0894: "Progeny"

This forum is for the individual discussion thread that goes with each new comic.

Moderators: Moderators General, Prelates, Magistrates

Avaviel
Posts: 5
Joined: Mon Dec 20, 2010 5:08 am UTC

0894: "Progeny"

Postby Avaviel » Wed May 04, 2011 5:52 am UTC

Image

alt text: I tell my children 'it's not whether you win or lose, it's how you play the game.' I'm trying to take the edge off their competitive drive to ensure that I can always beat them.




Bean. I'm thinking of Bean's last battles at Battle School that he lost. But, eh, it doesn't exactly match up with this, does it? Unless you think of the kids as the computers. (Wait, Jane and her money managing skills!)
Also: Wow, I even had time to change my password with the forums for posting.
Last edited by Avaviel on Wed May 04, 2011 6:00 am UTC, edited 4 times in total.

Brooks Hatlen
Posts: 24
Joined: Tue Feb 15, 2011 5:47 am UTC

Re: 0893: Progeny

Postby Brooks Hatlen » Wed May 04, 2011 5:54 am UTC

We'll always have more creativity and general eccentricity than the machines. And feelings. Even if they learn to reproduce they won't enjoy it. :D

User avatar
SW15243
Posts: 75
Joined: Fri Jan 30, 2009 5:56 am UTC
Location: The Frozen White North
Contact:

Re: 0893: Progeny

Postby SW15243 » Wed May 04, 2011 5:57 am UTC

Not really funny, kinda made me think a bit though.

Incidentally it's comic number 894, not 893

hujackus
Posts: 79
Joined: Sat Aug 22, 2009 4:30 am UTC

Re: 0893: Progeny

Postby hujackus » Wed May 04, 2011 5:57 am UTC

Is this about Watson, or is there another new computer intelligence?

azza-bazoo
Posts: 6
Joined: Tue Dec 12, 2006 2:26 am UTC
Location: Perth, Australia
Contact:

Re: 0893: Progeny

Postby azza-bazoo » Wed May 04, 2011 6:01 am UTC

I'm with you hujackus -- I think he's talking about IBM Watson winning Jeopardy. Or have the robot invaders beaten us at another game while I wasn't paying attention?

Also, I worry what'll happen if Randall ever actually has kids. SQL-injected names, diabetes, aversion therapy, and now this, destroying their sense of competition!

Brickman
Posts: 12
Joined: Mon May 02, 2011 7:25 pm UTC

Re: 0894: Progeny

Postby Brickman » Wed May 04, 2011 6:12 am UTC

Thanks to so much sci-fi that treats robots as shiny people, it's easy to forget what computers are: Tools. An extension of ourselves, of our reach. To compare my abilities to a computer's is like comparing my abilities to a hammer. Anything that a computer does, ever, is really something a human has done. That's true when they find the ten billionth prime number and it's true when they brute force through but-this-used-to-be-enough-bits encryption and it's true when they run into the same wall twenty times in a row. In fact it's especially true in that last case (and this entire post was totally not just an excuse to use that example).

hujackus
Posts: 79
Joined: Sat Aug 22, 2009 4:30 am UTC

Re: 0894: Progeny

Postby hujackus » Wed May 04, 2011 6:15 am UTC

I remember teaching my computer how to play bejeweled back in 2006. I still hold second place for the timed version.
http://zone.msn.com/en/bejeweled/articl ... scores.htm

philip1201
Posts: 201
Joined: Tue Nov 03, 2009 6:16 am UTC

Re: 0893: Progeny

Postby philip1201 » Wed May 04, 2011 6:42 am UTC

Brooks Hatlen wrote:We'll always have more creativity and general eccentricity than the machines. And feelings. Even if they learn to reproduce they won't enjoy it. :D

Creativity and eccentricity is simply the randomness in path-finding algorithms.
Feelings are the basic (i.e. above the level of individual programs) dispositions which determine the relative weighing of goals in decision making processes, and enjoyment is one of those feelings.
There is no reason an organic computer like ourselves can do things a silicon computer can't.
This discussion is the matter of the [urlhttp://en.wikipedia.org/wiki/Philosophical_zombie]Philosophical Zombie[/url]. A being programmed to pass every Turing test imaginable. Computers will, for a fact, be able to do that one day. They will be able of expressing outrage over you doubting their abilities to have feelings, to show all the signs of sexual enjoyment, creative thought, eccentric behavior, love, hate, fear, regret, sadness. They'll be able to express all those feelings as well as any human. Better, probably. You'll be able to have philosophical discussions with them about whether they feel, and they can answer with "I can't prove it to you, but I know I feel". They can be programmed to do all that. Is that any different from actually feeling those things, from actually believing them?

Brickman wrote:Anything that a computer does, ever, is really something a human has done. That's true when they find the ten billionth prime number and it's true when they brute force through but-this-used-to-be-enough-bits encryption and it's true when they run into the same wall twenty times in a row. In fact it's especially true in that last case (and this entire post was totally not just an excuse to use that example).

They can calculate things no man has ever calculated (No human has ever known the 2339846892670489628904356807245680425th digit of pi, but computers CAN calculate it). They can come with solutions no man has ever come up with. It's evolutionary, my dear Brickman.

User avatar
glasnt
Posts: 539
Joined: Fri Jan 25, 2008 5:18 am UTC
Location: SQUEE!

Re: 0894: Progeny

Postby glasnt » Wed May 04, 2011 6:43 am UTC

That's what she might have uttered.

http://mynorthwest.com/?nid=11&sid=474343

hujackus
Posts: 79
Joined: Sat Aug 22, 2009 4:30 am UTC

Re: 0894: Progeny

Postby hujackus » Wed May 04, 2011 6:43 am UTC

@Brickman Unlike hammers, computers and the programs they run have the potential to break the tool barrier. Most programs currently written can only be used as tools, but there are programs out there that act autonomously without the need for human feedback. More progress is being made towards self-learning systems every year, and there is plenty of room for improvement.

As a side note, just because something is used as tool doesn't mean it is incapable of what we consider intelligence.There are several examples that include birds, dogs, horses and other domesticated animals that are intelligent creatures that have been bred as tools for hunting, gathering, and transportation. In more extreme cases, humans can even be used as tools although this is becoming quite rare.

Farabor
Posts: 157
Joined: Wed Apr 06, 2011 9:46 am UTC

Re: 0893: Progeny

Postby Farabor » Wed May 04, 2011 6:52 am UTC

Not to be pedantic, but a human can easily compute said digit of pi also, at least if we're talking in hexadecimal or binary. Not sure if anyone's come up with a simple formula for base 10 yet (But a simple google search for "calculate digits of pi" will show plenty of simple ways to compute any digit of pi you want, with no computers involved. In base 16.

Okay, I lied, I did post this just to be pedantic :)

philip1201 wrote:They can calculate things no man has ever calculated (No human has ever known the 2339846892670489628904356807245680425th digit of pi, but computers CAN calculate it). They can come with solutions no man has ever come up with. It's evolutionary, my dear Brickman.

Brickman
Posts: 12
Joined: Mon May 02, 2011 7:25 pm UTC

Re: 0894: Progeny

Postby Brickman » Wed May 04, 2011 7:18 am UTC

Oh really? Humans can't calculate/solve/derive/etc <thing X that computers can do>? Sure we can! Here's the steps it takes:
1) Convince a few other humans to give you funding, or be able to fund this use of your and possibly others' time yourself.
2) Build a powerful computer.
3) Write a program that does task X.
4) Run the program.

Tah-dah! A human has just completed task X!

The machinery that is in place in the typical factory assembly line, at some point, probably came from an assembly line (at the least if it's made of steel it was probably processed using more machinery). A blacksmith who wanted to make a good hammer would generally need another hammer to make it with (I'm guessing. Slight factual inaccuracies in my analogies don't invalidate the point). Tools which can be or are used to make more tools, including more tools of the same class, are not new and not exclusive to AI.

As for slipping out of our control, a chemist experimenting with chemicals he knows little about might produce a mixture which blows up his lab or is poisonous to breath. He also might produce something incredibly useful that wasn't even what he was trying for. He also might just produce nothing special but increase his knowledge a bit. Computers are not a new class of thing for being able to do things and/or backfire in ways their creators didn't expect, although I'll admit they're by far the strongest example of doing non-trivial non-destructive non-detrimental things the creator didn't expect.


All that said, however:
@hujackus: I gotta cede you some points comparing them to domesticated animals. A lot of points in fact. I was not prepared for that response.
Last edited by Brickman on Wed May 04, 2011 7:20 am UTC, edited 1 time in total.

Turning_Green
Posts: 2
Joined: Thu Mar 17, 2011 12:28 pm UTC

Re: 0894: Progeny

Postby Turning_Green » Wed May 04, 2011 7:19 am UTC

Man can create a machine. Can machine create a man?

User avatar
libra
Posts: 151
Joined: Mon Oct 15, 2007 8:50 am UTC
Location: UK
Contact:

Re: 0894: Progeny

Postby libra » Wed May 04, 2011 7:58 am UTC

Humans can teach other humans. Humans can learn from machines.

But humans cannot "teach" a computer anything: only program it to adapt its algorithms until it can model something whose behaviour matches, and then exceeds, the expectations of human observers.

Well-programmed chess software does no more "play" chess than a human can run a spreadsheet program in his head.

No matter how many times the computer software can set up a winning condition in a game to favour itself over its opponent, whether at chess, Jeopardy or poker, the computer itself does not "play" these games, "outsmart" the humans or do anything that it does not have a software routine for.

It is just a tool, used to run programs.

It is the humans who are playing the games, winning and losing. The only ones who derive any meaning from the activity of playing against the computer are the humans: the computer just runs the program in its drive. That's all.

User avatar
Plasma Mongoose
Posts: 213
Joined: Tue Feb 01, 2011 1:09 am UTC
Contact:

Re: 0893: Progeny

Postby Plasma Mongoose » Wed May 04, 2011 8:00 am UTC

azza-bazoo wrote:I worry what'll happen IF Randall ever actually has kids. SQL-injected names, diabetes, aversion therapy, and now this, destroying their sense of competition!


IF being the operative word here.

(I hope Randall isn't too insulted by this)
A virus walks into a bar, the bartender says "We don't serve viruses in here".
The virus replaces the bartender and says "Now we do!"

User avatar
BlitzGirl
Posts: 9119
Joined: Mon Sep 20, 2010 11:48 am UTC
Location: Out of the basement for Yip 6! Schizoblitz: 115/2672 NP
Contact:

Re: 0894: Progeny

Postby BlitzGirl » Wed May 04, 2011 8:04 am UTC

I'm usually not so critical of xkcd, but this one was pretty "meh" for me. It's another "talking heads" comic: entirely text-based with the stick figures doing nothing. It would work exactly the same written out like this:

Person 1: Wow - Researchers taught a computer to beat the world's best humans at yet another task. Does our species have anything left to be proud of?
Person 2: Well, it sounds like we're pretty awesome at teaching.
Person 1: Huh? What good is that?


I'm not entirely sure why this bothers me so much since there are other comics that do this. Maybe it's because with those, there's emotion expressed using the faces or postures of the characters having the discussion. And I know Randall has pulled off emotions in stick figures before, but this one just doesn't manage it. This would feel entirely different if (for instance) it was a Calvin and Hobbes discussion, plus expressions, doing something visually interesting like riding a wagon down a ridiculous hill.
Knight Temporal of the One True Comic
BlitzGirl the Pink, Mopey Molpy Mome
Spoiler:
Image
Image
Image<Profile
~.Image~.FAQ->Image

FourTael
Posts: 55
Joined: Wed Apr 14, 2010 3:36 pm UTC

Re: 0894: Progeny

Postby FourTael » Wed May 04, 2011 9:03 am UTC

Poker

Lawton
Posts: 16
Joined: Fri Feb 25, 2011 7:57 am UTC

Re: 0894: Progeny

Postby Lawton » Wed May 04, 2011 9:31 am UTC

Ah, I see that one of the characters completely missed the point.

If you were to replaced the words "taught" and "teaching" with "program" and "programming" then this comic would be more accurate. Though it would probably also end with the obvious computer geek agreeing with the statement.

It seems to me, that computers are only capable of doing what their programmers are capable of doing, albeit much faster. (I could find any digit of pi given enough time) As such, a computer that could program would only be able to make computers as good as itself. A computer that could actually learn however, could be infinitely useful.
Last edited by Lawton on Wed May 04, 2011 9:40 am UTC, edited 1 time in total.

Faranya
Posts: 259
Joined: Tue Nov 18, 2008 3:10 am UTC

Re: 0894: Progeny

Postby Faranya » Wed May 04, 2011 9:31 am UTC

Turning_Green wrote:Man can create a machine. Can machine create a man?


Not yet.
Image

User avatar
Samik
Posts: 511
Joined: Mon May 02, 2011 11:14 am UTC

Re: 0893: Progeny

Postby Samik » Wed May 04, 2011 10:54 am UTC

philip1201 wrote:There is no reason an organic computer like ourselves can do things a silicon computer can't.



Of course you're correct. There's no reason, in principle, why the data storage and transmission mechanisms of the human brain can't be recreated with silicon and electricity (or vastly improved upon). I find that to be such an obvious assertion as to be entirely uninteresting.

The relevant question is not if it is possible, but instead: "Are we really anywhere near close to this now, or are Watson and Deep Blue nothing more than parlor tricks?"

I'm personally of the opinion that there is nothing on the planet now that even begins to resemble any sort of notable AI. Throwing ever more computational power and data storage capability at the problem is only going to take us so far - we will not begin to truly enter the age of AI until our skill at programming catches up to our skill at construction.


For those arguing that a human being cannot do the things a computer can do, I would suggest that it's less a matter of ability than it is simply a matter of time constraints (Randall seems to agree with me.). After all, we must program in the rules of any computational operation we want the computer to perform, which by default means we must understand those rules ourselves.



EDIT: My goodness. I didn't even notice it was you I was responding to, Philip, until after the fact. Your posts just seem to gravitationally attract my attention. Nothing personal.

jozwa
Posts: 144
Joined: Fri Sep 11, 2009 3:16 pm UTC
Location: Finland

Re: 0894: Progeny

Postby jozwa » Wed May 04, 2011 11:55 am UTC

Reminds me of the Skynet fuzz a couple of weeks ago. http://www.cbc.ca/news/technology/story ... ttack.html

This quote's interesting:
"Once a system is deployed, people often stop thinking of it as artificial intelligence. In the 1950s, for example, programming a computer using Fortran seemed like artificial intelligence. So, lots of the [currently] deployed systems use technology that was once considered artificial intelligence."

User avatar
Samik
Posts: 511
Joined: Mon May 02, 2011 11:14 am UTC

Re: 0893: Progeny

Postby Samik » Wed May 04, 2011 12:26 pm UTC

philip1201 wrote:This discussion is the matter of the Philosophical Zombie. A being programmed to pass every Turing test imaginable.



So I take it you're not a behaviorist?

BeagleFury
Posts: 11
Joined: Wed Feb 03, 2010 4:06 pm UTC

Re: 0893: Progeny

Postby BeagleFury » Wed May 04, 2011 12:49 pm UTC

Samik wrote:For those arguing that a human being cannot do the things a computer can do, I would suggest that it's less a matter of ability than it is simply a matter of time constraints. [..]. After all, we must program in the rules of any computational operation we want the computer to perform, which by default means we must understand those rules ourselves.


If it just a matter of "time", it's easy to push that off to the computer too. Simply write the (almost trivial) program to write all possible programs and enumerate thru them. At some point, it will generate the program that simulates a human thought process / observable behavior / comments and philisophical position.

And responding to some objection to using the word "teach" vs "program", I'd argue that is semantics. I see little difference between "teaching" and "programming"; both are an organized process to transfer (meta)-information from one human/processor to another human/processor.

User avatar
Samik
Posts: 511
Joined: Mon May 02, 2011 11:14 am UTC

Re: 0893: Progeny

Postby Samik » Wed May 04, 2011 1:17 pm UTC

BeagleFury wrote:If it just a matter of "time", it's easy to push that off to the computer too. Simply write the (almost trivial) program to write all possible programs and enumerate thru them. At some point, it will generate the program that simulates a human thought process / observable behavior / comments and philisophical position.


Heh, that is certainly true. However, bear in mind that all I was arguing is that a human mind, as presently constructed, could do the same thing, if given enough time. As could my 1GHz Pentium III A22m Thinkpad, if allowed to work at it's desired pace. I was not arguing against the possibility of robust artificial intelligence.


BeagleFury wrote:And responding to some objection to using the word "teach" vs "program", I'd argue that is semantics. I see little difference between "teaching" and "programming"; both are an organized process to transfer (meta)-information from one human/processor to another human/processor.


I wouldn't even think of making such an objection. I could not possibly agree with you more. As I've already said, "There's no reason, in principle, why the data storage and transmission mechanisms of the human brain can't be recreated with silicon and electricity (or vastly improved upon)."

User avatar
Samik
Posts: 511
Joined: Mon May 02, 2011 11:14 am UTC

Re: 0893: Progeny

Postby Samik » Wed May 04, 2011 1:37 pm UTC

BeagleFury wrote:If it just a matter of "time", it's easy to push that off to the computer too. Simply write the (almost trivial) program to write all possible programs and enumerate thru them. At some point, it will generate the program that simulates a human thought process / observable behavior / comments and philisophical position.



Ah, ok, I see what you're saying. Using the process you describe, the computer would be certain to eventually generate, and then utilize, programs that perform tasks that we do not have a current understanding of. I.E. it would gain knowledge and capabilities that we do not have and did not program into it. There wouldn't even have to be any real selection mechanism - if the computer runs each program in series, it will be guaranteed to eventually run a simulation, such as you describe, that gives it the disposition to stop the process and stay as it is - a functionally intelligent entity. Thus, the combination of massive computational power, arbitrarily large data storage capacity, and trivially simple programs could yield robust AI.

Of course, such a system has every possibility of generating Skynet before R. Daneel Olivaw.


EDIT: As well as an infinite number of simulated thought processes that have the disposition to terminate the procedure, but not much else to them of any substance (the set of simulations which have both the disposition to terminate the process and some level of robust intelligence being an infinitely small subset of the set up simulations which just have the disposition to terminate the process). So now you need to program in a selection mechanism to avoid these, which makes the original program slightly less trivially simple, but moves it closer to having something that can be meaningfully called a learning mechanism.



EDIT2: I should clarify why I am requiring the use of a termination mechanism, and not just letting the computer run each program through to its conclusion.

If the programs are written and executed in series, in a truly random fashion, it seems likely that an extraordinarily simple loop will be generated long before anything like a robust artificial intelligence, thus stagnating the process completely.

If the programs are written and executed in parallel, even infinitely so, then it must be considered that the set of programs that can be meaningfully called intelligent, while infinite, is still an infinitely small subset of all possible programs.* Thus, the computational power required to be guaranteed to achieve the objective we want will also be infinite.

*(Ok, I'm less than sure about this one. For every program providing robust intellect, there are an infinite number of programs that do not, but the same can be said in reverse. What proportion of all possible programs simulate a robust intellect? My intuition tells me it would be microscopically small, but I can't even think about how I would prove that.)


The point is that it seems to me that you are ultimately going to have to program in so many guiding constraints to your random program generator, that it will cease being meaningfully random, and instead have what amounts to a very robust learning mechanism, provided by the programmer. Which brings us right back to where we started.

User avatar
Karilyn
Posts: 282
Joined: Thu Oct 15, 2009 6:09 pm UTC

Re: 0893: Progeny

Postby Karilyn » Wed May 04, 2011 2:32 pm UTC

philip1201 wrote:
Brooks Hatlen wrote:We'll always have more creativity and general eccentricity than the machines. And feelings. Even if they learn to reproduce they won't enjoy it. :D

Creativity and eccentricity is simply the randomness in path-finding algorithms.
Feelings are the basic (i.e. above the level of individual programs) dispositions which determine the relative weighing of goals in decision making processes, and enjoyment is one of those feelings.
There is no reason an organic computer like ourselves can do things a silicon computer can't.
This discussion is the matter of the [urlhttp://en.wikipedia.org/wiki/Philosophical_zombie]Philosophical Zombie[/url]. A being programmed to pass every Turing test imaginable. Computers will, for a fact, be able to do that one day. They will be able of expressing outrage over you doubting their abilities to have feelings, to show all the signs of sexual enjoyment, creative thought, eccentric behavior, love, hate, fear, regret, sadness. They'll be able to express all those feelings as well as any human. Better, probably. You'll be able to have philosophical discussions with them about whether they feel, and they can answer with "I can't prove it to you, but I know I feel". They can be programmed to do all that. Is that any different from actually feeling those things, from actually believing them?

As far as I'm concerned, there isn't. At the end of the day, even the human brain runs on it's own version of zeros and ones. Just little bleeps of electrical charge, and no electrical charge.

libra wrote:Well-programmed chess software does no more "play" chess than a human can run a spreadsheet program in his head.

No matter how many times the computer software can set up a winning condition in a game to favour itself over its opponent, whether at chess, Jeopardy or poker, the computer itself does not "play" these games, "outsmart" the humans or do anything that it does not have a software routine for.

It is just a tool, used to run programs.

It is the humans who are playing the games, winning and losing. The only ones who derive any meaning from the activity of playing against the computer are the humans: the computer just runs the program in its drive. That's all.

All it really means is that humans are really good at convincing themselves that they are doing something special. That we have souls, and are not simply electrical blips.
Gelsamel wrote:If you punch him in the face repeatedly then it's science.

User avatar
Introbulus
Posts: 84
Joined: Fri Aug 21, 2009 4:09 am UTC
Location: New York
Contact:

Re: 0894: Progeny

Postby Introbulus » Wed May 04, 2011 2:34 pm UTC

My thoughts on the comic:

1: More proof that teachers are undervalued? Nah, probably just proof that Randall thinks teachers are undervalued. Still, for as important the task is that they perform, why are they not treated better?

Bah...but I suppose they at least get the biggest vacations ever. Except professors who work over Summers.

2: This occurred to me the very instant I started to seriously consider Watson the Jeapordy-playing computer. So yes, Watson is a marvel and a wizard at Jeapordy, but it literally took him YEARS to get to that point, and that's NOT counting the time it took to build him (which I wouldn't), or the fact that he is literally incapable of doing anything else without a LOT Of re-programming and re-teaching.

That's basically the equivalent of a man who spends his entire life learning to play Jeapordy, who then forgoes everything else in the entire world to prove he is the absolute best, then realizes afterward that his life is essentially over. That he has done exactly what he set out to do, and there is nothing left for him that he is mentally or physically capable of doing, because all he knows is how to play Jeapordy.

I suppose Watson will be doing what that man would then be doing at that point - re-learning everything. Integrating himself into society to become a useful, productive member. But, and here's the important thing to remember, he can only ever do one thing.

I'm not touting this as human superiority. It's actually kind of sad that Watson has to dedicate his entire existence to one function forever. If he ever does gain sentience, it will be a very, very sad and resigned sentience of perpetual effort towards a goal for which there is no reward.

3: After typing all that, I rubbed the front of my computer case with the back of my hand. I don't think she can feel it, but I hope she feels better knowing I appreciate her.
If you can read this, you are wasting your time.

User avatar
Samik
Posts: 511
Joined: Mon May 02, 2011 11:14 am UTC

Re: 0894: Progeny

Postby Samik » Wed May 04, 2011 2:45 pm UTC

Introbulus wrote:So yes, Watson is a marvel and a wizard at Jeapordy, but it literally took him YEARS to get to that point, and that's NOT counting the time it took to build him (which I wouldn't), or the fact that he is literally incapable of doing anything else without a LOT Of re-programming and re-teaching.


I have half a mind to write a short story about basically this, in response to near-future Singularitarians.

Plot: Some genius programmer finally designs an AI capable of learning / self-improvement (modifying it's own code in beneficial ways). He executes the program, fully expecting runaway growth, only to spend the next few decades trying to motivate a lazy program who just wants to hang around and have fun with the other programs, rather than doing its schoolwork. In the end, he spends just as much time with it as he would have a kid, and winds up with roughly the same result.

Brickman
Posts: 12
Joined: Mon May 02, 2011 7:25 pm UTC

Re: 0894: Progeny

Postby Brickman » Wed May 04, 2011 3:04 pm UTC

This stuff about computers generating an infinite series of programs... two problems. First, the set of possible programs is an infinite set, and not just infinite but it's a large infinity (the same way 2^infinity is bigger than 1+infinity). It's certainly a larger infinity than an infinite series of things back-to-back, though it doesn't really have to test every possibility since there'll be an infinite number of variations on the goal too. So I guess that's ok, sorta.

Second, and more importantly, it's meaningless if you just run all the outputted programs at once--sure, one will be correct but one being correct in a billion isn't very meaningful since it's drowned out. You need to select the right one... but how can you write a computer to select for "robust intelligence" if the computer's not robustly intelligent? You can't; if you try to test for a goal more complex than what you have you're almost guaranteed to instead get fakers. The only thing that can separate out the winning computer is a human, manually testing every one (or every one that passes some preliminary automated tests). And with the goal you stated that human will probably be performing, you guessed, a classic Turing test.

And for similar reasons, it'd be logically impossible for us to test for something intellectually superior to us in the same way that a human is to a dog. Better at some task, yes, but not better at being intelligent. The only conceivable way I can imagine to do that is to just select for something slightly smarter than yourself and have it test for something slightly smarter than itself, in a chain... assuming that's even possible, which it may not be.

philip1201
Posts: 201
Joined: Tue Nov 03, 2009 6:16 am UTC

Re: 0894: Progeny

Postby philip1201 » Wed May 04, 2011 3:06 pm UTC

Brickman wrote:Oh really? Humans can't calculate/solve/derive/etc <thing X that computers can do>? Sure we can! Here's the steps it takes:
1) Convince a few other humans to give you funding, or be able to fund this use of your and possibly others' time yourself.
2) Build a powerful computer.
3) Write a program that does task X.
4) Run the program.

Tah-dah! A human has just completed task X!


That would make the word "can" completely meaningless. Can a human collapse a star? Sure we can, just invent matter-to-energy technology, attach an energy drive to the Earth and pull a black hole into the star.
Computers can do things humans can't do without the help of computers, especially if you set a time limit.
A computer, unlike other artifacts, can, in the future, design more efficient computers than itself.

Samik wrote:I'm personally of the opinion that there is nothing on the planet now that even begins to resemble any sort of notable AI. Throwing ever more computational power and data storage capability at the problem is only going to take us so far - we will not begin to truly enter the age of AI until our skill at programming catches up to our skill at construction.

*agrees*
So I take it you're not a behaviorist?

I think a philosophical zombie, for all intents and purposes, is identical to a human. I'm not familiar enough with 19th century philosophy to answer your statement, though that pretty much answers your question with a "no".

User avatar
Samik
Posts: 511
Joined: Mon May 02, 2011 11:14 am UTC

Re: 0894: Progeny

Postby Samik » Wed May 04, 2011 3:22 pm UTC

Brickman wrote:Second, and more importantly, it's meaningless if you just run all the outputted programs at once--sure, one will be correct but one being correct in a billion isn't very meaningful since it's drowned out.


Well, yes and no. Remember that Beagle was just responding to my assertion that, in principle, a program can only ever do things that we have programmed it to do. S/he wasn't suggesting his/her random programs generator as functional to any real degree.

You need to select the right one... but how can you write a computer to select for "robust intelligence" if the computer's not robustly intelligent? You can't; if you try to test for a goal more complex than what you have you're almost guaranteed to instead get fakers.


Exactly. As I said, I expect that by the time you've programmed in enough intelligent selection criteria to allow you to end up with something useful, you've basically done a huge amount of the required work already anyway.


The discussion of the random program generator was only meant (I hope) to make an in principle argument.

User avatar
Samik
Posts: 511
Joined: Mon May 02, 2011 11:14 am UTC

Re: 0894: Progeny

Postby Samik » Wed May 04, 2011 3:26 pm UTC

philip1201 wrote:I think a philosophical zombie, for all intents and purposes, is identical to a human. I'm not familiar enough with 19th century philosophy to answer your statement, though that pretty much answers your question with a "no".


Then you are at least sympathetic to behaviorism. A behaviorist would consider the concept of a philosophical zombie to be incoherent. If it walks like a duck, swims like a duck, quacks like a duck, and performs every variety of duck-like actions at reasonably duck-like frequencies, while eschewing non-duck-like behaviors, and successfully passing all possible duck-calibrated Turing tests, it's a duck.




On a related note, for the last three years I have gone to every Halloween party as a philosophical zombie.

No one ever gets it.

Brickman
Posts: 12
Joined: Mon May 02, 2011 7:25 pm UTC

Re: 0894: Progeny

Postby Brickman » Wed May 04, 2011 3:38 pm UTC

philip1201 wrote:
Brickman wrote:Oh really? Humans can't calculate/solve/derive/etc <thing X that computers can do>? Sure we can! Here's the steps it takes:
1) Convince a few other humans to give you funding, or be able to fund this use of your and possibly others' time yourself.
2) Build a powerful computer.
3) Write a program that does task X.
4) Run the program.

Tah-dah! A human has just completed task X!


That would make the word "can" completely meaningless. Can a human collapse a star? Sure we can, just invent matter-to-energy technology, attach an energy drive to the Earth and pull a black hole into the star.
Computers can do things humans can't do without the help of computers, especially if you set a time limit.
A computer, unlike other artifacts, can, in the future, design more efficient computers than itself.


By that logic, a saw can do things that a human can't do without a saw. I shudder to think what my house would look like if noone making it had enlisted the help of a saw. A saw needs a human holding it and applying force to it and deciding where to use it, but how's that really, fundamentally different from a computer that needs a human to explain to it in detail that'd make a toddler feel condescended the exact process of doing what you want? I guess I'll grant that my definition of what a human "can" do was a little too broad since it does include things we're yet to do even once, or verify that we'll ever be able to do. But remember that ultimately, there was a human who wanted to find out what the 500th prime number was and he and some friends settled down to work and eventually found the first 500 prime numbers.

@Samik: Isn't a random program generator just a computer programmed to do the human's job for him, ie write more programs? It won't make anything that we couldn't have unless given infinite time, and I think giving one side too much more time than the other is what I just got called out on above.

TheCycoONE
Posts: 13
Joined: Wed Nov 17, 2010 7:40 pm UTC

Re: 0894: Progeny

Postby TheCycoONE » Wed May 04, 2011 4:06 pm UTC

I always wondered if genetic programming techniques might not eventually replace human efforts at designing better AI - after all it can design better antenna's than people, design walking robot schematics, and write programs to beat other programs it writes at any simple game.

So far applications seem limited outside of engineering and toy problems though :(

User avatar
Introbulus
Posts: 84
Joined: Fri Aug 21, 2009 4:09 am UTC
Location: New York
Contact:

Re: 0894: Progeny

Postby Introbulus » Wed May 04, 2011 4:11 pm UTC

Samik wrote:
Introbulus wrote:So yes, Watson is a marvel and a wizard at Jeapordy, but it literally took him YEARS to get to that point, and that's NOT counting the time it took to build him (which I wouldn't), or the fact that he is literally incapable of doing anything else without a LOT Of re-programming and re-teaching.


I have half a mind to write a short story about basically this, in response to near-future Singularitarians.

Plot: Some genius programmer finally designs an AI capable of learning / self-improvement (modifying it's own code in beneficial ways). He executes the program, fully expecting runaway growth, only to spend the next few decades trying to motivate a lazy program who just wants to hang around and have fun with the other programs, rather than doing its schoolwork. In the end, he spends just as much time with it as he would have a kid, and winds up with roughly the same result.


I hope you realize that this story is essentially "Frankenstein 2012".

With the exception of the programmer here actually being a decent father to his artificial creation, rather than the total careless hack that Dr. Frankenstein was.
If you can read this, you are wasting your time.

User avatar
Samik
Posts: 511
Joined: Mon May 02, 2011 11:14 am UTC

Re: 0894: Progeny

Postby Samik » Wed May 04, 2011 4:14 pm UTC

Brickman wrote:@Samik: Isn't a random program generator just a computer programmed to do the human's job for him, ie write more programs? It won't make anything that we couldn't have unless given infinite time, and I think giving one side too much more time than the other is what I just got called out on above.


Er. I agree completely. The idea of the random program generator wasn't meant to show that a computer could do things a human couldn't. It was introduced by BeagleFury to counter my argument that a computer can't do anything that hadn't been directly programmed into it by a human. Different issue.


The first issue is one that I think most of us here agree on - give any computer, biological or otherwise, enough time, and ability to work at its own pace, and it can eventually run every possible program (again, suggested here).

The second issue tries to address the minimum complexity for what many consider an integral feature of any prospective artificial intelligence - a learning mechanism.

User avatar
Samik
Posts: 511
Joined: Mon May 02, 2011 11:14 am UTC

Re: 0894: Progeny

Postby Samik » Wed May 04, 2011 4:17 pm UTC

Introbulus wrote:I hope you realize that this story is essentially "Frankenstein 2012".


Awesome.

It's been ages since I read that book, but I still should have made the connection.


That said, now I have an even stronger desire to write it.

MagentaHawk
Posts: 8
Joined: Wed Feb 09, 2011 3:52 pm UTC

Re: 0894: Progeny

Postby MagentaHawk » Wed May 04, 2011 5:16 pm UTC

So I know I am going to be "that guy", but the reason a computer will never match up to us is because we aren't simply computing power. I love science and mathematics, but that's not all of it. Kind of like when you read Zen and the art of motorcycle maintenance and he deals with the problem he found in the idea, where do we get these hypotheses? The scientific method works so completely well once you have one, but how did we choose one? Off a hunch? Then what is this said hunch and how do we get it? How do we program computers to test hunches? I'm not sure if we can or can't, but the inquiry into inquiries and lines of thinking is interesting.

Because of my religion, and my logical deduction that supports this claim, I believe that we can never create a computer like a human being since computers are things to be acted upon. They do what they do from a predetermined set of responses to stimuli in so that the stimuli really determine the computer's actions. The reason we can "randomly" choose a hypothesis to test or other examples is that we are things that act. When you choose an action that occurred you can ask: why did that happen. Get a response and then ask why did that happen. You can continue this path with anything really, (why wind blows, why that tree fell, why it is warm today) and eventually you will get down to one of two things: Either a human decided or God decided. Well if you believe he made the universe etc. I would pose the claim that that is because we are the only true source of action and everything else just reacts to us. We aren't like science in the idea that if you were somehow able to put someone in the exact same circumstances twice a different thing might happen. I didn't like this idea at first, but it makes more sense to me. Taking this further, if we are the only perpetrators of true action then a machine or computer could never do what we do in that it would only be reacting. It would be as many people believe we are: simply reacting to chemical reactions by firing off another chemical. It would simply be a product of electrical and chemical reactions. It would never be able to do what any human can do; that is, if you believe that we are things to act and not things to be acted upon.

Twigshusband
Posts: 51
Joined: Thu Feb 18, 2010 2:00 pm UTC

Re: 0894: Progeny

Postby Twigshusband » Wed May 04, 2011 5:20 pm UTC

World of Warcraft researchers?

charonme
Posts: 141
Joined: Sun May 18, 2008 11:18 am UTC

Re: 0894: Progeny

Postby charonme » Wed May 04, 2011 5:42 pm UTC

(some) humans are still better than IBM watson at jeopardy, because they achieve approximately the same results with just a fraction of:
- the raw processing power (a single CPU is faster than a human at basic computing tasks)
- storage space (eg. watson contains the whole wikipedia offline)
- power consumption (a human runs at roughly 100Watts)
- mass, volume, portability etc
- maintenance and assembling costs

Although it probably takes considerably longer time to build/raise/train/teach a human for the task.


Return to “Individual XKCD Comic Threads”

Who is online

Users browsing this forum: dtobias and 107 guests