-hour period. “Six machines times 14,000 games is a lot of games,” he says. And with every game played, the database gets deeper and richer.

The result of Lefler’s busily whirring machines is the ever-increasing prowess of Komodo. There is a rating system for chess ability that applies to both humans and computers called the “Elo rating system”, after Arpad Elo, a physics professor who invented it.

Magnus Carlsen’s Elo is currently 2,850 while Komodo’s sits comfortably above that – at 3,350.
There is absolutely no reason that I can think of other than willingness why Carlsen couldn't challenge a top computer program. I highly doubt they are ducking him. Pretty sure these new programs are better than the one that beat Kasparov back in the day, too.

Just seems like an odd thing to doubt.
__________________
A fool thinks himself to be wise, but a wise man knows himself to be a fool.
William Shakespeare
Puppycow is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 02:12 AM   #50
Darat
Lackey
Administrator
 
Darat's Avatar
 
Join Date: Aug 2001
Location: South East, UK
Posts: 84,224
It will be some semantic twiddle such as "computers can't play, they are not thinking"
__________________
I wish I knew how to quit you
Darat is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 02:49 AM   #51
lionking
In the Peanut Gallery
 
lionking's Avatar
 
Join Date: Jan 2007
Location: Melbourne
Posts: 42,607
Originally Posted by Peregrinus View Post
The king can be a valuable, even potent force in some end games; strategic sacrifices can indeed paralyze (or at least severely hamper) one's opponent - see some of Fischer's surprise wins against Spassky.
Oh, yes indeed, but Alpha Zero has been using the king earlier and more creatively.

Chess has become a war of attrition with grandmasters rarely losing when playing white, but also not often winning. This AI has brought beauty back to chess in my view, and can't wait to see Alpha Zero play another AI with similar characteristics. This seems to be the only challenge Alpha Zero will have. It can easily beat the best chess programs, and the best humans in my view.
__________________
A fanatic is one who can't change his mind and won't change the subject.

Sir Winston Churchill
lionking is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 03:36 AM   #52
3point14
Pi
 
3point14's Avatar
 
Join Date: Nov 2005
Posts: 16,256
Originally Posted by lionking View Post
I don't see the unfairness. Each "player" was given a minute to move. Stockfish can analyse 70 million positions a minute, Alpha Zero a mere 80k. Stockfish failed to win a single game.

Because what they're playing isn't traditional chess. They've amended the rules to accommodate their AI.

Essentially, they've managed to come up with a machine that's really good at playing a game that's very nearly chess. My weak inference from this is that they couldn't come up with a machine that's as good at playing chess as this machine is at playing nearly chess.

It's a fudge.
__________________
Up the River!
3point14 is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 03:39 AM   #53
Darat
Lackey
Administrator
 
Darat's Avatar
 
Join Date: Aug 2001
Location: South East, UK
Posts: 84,224
Originally Posted by 3point14 View Post
Because what they're playing isn't traditional chess. They've amended the rules to accommodate their AI.

Essentially, they've managed to come up with a machine that's really good at playing a game that's very nearly chess. My weak inference from this is that they couldn't come up with a machine that's as good at playing chess as this machine is at playing nearly chess.

It's a fudge.
Time limits on moves are a part of chess tournaments: https://en.wikipedia.org/wiki/Time_control
__________________
I wish I knew how to quit you
Darat is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 03:52 AM   #54
3point14
Pi
 
3point14's Avatar
 
Join Date: Nov 2005
Posts: 16,256
Originally Posted by Darat View Post
Time limits on moves are a part of chess tournaments: https://en.wikipedia.org/wiki/Time_control

Yes, in a very specific way.

The AI in this instance does not use a time-clock in the same way people and nearly all other chess programs do.


From the article:

"Next, DeepMind's paper stated that both systems, AlphaZero and Stockfish, were given one minute to make a move."


One minute to make a move is not how chess usually works.
__________________
Up the River!
3point14 is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 04:47 AM   #55
lionking
In the Peanut Gallery
 
lionking's Avatar
 
Join Date: Jan 2007
Location: Melbourne
Posts: 42,607
Originally Posted by 3point14 View Post
Yes, in a very specific way.

The AI in this instance does not use a time-clock in the same way people and nearly all other chess programs do.


From the article:

"Next, DeepMind's paper stated that both systems, AlphaZero and Stockfish, were given one minute to make a move."


One minute to make a move is not how chess usually works.
How much time do you want? Stockfish analyse tens of millions positions every second.

And explain how this is massively different from speed chess rules?
__________________
A fanatic is one who can't change his mind and won't change the subject.

Sir Winston Churchill
lionking is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 05:30 AM   #56
Fast Eddie B
Philosopher
 
Fast Eddie B's Avatar
 
Join Date: Sep 2010
Location: Mineral Bluff, GA
Posts: 5,324
The time allowed per move in a chess tournament has evolved over the years.*

Yes, giving computers less time is dumbing them down a bit. A level playing field would give them equal time.

I suppose the shortened time was to allow for the fundamental speed advantage of silicon over biological thinking machines.

Other than that, chess programs are playing under exactly the same rules of chess as humans.

As an aside, there’s a nice little documentary called “Magnus” available on Netflix. Well done, and has both human interest and chess related appeal.


*There used to be a forced adjournment on the 40th move. I believe black’s 40th move had to be made, but was “sealed”, only to be revealed the following day. Teams of coaches would help each side analyze the position overnight. I think the advent of computers helped do away with that protocol - games are now played to their conclusion in one sitting.
__________________
“I do solemnly swear (or affirm) that...I will to the best of my ability, preserve, protect and defend the Constitution of the United States.” - President Donald J. Trump, January 20, 2017.
"And it's, frankly, disgusting the way the press is able to write whatever they want to write. And people should look into it." - President Donald J. Trump, October 11, 2017.

Last edited by Fast Eddie B; 19th December 2017 at 05:35 AM.
Fast Eddie B is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 05:35 AM   #57
3point14
Pi
 
3point14's Avatar
 
Join Date: Nov 2005
Posts: 16,256
Originally Posted by lionking View Post
How much time do you want? Stockfish analyse tens of millions positions every second.

I don't really want anything. I think it's notable that this machine isn't playing chess by the standard rules of chess.



Quote:
And explain how this is massively different from speed chess rules?

This machine gets a minute to make each move.

Usual chess rules, including speed chess variants, allow x minutes in total to play all of one's moves (normally with + x seconds per move made).

One of the skills of chess is managing the clock. This machine doesn't have to do that.
__________________
Up the River!
3point14 is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 05:52 AM   #58
Meadmaker
Penultimate Amazing
 
Meadmaker's Avatar
 
Join Date: Apr 2004
Posts: 18,353
Originally Posted by lionking View Post
How much time do you want? Stockfish analyse tens of millions positions every second.

And explain how this is massively different from speed chess rules?
The point 3.14 is making doesn't have anything to do with the amount of time. It's about how the time is allocated.


In a tournament, chess players are given a certain amount of time to make a certain number of moves. This is true whether playing speed chess, which is typically 5 minutes to make all moves, or championship chess, where players are given 4 hours. If a player makes one move very rapidly, he can use more time on the more difficult moves. Stockfish is programmed to make use of the amount of time remaining as an input determining how deep the search should be for that particular move. AlphaZero was not programmed that way. By giving Stockfish a fixed amount of time per move, instead of one of the standard sorts of time controls, they were playing to AlphaZero's strength.

I would be very surprised if that turned out to be a major factor in AlphaZero's victory, but it helped. I suspect that it was done that way just because the "tournament" was probably conducted by software, and there was no way to set asymmetric time controls, where AlphaZero got a fixed one minute for each move, but Stockfish got 60 minutes, or 45 minutes, per game, to allocate as it saw fit.

More telling was the decision not to allow Stockfish to use its opening book. That decreased Stockfish's rating considerably. I have heard it said that this was just "fairness" because AlphaZero didn't use an opening book. Well, ok, that's fine, but, simply put, it would have been more impressive if AlphaZero had beaten Stockfish badly while Stockfish was using an opening book, and it would have been a trivial exercise to allow Stockfish to use an opening book. In fact, it is so trivial that I am confident they did it, and they chose to publish the results from the version without the opening book, because the results were more impressive.

Still, it is an impressive result. The greatest thing about this particular win, in my opinion, is the style of chess that was played by AlphaZero. After Deep Blue defeated Kasparov, the Chess world was not abuzz with the commentary about how Deep Blue played. Deep Blue played nearly flawless, but uninspiring, Chess. it just never made a mistake. After this tournament, people described AlphaZero's playing style as "alien", making moves that no human would have ever made, but which turned out to be devastating to its opponent.

In terms of the grander significance, whether this turns out to be a major breakthrough or a parlor trick, we'll see if the folks at DeepMind can use similar algorithms to do something useful. I would even like to see if it can be turned to other board games that have different techniques. Can it play a hex-and-counter wargame? Or even "Ticket to Ride"?
Meadmaker is online now   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 06:51 AM   #59
Fast Eddie B
Philosopher
 
Fast Eddie B's Avatar
 
Join Date: Sep 2010
Location: Mineral Bluff, GA
Posts: 5,324
To help establish my nerd cred, this is from my library:



Bear in mind this match was in 1996! A lot has changed in 21 years.

Anyway, some reflections in the final paragraph here:

__________________
“I do solemnly swear (or affirm) that...I will to the best of my ability, preserve, protect and defend the Constitution of the United States.” - President Donald J. Trump, January 20, 2017.
"And it's, frankly, disgusting the way the press is able to write whatever they want to write. And people should look into it." - President Donald J. Trump, October 11, 2017.
Fast Eddie B is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 09:48 AM   #60
3point14
Pi
 
3point14's Avatar
 
Join Date: Nov 2005
Posts: 16,256
Originally Posted by Meadmaker View Post
The point 3.14 is making doesn't have anything to do with the amount of time. It's about how the time is allocated.
(truncated by me)


Thank you for explaining so well what I was struggling to.

I don't claim to be able to play chess (I know how the pieces move, but that's not the same thing.) but I do occasionally enjoy watching Masters play blitz on the internet.


This is my favourite:

https://www.youtube.com/watch?v=ct-BsxbYI0k
__________________
Up the River!
3point14 is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 19th December 2017, 10:45 PM   #61
rjh01
Gentleman of leisure
Tagger
 
rjh01's Avatar
 
Join Date: May 2005
Location: Flying around in the sky
Posts: 24,109
I am waiting on a chess tournament where all the players are computers. There would have to be certain rules on how big the computers can be, plus timing rules, plus direct computer to computer communications with all computers displaying the board and moves on a screen.
__________________
This signature is for rent.
rjh01 is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th December 2017, 12:27 AM   #62
lauwersw
Scholar
 
Join Date: Mar 2008
Posts: 121
Originally Posted by lionking View Post
I don't see the unfairness. Each "player" was given a minute to move. Stockfish can analyse 70 million positions a minute, Alpha Zero a mere 80k. Stockfish failed to win a single game.

This is indeed an AI breakthrough.
Another point of unfairness raised in that linked article is the fact that Stockfish was given a standard PC while the AI required 5000 cloud instances. That's an immense difference in computing power.

I'm not denying it's a breakthrough for AI, but it's not quite on par with classic computing for practical purposes.
lauwersw is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th December 2017, 12:52 AM   #63
Roboramma
Penultimate Amazing
 
Roboramma's Avatar
 
Join Date: Feb 2005
Location: Shanghai
Posts: 11,394
Originally Posted by lauwersw View Post
Another point of unfairness raised in that linked article is the fact that Stockfish was given a standard PC while the AI required 5000 cloud instances. That's an immense difference in computing power.

I'm not denying it's a breakthrough for AI, but it's not quite on par with classic computing for practical purposes.
I'm not quite sure I understand what the hardware used was, but I think what you wrote here is incorrect. I think that they tried to use similar hardware for both but because Alpha Zero doesn't run on a conventional computer it's difficult to make a comparison from one to the other.

I might also be wrong of course.
__________________
"... when people thought the Earth was flat, they were wrong. When people thought the Earth was spherical they were wrong. But if you think that thinking the Earth is spherical is just as wrong as thinking the Earth is flat, then your view is wronger than both of them put together."
Isaac Asimov
Roboramma is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th December 2017, 05:54 AM   #64
Fast Eddie B
Philosopher
 
Fast Eddie B's Avatar
 
Join Date: Sep 2010
Location: Mineral Bluff, GA
Posts: 5,324
Originally Posted by lauwersw View Post

I'm not denying it's a breakthrough for AI, but it's not quite on par with classic computing for practical purposes.
To clarify, the reason I called it a “landmark” was not based on it winning or playing “better”, but on the fact that it taught itself from a virtual tabla rasa, with no strategies or weighing or “book” fed to it by humans. That it did so in a stunningly short time is just gravy.

For those protesting it did not need to deal with the vagaries of conventional time controls, I have little doubt it could teach itself how to deal with any time controls one could throw at it.
__________________
“I do solemnly swear (or affirm) that...I will to the best of my ability, preserve, protect and defend the Constitution of the United States.” - President Donald J. Trump, January 20, 2017.
"And it's, frankly, disgusting the way the press is able to write whatever they want to write. And people should look into it." - President Donald J. Trump, October 11, 2017.
Fast Eddie B is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th December 2017, 09:24 AM   #65
casebro
Penultimate Amazing
 
casebro's Avatar
 
Join Date: Jun 2005
Posts: 15,975
So if we judge AI by ability to learn chess, they seem to be at least on a par with humans.

Next mile stone will be to play a game with no rules. We humans call it Life.
__________________
Great minds discuss ideas.
Medium minds discuss events.
Small minds spend all their time on U-Tube and Facebook.
casebro is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 20th December 2017, 09:29 AM   #66
GlennB
Loggerheaded, earth-vexing fustilarian
 
GlennB's Avatar
 
Join Date: Sep 2006
Location: Arcadia, Greece
Posts: 23,819
Originally Posted by casebro View Post
So if we judge AI by ability to learn chess, they seem to be at least on a par with humans.

Next mile stone will be to play a game with no rules. We humans call it Life.
That idea sent a shiver down my spine. Imagine if the AI used the behaviour of 'successful' people like Trump and Putin as its models
__________________
"Even a broken clock is right twice a day. 9/11 truth is a clock with no hands." - Beachnut
GlennB is online now   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 22nd December 2017, 02:38 PM   #67
P.J. Denyer
Illuminator
 
Join Date: Aug 2008
Posts: 4,738
Originally Posted by GlennB View Post
That idea sent a shiver down my spine. Imagine if the AI used the behaviour of 'successful' people like Trump and Putin as its models
Red Dwarf Season 10 Episode 2 (Fathers and Suns) gives a pretty good idea of what that would look like.
__________________
"I know my brain cannot tell me what to think." - Scorpion
P.J. Denyer is online now   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 8th January 2018, 08:06 PM   #68
barehl
Master Poster
 
barehl's Avatar
 
Join Date: Jul 2013
Posts: 2,649
Originally Posted by Roboramma View Post
Can you clarify what you're waiting for? It seems like by "better" you mean something other than "better at winning games of chess", as, at least as I understand it, they already are.

I'm probably just being dense here, but what exactly are you waiting for?
The game of chess was designed with human limitations in mind. It's more likely that the chess playing programs are exploiting aspects of the game that were never considered because of human limitations.

Imagine this scenario. Robots are playing a game with dice. These robots have a cup-like device on their left arm with a concave bottom. The manipulator on their right arm cannot reach the bottom of this cup. They pick the dice up one at a time and drop them into the cup. When all the dice are in the cup they open the bottom of the cup and drop the dice onto the board. When you read through the rules you realize that they are based on the limitations of these robots and that you can exploit them. You pick up the dice one at a time and put them in your other palm. You arrange the dice to a favorable pattern and then pick them all back up and simply set them on the board. The rules contain no restrictions about shaking the dice or dropping them because it is physically impossible for the robots to use the dice without randomizing them. Using this strategy you win every game. Are you actually a better player?

I'm waiting for a game where whatever advantages a non-human player might have are removed. We might already be at the point where a computer chess program is better, but without a proper game I have no way of knowing.

Last edited by barehl; 8th January 2018 at 08:09 PM.
barehl is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 8th January 2018, 08:21 PM   #69
Meadmaker
Penultimate Amazing
 
Meadmaker's Avatar
 
Join Date: Apr 2004
Posts: 18,353
Originally Posted by barehl View Post
The game of chess was designed with human limitations in mind. It's more likely that the chess playing programs are exploiting aspects of the game that were never considered because of human limitations.

Imagine this scenario. Robots are playing a game with dice. These robots have a cup-like device on their left arm with a concave bottom. The manipulator on their right arm cannot reach the bottom of this cup. They pick the dice up one at a time and drop them into the cup. When all the dice are in the cup they open the bottom of the cup and drop the dice onto the board. When you read through the rules you realize that they are based on the limitations of these robots and that you can exploit them. You pick up the dice one at a time and put them in your other palm. You arrange the dice to a favorable pattern and then pick them all back up and simply set them on the board. The rules contain no restrictions about shaking the dice or dropping them because it is physically impossible for the robots to use the dice without randomizing them. Using this strategy you win every game. Are you actually a better player?

I'm waiting for a game where whatever advantages a non-human player might have are removed. We might already be at the point where a computer chess program is better, but without a proper game I have no way of knowing.
When it comes to chess, there's really no contest. Computers are playing with the exact same limitations as humans, as far as the rules go, and they win. When Deep Blue defeated Kasparov, you needed a special computer to be as good as the best human. By today, a program running on your laptop can do it, and do it with no special exploitation of human weakness, unless you count being forced to use a carbon based brain as a weakness.

As for alphazero, first indications are that this thing is really, really good at perfect information, abstract strategy games. Although the only games that it has been tried on, with published results, are Chess, Shogi, and Go, I'm guessing it could play a mean game of Chinese Checkers or Mancala, too.
Meadmaker is online now   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 11th January 2018, 12:43 AM   #70
Sceptic-PK
Illuminator
 
Join Date: Jun 2010
Posts: 3,720
I'm more interested in the ability of AI to beat human opponents in real-time environments (though, that recent Texas Hold'em win discussed here was pretty impressive).

Games that are real-time however, have a lot more going on and a lot more factors to consider, because your opponent isn't restricted to wait their turn to move. The execution of strategy is a lot more complex in real-time, because your opponent is reacting to you simultaneously.

The top tournament strategy games in 2018 are more complex than chess, I can't imagine the crazy coding that would be required to beat a professional human opponent in something like a proper DOTA2 match. Elon Musk & friends had a crack and designed an AI that could beat pros, but with pretty severe caveats that make it worthless in a real game.

https://arstechnica.com/gaming/2017/...ta-2-and-wins/
Sceptic-PK is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 11th January 2018, 04:47 AM   #71
Meadmaker
Penultimate Amazing
 
Meadmaker's Avatar
 
Join Date: Apr 2004
Posts: 18,353
Originally Posted by Sceptic-PK View Post
I'm more interested in the ability of AI to beat human opponents in real-time environments (though, that recent Texas Hold'em win discussed here was pretty impressive).

Games that are real-time however, have a lot more going on and a lot more factors to consider, because your opponent isn't restricted to wait their turn to move. The execution of strategy is a lot more complex in real-time, because your opponent is reacting to you simultaneously.

The top tournament strategy games in 2018 are more complex than chess, I can't imagine the crazy coding that would be required to beat a professional human opponent in something like a proper DOTA2 match. Elon Musk & friends had a crack and designed an AI that could beat pros, but with pretty severe caveats that make it worthless in a real game.

https://arstechnica.com/gaming/2017/...ta-2-and-wins/
Agreed, mostly.

Perfect information abstract strategy games, which are any games where there is no luck, all information is known, and players move alternately, are very well suited to being able to represent in the form a neural network can deal with. Real problems that humans have to deal with, including other sorts of games, often share none of those characteristics. Whether it's a real time strategy game, or even something like a hex and counter war game, you have issues of randomness, risk versus reward, large tasks that have to be broken down into smaller problems, but where the pieces still have to all work together.

It will be interesting to see where the technology goes from here.
Meadmaker is online now   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 11th January 2018, 06:26 AM   #72
Fast Eddie B
Philosopher
 
Fast Eddie B's Avatar
 
Join Date: Sep 2010
Location: Mineral Bluff, GA
Posts: 5,324
My feeling is that in any situation where there are “rules”, AI will be able to cope.

The question is, can virtually anything be reduced to “rules”? I lean towards that being true.

I saw a recent study where AI learned to spot cancer cells in biopsies. It may not be this one, but the conclusion was that humans do better in one regard, AI does better in another, but combined they complement each other to provide very high accuracy.

http://fortune.com/2017/10/30/ai-ear...cer-detection/

So, rules can be applied for cancer detection. Could not most medical diagnoses, legal matters, engineering problems and the like ultimately be reduced to rules?

Problems involving human interaction are a tougher nut, but unless something spiritual or supernatural is invoked, I suspect those can be reduced to rules as well.
__________________
“I do solemnly swear (or affirm) that...I will to the best of my ability, preserve, protect and defend the Constitution of the United States.” - President Donald J. Trump, January 20, 2017.
"And it's, frankly, disgusting the way the press is able to write whatever they want to write. And people should look into it." - President Donald J. Trump, October 11, 2017.

Last edited by Fast Eddie B; 11th January 2018 at 06:27 AM.
Fast Eddie B is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 11th January 2018, 06:47 AM   #73
Meadmaker
Penultimate Amazing
 
Meadmaker's Avatar
 
Join Date: Apr 2004
Posts: 18,353
Originally Posted by Fast Eddie B View Post
My feeling is that in any situation where there are “rules”, AI will be able to cope.

.....

The question is, can virtually anything be reduced to “rules”? I lean towards that being true.
I agree. The question in my mind is whether alphazero represents the breakthrough that will allow AI to be applied in a wide variety of situations, or if it is limited to a very narrow range of problems, like chess.
Meadmaker is online now   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 11th January 2018, 08:14 AM   #74
Cheetah
Graduate Poster
 
Cheetah's Avatar
 
Join Date: Feb 2010
Posts: 1,077
How about training an AI with lots of images as input and then giving it a low quality photo and require it to fill in the details?

What type of a problem would that be?

More a pattern recognition and memory recall problem? I'm not sure, it doesn't really seem to have any "rules".

Here are three images, the first is the input image, the second the AI's guess at filling in the details and the third is the original photo.



It is by no means perfect, but still incredibly impressive to me.

Single Image Super-Resolution Through Automated Texture Synthesis

The things AI are good at now are not really problems with rules, but simple enough problems that a single trained NN can cope with it.
More complex stuff will need multiple specialized NN's to work together.

Early days, I'm sure it won't be long before we have AI's capable of coping with more complex and random problems.
__________________
"... when you dig my grave, could you make it shallow so that I can feel the rain" - DMB
Cheetah is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 11th January 2018, 08:30 AM   #75
Fast Eddie B
Philosopher
 
Fast Eddie B's Avatar
 
Join Date: Sep 2010
Location: Mineral Bluff, GA
Posts: 5,324
That's pretty cool - thanks!

A lot of that sort of image processing goes on in the best cellphone cameras. Taking imperfect "raw" data and massaging it to get a more pleasing image.

The number of megapixels utilized is becoming less and less important, and the image processing more and more so.

Apple, somewhat conceitedly I think, even calls one of their processors a "bionic" chip!
__________________
“I do solemnly swear (or affirm) that...I will to the best of my ability, preserve, protect and defend the Constitution of the United States.” - President Donald J. Trump, January 20, 2017.
"And it's, frankly, disgusting the way the press is able to write whatever they want to write. And people should look into it." - President Donald J. Trump, October 11, 2017.
Fast Eddie B is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 11th January 2018, 08:35 AM   #76
Cheetah
Graduate Poster
 
Cheetah's Avatar
 
Join Date: Feb 2010
Posts: 1,077
This is also quite cool, you can upload a selfie and have it reconstruct a 3D version of your face. Also not perfect but impressive as well.

3D Face Reconstruction from a Single Image

How about creating fake celebrities?



These are not real photos of real people.

When two competing neural networks result in photorealistic face.
__________________
"... when you dig my grave, could you make it shallow so that I can feel the rain" - DMB
Cheetah is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 12th January 2018, 09:25 AM   #77
Beelzebuddy
Philosopher
 
Beelzebuddy's Avatar
 
Join Date: Jun 2010
Posts: 6,369
Originally Posted by Cheetah View Post
How about training an AI with lots of images as input and then giving it a low quality photo and require it to fill in the details?

What type of a problem would that be?

More a pattern recognition and memory recall problem? I'm not sure, it doesn't really seem to have any "rules".
Sure it does. Before it can do such impressive feats, you have to feed it millions of similar tasks with the correct answers to teach it. The "AI" is just a solid block of inference rules, whose weights are refined over the training.
Beelzebuddy is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 12th January 2018, 09:39 AM   #78
Fast Eddie B
Philosopher
 
Fast Eddie B's Avatar
 
Join Date: Sep 2010
Location: Mineral Bluff, GA
Posts: 5,324
Originally Posted by Beelzebuddy View Post
Before it can do such impressive feats, you have to feed it millions of similar tasks with the correct answers to teach it. The "AI" is just a solid block of inference rules, whose weights are refined over the training.
I think that misses the significance of the chess learning program referenced in the OP.

It was specifically not fed "millions of similar tasks with the correct answers to teach it." It figured those out on its own from the basic rules of the game. It was never "taught" that a queen is "worth more" than a pawn - it had to figure that out on its own.
__________________
“I do solemnly swear (or affirm) that...I will to the best of my ability, preserve, protect and defend the Constitution of the United States.” - President Donald J. Trump, January 20, 2017.
"And it's, frankly, disgusting the way the press is able to write whatever they want to write. And people should look into it." - President Donald J. Trump, October 11, 2017.
Fast Eddie B is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 12th January 2018, 09:45 AM   #79
Cheetah
Graduate Poster
 
Cheetah's Avatar
 
Join Date: Feb 2010
Posts: 1,077
Originally Posted by Beelzebuddy View Post
Sure it does. Before it can do such impressive feats, you have to feed it millions of similar tasks with the correct answers to teach it. The "AI" is just a solid block of inference rules, whose weights are refined over the training.
Are you saying it is functionally like a bunch of "if low-rez looks like this, then the detail/texture should look like this" statements?
Are those the rules you mean?
__________________
"... when you dig my grave, could you make it shallow so that I can feel the rain" - DMB
Cheetah is offline   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Old 12th January 2018, 11:33 AM   #80
Meadmaker
Penultimate Amazing
 
Meadmaker's Avatar
 
Join Date: Apr 2004
Posts: 18,353
Originally Posted by Fast Eddie B View Post
I think that misses the significance of the chess learning program referenced in the OP.

It was specifically not fed "millions of similar tasks with the correct answers to teach it." It figured those out on its own from the basic rules of the game. It was never "taught" that a queen is "worth more" than a pawn - it had to figure that out on its own.
Sort of. It didn't get fed "the correct answers", it just tried things to see if it ended up with "the correct answer", i.e. a victory.

And it's hard to say exactly what it figured out. One thing about neural nets is that they are pretty opaque. No one really knows what "rules" it was using, or if they do they haven't published it yet.


For what it's worth, one of the things that made its style of chess so "alien", as it was described in the press conference, was that it seemed incredibly willing to sacrifice material to gain a positional advantage. I'm sure it didn't trade queens for pawns all that often, but it would be more likely to do that than a human player would.
Meadmaker is online now   Quote this post in a PM   Nominate this post for this month's language award Copy a direct link to this post Reply With Quote Back to Top
Reply

International Skeptics Forum » General Topics » Science, Mathematics, Medicine, and Technology

Bookmarks

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump


All times are GMT -7. The time now is 04:39 AM.
Powered by vBulletin. Copyright ©2000 - 2018, Jelsoft Enterprises Ltd.

This forum began as part of the James Randi Education Foundation (JREF). However, the forum now exists as
an independent entity with no affiliation with or endorsement by the JREF, including the section in reference to "JREF" topics.

Disclaimer: Messages posted in the Forum are solely the opinion of their authors.