Arimaa Forum - Print Page


    
      
        Arimaa Forum
        (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
      

        Arimaa >> General Discussion >> when will computers win ?
        
(Message started by: PMertens on May 24^th, 2005, 1:40am)

Title: when will computers win ?
Post by PMertens on May 24^th, 2005, 1:40am

some people seem to think much higher of computers then I do ... and they get paid a hell of a lot better than me :-(

http://edition.cnn.com/2005/TECH/05/23/brain.download/

Title: Re: when will computers win ?
Post by omar on May 24^th, 2005, 4:54pm

Neat stuff.

Also check out:

http://www.kurzweilai.net/meme/frame.html?main=/articles/art0063.html

That's why there's a time limit on the Arimaa challenge till the start of year 2020. My guess is that by 2020 the hardware should be fast enough that a current software version of a decent Arimaa program running on that hardware should be able to win the challenge. But I really beleive that the challenge can be acheived before that if there are major advances in AI software.

Title: Re: when will computers win ?
Post by PMertens on May 24^th, 2005, 11:34pm

well ... computers for president :-)

I always wanted to outsource the thinking ... so I can concentrate on the important things in live:

sleeping, eating, drinking and some things I wont talk about since there might be minors in this forum ;-)

Title: Re: when will computers win ?
Post by 99of9 on May 25^th, 2005, 2:13am

Yeah Paul, it's better not to mention programming in the presence of minors.

Title: Re: when will computers win ?
Post by 99of9 on May 28^th, 2005, 4:14am

on 05/28/05 at 02:37:56, Arimanator wrote:

Now about the omniscient computer , I bet a golden Cadillac that in 20 years there will still be articles about an incredible breakthrough in the next... 20 years ;D that seems to be the standard horizon for these kind of predictions

Two questions:

1) Are you denying that there have been incredible computer breakthroughs in the last 20 years?

2) Are you willing to put your golden Cadillac where your mouth is in a testable arena, arimaa? Specifically are you willing to bet that the arimaa challenge will not be won by 2020? Perhaps that's what Paul was angling at when he said it would be testable.

(Of course I agree that in 20 years there will be predictions of more breakthroughs... perhaps they will set their sights on Go by then.)

on 05/28/05 at 02:37:56, Arimanator wrote:

Let me just give you an example, if the speculations of 20 years ago about the present were verified there should be autonomic permanent dwellings on the moon, manned orbital stations around Mars, in the asteroid belt and why not around Jupiter and Saturn. And the people who predicted that were the very ones who conceived the rockets at that time so as the say goes it doesn't take a rocket scientist to know the value of such extrapolations.

If you like making bets about this kind of thing, I can recommend another game I play:
www.ideosphere.com

Title: Re: when will computers win ?
Post by PMertens on May 29^th, 2005, 5:11pm

you are absolutely correct:
currently we do not know how to "implement" intelligence.
To compare a massively parrallel but fuzzy computer (brain) with a normal PC does not make any sense and saying a PS3 has 1% of the human brains capacity is a useless bit of information.

There are plenty of things machines can do much better then we humans can - and obviously plenty of things we can do better.
Unfortunately for us its slowly shifting to the machines (even you wont argue that - if you do I can provide plenty of examples :-P )

Conscience and common sense might be on the far end of the list of things computers will eventually achieve - but they are on the list.

Do you believe a human brain could be simulated 1:1 if we just had enough processing power or not ?
If yes - then we could not only simulate but copy (and improve)
If no - why is that ?

A computer winning Arimaa is not yet a sign for intelligent computers but for a better understanding of the human brain.
And by the way:
my bet is that the arimaa challenge will be met before 2010 !
I really doubt we will have any self aware computers at that time though :-)

But the day the first truly conscious computer starts thinking will come - and when that happens it will speed up the process again (because a more intelligent computer might build a more sophisticated solution.

Some things are still beyond the reach of mankind, but not for ever.
We will build intelligent computers
We will understand the remarkable programming language called genetic code (something I think is much further away than AI)

And hopefully we wont self-destruct us in the process :-/

Title: Re: when will computers win ?
Post by 99of9 on May 30^th, 2005, 4:58am

Of course you have a point. In fact you have many ;-). There's no way I can give my opinions on the whole thread, but I'll give you a few.

Computer hardware is still improving. It has improved very fast, and continues to do so (albeit at a slightly declining rate). I think this is likely to continue for some time. This already means that if we can figure out how to write good software for a certain problem, the computer is usually better and quicker at solving it than we are. [I'm thinking simple things like solving equations etc].

Writing software is difficult, and the person doing it has only human intelligence.

It is especially difficult to get a computer to do something that requires embodiment (as in the spider/monkey/marmoset/pickup-things robots you were hoping for), because then you need specialised electronics and mechanics for the particular task as well as software to control it. Significant financial gains must be available, and I'm not sure those incentives are really there for making a toy insect (I agree, this is all the software would be able to handle at the moment).

How about the turing test? That doesn't require embodiment! True... but I think the Turing test is a pretty high bar to set out as the requirement for a "computer breakthrough":

For starters the computer would have to understand a language that is not it's own so well that it is impossible to detect. Not many humans can do that! (I admit that many european language native speakers can do a pretty good job of it in English, but there's usually something that gives them away, even if it is only cultural.) (And hey, I'd probably fail at emulating an American, so what hope does a computer have ;-))

Secondly I am of the opinion that the Turing test is basically an all or nothing problem. Toddlers go very quickly from "total failure at turing test" to "pretty much a pass". There's not much of a gradient to see how much or whether you're improving at understanding a conversation.

Thirdly if a computer *could* pass the Turing test, then it could automatically solve any problem you could reasonably expect an average human to be able to solve. So we could (through a conversation involving practice games of arimaa) teach it to play arimaa at least as well as the average human.

To pass the Turing test, the computer will also already need to have common sense. As you say, we have no idea how to implement it. I think that will continue to be the case. Rule based knowledge-bases are not going to ever solve the problem, because as you say, there are millions of rules we learnt without thought.

Will computers ever get this common sense? Not for a long time. In fact not until after one of them has grown up in the real world, interacting as a human would, and learning from it's experiences. I know some researchers are trying to do that now, but I personally believe they do not have sufficient "innate learning software" programmed in to get their robo-child very far. I expect we will only have this software once computers are routinely used to *write* software, given an objective defined in a limited and well defined "language".

Certainly not by 2020. More like 2100 or 2200.... or Fritzlein's maximum rating ;-). Which poses the problem, what if his rating continues to skyrocket forever - then common sense will never be implemented!

Brain downloads? Raw simulations of the human mind? Not until at least 2075. Those guys only get attention because they claim an early deadline. It is little wonder that by 2020 someone will still be claiming we are on the verge of discovering it.

I think breaking the arimaa challenge is a priori more likely than the Turing test. I am encouraged in this opinion, because despite a few tussles with Kasparov et al, it seems to me that computers (not your amiga btw) have now surpassed humans at chess. I know some people say this is "just calculation not real intelligence", but when chess programmers started out, the prevailing view was that chess represented one of the unrepeatable mysteries of the human brain. To them I say that moving the goalposts as to what constitutes AI is not a fair contest. I think arimaa is a little harder for a computer than chess, but not amazingly so. Perhaps our difference in perspective about conquering humanity at arimaa comes about because I know from experience how fallible the top human player is?

[The server told my my message was too long, so I've cut it in half :-)]

Title: Re: when will computers win ?
Post by 99of9 on May 30^th, 2005, 4:58am

Some "AI" is better for humanity than other "AI". Surpassing us at chess has not really enriched our world all that much - it has just meant that the GMs start learning from the computer and preparing with it. I don't think that conquering arimaa will achieve that much in itself either. [We shall see if it has any other ai spinoffs as Omar is hoping.] I think some other applications of computers have already somewhat enriched our world, which is why I asked whether you were claiming that not a single computer breakthrough had yet occurred. My job, and the understanding it can give of the physics and chemistry of condensed matter, would not be possible without the achievements to date in computing.

Will we make progress on cancer in the next 15 years? Yes, I think we probably will, but not to the extent that we would all hope for.

But let me be clear, I am no humanist. These "advancements" may help, but they do not solve the biggest problems of this world. My understanding is that many of the biggest difficulties in this world stem from the corrupt and self-serving human heart. Will we make progress on malaria? I am told that this kills more monthly than the tsunami. This is a problem of the heart and the wallet, not a scientific problem anymore (though perhaps medical science can make theh cure cheaper). Will we stop blowing up each other's buildings and cities? Will dictatorships be abolished? I think unlikely, although I respect those trying, and am willing to try to help.

I will save the topic of betting for another day. Stay tuned ;-).

Title: Re: when will computers win ?
Post by Fritzlein on Jun 4^th, 2005, 10:19pm

on 06/03/05 at 10:48:43, Arimanator wrote:

Now if they're a significant* number of players by 2010, which is in 5 years only by the way, I am dead certain that not only the challenge will go unanswered but also that the distance between the best player and the best bot will have increased.

I think it is an important point that the size of the human community playing Arimaa will have a significant impact on the quality of human play. I expect it is more significant for the human side than the size of the programming community working on the bot side. It is easier for humans to teach each other (directly or indirectly) about Arimaa than it is for bot developers to swap useful insights about Arimaa, isn't it?

Title: Re: when will computers win ?
Post by PMertens on Jun 6^th, 2005, 4:45pm

http://www.tomshardware.com/hardnews/20050606_173339.html