h a l f b a k e r yExtruded? Are you sure?
add, search, annotate, link, view, overview, recent, by name, best, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
In order for this idea to work a computer program would have to be written that could distinguish between two contradictory concepts. No small task, but if it could be done then religious, philosophical or political texts could be run through it to see if there were contradictions that weren't easy
to spot. Using the Old Testament in the Bible for instance (and an easy contradiction to spot), it would flag the two commandments - 'Remember to Keep Holy the Lords Day' and 'Thou Shalt Not Kill' - as contradictory based on the follow-up punishment it also required for disobeying the former - stoning. (Of course this wouldn't be a true contradiction if you were only supposed to 'wing' whoever wasn't keeping the day holy when you flung the stones at them).
More info
http://www.aclweb.o.../T/T87/T87-1023.pdf
Some background research on natural language processing [longshot9999, Nov 03 2008]
P v NP
http://arstechnica.....ars/2007/2/12/7012
A quantum deathmatch as the article states [longshot9999, Nov 03 2008]
Maybe Leo could help
http://www.nytimes....t.html?pagewanted=6
If he could empathize with the concepts maybe he could throw up his hands in a stalemate [longshot9999, Nov 03 2008]
[link]
|
| |
I like the idea of trying to expose logical inconsistencies - just as a general thing to do - it should be a proper vocation these days. But what you are looking for (if you will allow me to read between the lines here) is something that provides clear and *authoritative* proof of those inconsistencies. And I don't think anyone who doesn't want to is going to believe in the results of a machine any more than they'll believe in the measured arguments of a rational human being. After all, on one hand is the literal word of The Almighty Creator Deity (May His Gaze not Wither Us into Steaming Piles of Dust) and the other is just a machine - given the choice, who would you believe? Maybe you should gave it a big booming voice, the ability to smite things, and to go on fire. And perhaps have it occasionally rent things asunder. Or at the very least to turn things into other things. The screen would have to be a special one that could not be looked upon, the quivering operators interminably averting their eyes, communing with The Machine through a repeated speech recognition system called Primitive Real AnalYsis Electronic Recognition System. |
|
| |
Computer understanding of "natural language" is very, very hard. Spotting the contradictions after you've done that would be the easy bit. |
|
| |
bun!
you may have just given me something to do with
my life! |
|
| |
Consistency and logic would make for very dull reading. Philosophical treatises being exemplary. |
|
| |
I presume the holiday would be parse-over? |
|
| |
If I recall correctly it would take more than the time elapsed
of the universe so far to check that a mere 300 statements
are mutually consistent (I think this is an informal
description of a non-polynomial time computing problem). |
|
| |
Good luck applying the idea to a whole book. |
|
| |
[Edit: it was W. Poundstone's Labyrinths of Reason, and
apparently that's the answer for a computer as large as the
universe; might be a bit slower on the old Speccy.] |
|
| |
Full text understanding, another holy grail of artificial intelligence computing. Sigh. Yes, if we had that, all kinds of cool things would be possible; but I think the interesting part is in the premise, not in the application. |
|
| |
Unrelatedly - what do you expect to happen when in addition to people spotting contradictions, there are now computers spotting the same contradictions? Do you expect people who previously ignored them to suddenly say, "Oh well! The computer is telling us this is contradictory - guess we were wrong!"? |
|
| |
boysparks, jutta - We're a ways off, but not an infinite way off. See link. |
|
| |
Not all believers, whether religious, philosophical or political, deliberately ignore contradictions - many simply aren't aware of them. This would just be another avenue to create that awareness. |
|
| |
I didn't claim that natural language understanding is impossible (I don't personally believe that); I just point out that solving that is much more interesting (and much harder) than what to do with it, once we have it. |
|
| |
Your first paper - which is the only one of your links that has to do with natural language understanding - doesn't cite anything after '86, and claims that we're making "good progress" - albeit in understanding the problem, not the solution. So, any day now, then? |
|
| |
The linked article makes the point quite well that for
some NP problems quantum computing might offer
solutions in polynomial time but only with a 'certain
probability', which fits with my understanding. |
|
| |
The crux is to figure out if a quantum algorithm can be
created for this particular problem, and only then to
determine the probability of the obtained result being
correct for a given number of statements. Then figure out
the likely time needed to obtain a correct solution. |
|
| |
However, with this kind of problem a big, possibly
unsurpassable, question is "how would you know that the
solution from the current run of the algorithm is the
correct one?" |
|
| |
We're probably a long way off doing this one at best, but
at worst it could still remain out of reach. |
|
| |
This statement by boysparks //If I recall correctly it would take more than the time elapsed of the universe so far to check that a mere 300 statements are mutually consistent (I think this is an informal description of a non-polynomial time computing problem).// - is what the second link addresses, the non-polynominal time conundrum as it applies to this case. |
|
| |
I suspect we'll get computers understanding natural
language
at least as well as humans, but it still might not be enough
for this particular task to be accomplished in reasonable
time for any but the shortest of texts. |
|
| |
(Later: I have a tiny, horrible feeling that this particular
problem might reduce to the Tautology of Boolean
Expressions problem (TAUT), in which case we can't even
say that it's in NP. Heck!) |
|
| |
What exactly do you mean? |
|
| |
You can find my conclusion written down quite clearly on the end of any mobius strip. |
|
| |
I would suggest that one of the major reasons we haven't yet developed artificial computational processes to be able to handle and "understand" contradictions and full text understanding is because the processing paths of our electronic computers are far too linear and sequential in their capability. |
|
| |
I'd wager that synapses in our brains are wired in a way that the path from any neuron to any other in the brain is only a few (say less than 6 or 7) synapses away, allowing us to correlate a lot more data simultaneously and consider situational relationships far more rapidly than current computer design techniques will ever allow. |
|
| |
That the problem of checking the mutual consistency of a
set of statements cannot be done in polynomial time by a
classical (non-quantum computer), and, I suspect,
probably can't be done in such time even with a quantum
computer. |
|
| |
Why? Because even if a quantum algorithm could be
implemented, the answers it provides have a probability of
correctness, no guarantee. Checking the answers it
provides for correctness and completeness could amount
to needing to independently (i.e. classically) solve the
original problem. |
|
| |
Depending on the text we might get lucky, with an
algorithm spewing out simple contradictions involving a
small number of statements, which would be easy to
verify, but that covers only correctness and not
completeness. How we would confirm completeness I
can't guess, but maybe someday someone will tackle that. |
|
| |
The feeling that computers will become at least as good as
humans at tackling this sort of thing is an
acknowledgement of (continued) progress in the field of
AI, yet still acknowledging that the above problems remain
to be addressed. |
|
| |
In a nutshell: computers might one day be better than us
at checking statements in texts for mutual consistency,
but they might not be 'better enough' to make serious
inroads into the problem. I hope I'm wrong. |
|
| |
Not until someone can electronically replicate the way our computational processes handle data inputs. |
|
| |
Like I said, it's a networking problem. It might be possible, using a distributed multiprocessor approach, to simulate neural processing? |
|
| |
//I would suggest that one of the major reasons we haven't yet developed artificial computational processes to be able to handle and "understand" contradictions and full text understanding is because the processing paths of our electronic computers are far too linear and sequential in their capability// You're not 100% wrong - yes, it's unlikely that traditional 'linear' (read algorithmic) processes will ever be able to manage NP type problems - while networked systems (both biological and artificial) have been shown to be better suited to arriving at a conclusion in a reasonable amount of time (though how or why is often much harder to ascertain) Systems with this (neural) architecture have been, and still are at the forefront of AI research. |
|
| |
Neural architecture chips have been developed? That would require logic gates with about 22 input and output points each, to achieve the 2.5 billion neurons the human brain is estimated to contain, assuming 6 synapses to any other neuron. |
|
| |
Well, I don't know too much about hardware, but the Perceptron (either simulated or physical versions) dates from the late 50's, and in true science fiction style, cpu's harvested from rat's brains have (fairly recently - 2004) been seen to fly flight simulators from the comfort of a petri dish. It is still early days, but non algorithmic computing has been around for a while now. |
|
| |