Semantic web : Lost in Translation

November 8, 2003 6:59 AM

Clay Shirky smacks syllogism around. Nice criticism of the semantic web and the present (and increasing) hype of the "semantic web revolution". The most damning part of the essay is the part about languages and categories being deeply intertwined with worldview and with culture—if there's no good definition for the word "bachelor" (see), how can there be an encoding of "friend", "lover" (see article for the classic AI example of "John loves Mary") or anything else that isn't zipcode?
posted by zpousman (62 comments total)

Day by day, I think of things I'd like to say
And all the answers to the questions that are facing me

But sometimes I just get pissed
I'm never in time I'm never in the right line
And I never go to the places that I should stay.

I'm not quite right, my timing or my place
And aggravation, it should show up on my face...

Sometimes when I'm down, I think of you, my friend Ringo
Like I have so many times before all through my life, my friend Ringo.
posted by stavrosthewonderchicken at 7:35 AM on November 8, 2003

Cory Doctorow's essay on metacrap is well worth a read. Clay points to it.
posted by Nelson at 8:27 AM on November 8, 2003

not sure it matters if there's no *global* defiition of bachelor. everyone reading metafilter understands the word. that seems like a large enough scope to me. (from a practical, rather than a culturally imperialist, pov).
posted by andrew cooke at 8:49 AM on November 8, 2003

[after reading article]. yawn. says hype is hyped, but something is probably useful. well who'd have guessed?
posted by andrew cooke at 8:55 AM on November 8, 2003

Excellent article. (andrew: Yes, to those to whom all of it is obvious, it's all obvious. You are perhaps not considering that everyone is not just like you.)

The Semantic Web's philosophical argument—the world should make more sense than it does—is hard to argue with.

This describes the basic problem with many more intellectual systems than the Semantic Web.
posted by languagehat at 9:03 AM on November 8, 2003

Let a writer tell me he lives in Brooklyn and he has lost me immediately.
posted by Postroad at 9:33 AM on November 8, 2003

Hrm, according to my AI book, language does not encode information, rather it is used to communicate. Simple sentances like "Look!" or "Fuck!" can tell people alot, but obviously do not contain much information themselves.

Anyway, any attempt to get this to actualy work will quickly be hijacked by pornographic spammers.
posted by delmoi at 9:51 AM on November 8, 2003

Delmoi: What does it mean to communicate other than to pass information from one place to another? Sure, the examples you give aren't informative, but then how did the person at the other end get "a lot" out of the utterance/statement/message? 'Cause information was passed between the two. How this is done, I have no idea, and nor do many people (certainly not linguists, and definitely not AI researchers).
posted by zpousman at 10:06 AM on November 8, 2003

Semantic web.... seems to me that the idea has been around for a long time without anyone actually doing anything about it...

Is anyone actually trying to develop for this anymore?
posted by tiamat at 10:10 AM on November 8, 2003

I have one issue with his use of Freenet in an analogy to the semantic web, as another piece of coding that has underlying philosophy. I think this is maybe an easy analogy, the point is that the semantic web has underlying philosophy, that's the key, philosophy that the architects of which are possibly unaware. The Dewey Decimal system is a better example, Dewey probably wasn't meaning to exclude other cultures decades later, he just didn't have the foresight.

Freenet is philosophical, and they make no effort to hide why they do it, or their philosophy, in fact their political and philosophical views about freedom are what made them do it, Dewey didn't come up with his system to exclude other religions, and the creators of the idea of the semantic web didn't do so in order to force people into standard definitions of all things, they probably think that's the easy part as he says, their goal is simply organization of data that they seem to think is fairly easy to agree on.
posted by rhyax at 10:16 AM on November 8, 2003

I also disagree that "Look!" and "Fuck!" aren't informative, they're just shorthand that everyone understands, Look!="Look at what I'm looking at because I think you will find it of interest!" Fuck!="Something has gone unexpectedly wrong!" just because humans are good at communication, and on the surface communication seems to short to contain data, or too context-based to contain machine readable data doesn't mean no data is being communicated.
posted by rhyax at 10:20 AM on November 8, 2003

languagehat - who doesn't know that hype is, well, hype, and that normally if lots of people are excited by something then, finally, something will come out of it, but not as much as was promised by those hyping it? these are social laws known to anyone with an ounce of common sense cynicism. how unlike me do you have to be before any of that is news? i mean maybe it's shocking for a sea-slug, but there are restrictions on how far you can go - if you can't read english or don't use the net, then you won't be getting much from the article anyway...

more pointedly, it seems to me that if saying hype is hype, and that, in the end, something will happen anyway, makes you a pundit, then i could write a program that does punditry. hmm. i have a sinking suspicion that someone has already had that idea.

rhyax - the book is probably trying to restrict "information" to have a technical meaning related to statistics. the kind of thing where yes/no is one bit of information, which isn't much (and "look"/"fuck" don't contain much more), while a stream of random numbers is just bursting with the stuff. so delmoi will doubtless be fascinated by 1001000001110110101001000001110101011011110, even if the rest of us don't think it's much use.
posted by andrew cooke at 10:51 AM on November 8, 2003

Shouldn't there be a cost associated with metadata? I.e. to broadcast a certain amount of metadata, one must spend some hashcash or some other resource (bandwidth expense alone is not adequate. )
posted by sonofsamiam at 11:31 AM on November 8, 2003

sonofsamiam: Well, there is the time spent creating the data you are describing, for one thing. Anyway, I don't really think we are talking about a broadcast system.
posted by delmoi at 12:11 PM on November 8, 2003

YHBT. YHL. HAND.
posted by xian at 12:22 PM on November 8, 2003

I agreed with the article, but damn, some of the conclusions Shirky draws from his own syllogisms are wrong.

Consider the following assertions:

- Count Dracula is a Vampire
- Count Dracula lives in Transylvania
- Transylvania is a region of Romania
- Vampires are not real

You can draw only one non-clashing conclusion from such a set of assertions -- Romania isn't real. That's wrong, of course, but the wrongness is nowhere reflected in these statements. There is simply no way to cleanly separate fact from fiction, and this matters in surprising and subtle ways that relate to matters far more weighty than vampiric identity. Consider these assertions:

- US citizens are people
- The First Amendment covers the rights of US citizens
- Nike is protected by the First Amendment

You could conclude from this that Nike is a person, and of course you would be right.

Neither of these conclusions follows from the premises stated. To support Shirky's first conclusion, we would need something like "unreal creatures do not live in real countries" in the first. In the second, we need the addition of "only" to the proposition about the 1st Amendment.

Nitpicking, but it bugs me a lot. I hate it when something basically right loses credibility from getting the details wrong.
posted by i_am_joe's_spleen at 12:27 PM on November 8, 2003

delmoi: re: broadcast: Poor choice of words. I mean "communicate."
The point is that there's no incentive for people to be honest. It should be more costly to lie than to tell the truth in their metadata. The "trust networks" proposed to keep people from gaming the system assume trust is associative. (It ain't.)

For a system that so completely depends on trust, there seems to have been almost no consideration of mechanism design.
posted by sonofsamiam at 12:37 PM on November 8, 2003

No spleen - both of those are valid, you could plug them into formal truth tables and they'd come out just fine. Which is just his point - the semantic web is working on the assumption that semantic logic is bester than everything, and that's just not the case.
posted by kavasa at 1:18 PM on November 8, 2003

I also disagree that "Look!" and "Fuck!" aren't informative, they're just shorthand that everyone understands

I have no background in anything, what so ever, but I'm going with delmoi on this one. "Look" and "Fuck" only contain information within a certain set of contexts. When the contexts surrounding language are removed, so are the meanings of each term. For language to have meaning, it must derive that meaning from something beyond itself. One might argue that computer science or mathematics create meaning beyond contexts, but it could be argued back that those disciplines are not languages meant to communicate, but formal systems meant to describe interactions within themselves.
posted by elwoodwiles at 1:30 PM on November 8, 2003

kavasa: huh? joe's_spleen is right and you are just way wrong. And Clay Shirky, whose name I had somehow associated with intelligence, is very silly. He seems to think you can "conclude" anything you can't disprove.

Given that US citizens are people, the First Amendment covers the rights of US citizens, and Nike is protected by the First Amendment, I can assert irrefutably that bagels are made of hair. So what?
posted by nicwolff at 1:49 PM on November 8, 2003

Well, it seems to me that a system that can detect lies and make them more 'expensive' is, erm, pretty much imposible. If you could tell how well real data matched with metadata, then what would be the point of the metadata?
posted by delmoi at 1:50 PM on November 8, 2003

It's not impossible. There is already tons of natural metadata sitting around. That's why Google works. That's why Bayesian spam blockers work.

The point is that in an honest signalling system you don't have to tell how well the metadata matches. If you know that it's not worth someone's while to lie, then you can trust it.

There seems to be plenty incentive for spammers, etc. to try to game the semantic web as it is currently envisioned.
posted by sonofsamiam at 2:07 PM on November 8, 2003

- Count Dracula is a Vampire
- Count Dracula lives in Transylvania
- Transylvania is a region of Romania
- Vampires are not real

You can draw only one non-clashing conclusion from such a set of assertions -- Romania isn't real.

I think Shirky is trying to qualify his argument with the words "non-clashing conclusion." What other relevant conclusion could we come to within the terms of the formal system he describes? "Count Dracula is not a vampire" doesn't work, as it clashes with the first statement, etc. The only assertion using the terms of the system that is consistent with those terms is "Romania isn't real."

Given that US citizens are people, the First Amendment covers the rights of US citizens, and Nike is protected by the First Amendment, I can assert irrefutably that bagels are made of hair.

Well, no, you couldn't. Hairy bagels aren't part of the system we are trying to derive a conclusion from. Think of this as a though experiment: We can only come to a conclusion that follows the given statements. So:
•US citizens are people
•The first amendment protects US citizens
•Nike is protected by the first amendment
∴
•Nike is a person.

The logic is consistent, but incorrect, and this is the problem of a semantic web. The semantic web would suffer the same weakness of the current internet: You could ask it any question and it will tell you 1000 answers, most of which are wrong.
posted by elwoodwiles at 2:12 PM on November 8, 2003

Daisy
Daisy
give me your answer true..............
posted by clavdivs at 2:46 PM on November 8, 2003

OK, now elwood seems to think that "therefore" means "therefore it is possible that". You want to talk logic? Let's state Shirky's terms as logical propositions:

1. All US citizens are people
2. All US citizens are First Amendment-protected
3. All Nikes are First Amendment-protected

This set of terms says absolutely nothing about whether all or any Nikes are people; that's not a conclusion you can draw, and in fact the only conclusion you can draw from this system is that if any US citizens exist, then some people are protected by the First Amendment.

You and Shirky are both trying to claim something about the applicability of formal logic to the real world, and you may be right, but your argument would be more likely to be taken seriously if you demonstrated some understanding of logic.

(Oh, and as to vampires: the conclusions you could draw from the first system are "Count Dracula is not real", "Count Dracula lives in Romania", and "some things that are not real live in Romania". But as joe's_spleen pointed out, unless you also assert that "all places where not-real things live are not real", you cannot conclude that "Romania is not real".)
posted by nicwolff at 2:47 PM on November 8, 2003

Which is a more interesting point about semantic logic than Shirky thought to make. Because if you're trying to apply a logical system of propositions semantically, you're going to have to accomodate a statement like "Superman lives in Metropolis, and Batman lives in Gotham City, but the Fantastic Four live in New York City." So you can't assume that "all places where not-real things live are not real"; rather, you should assert "all places where real things live are real" and leave it at that if you want to keep it simple.

As your "semantic web" grows, though, you'll have to separate your propositions about real things living in real places from those about not-real things being described as living in real or not-real places, and more generally require that not-real things have only attributes that start with "is described as". In fact, I wonder if this can be done at all without meta-propositions like that.
posted by nicwolff at 3:09 PM on November 8, 2003

thanks, nic. The rest of you have to do Formal Logic 101 again, I'm afraid.

Understand, I do think Shirky is right. Formal logic comes apart because we can't formulate propositions with the rigour and clarity required. It's a shame his examples don't work well. (He would have been better off examining the ambiguity in the verb "lives".

Incidentally, we can deduce that Count Dracula lives in Romania, that both Romania and Transylvania have unreal inhabitants, and that there is at least one entity V for whom the statement "V is an unreal person living in Transylvania" is true, from the propositions stated. The point Shirky ought to be making is that there is very little utility in this.
posted by i_am_joe's_spleen at 3:20 PM on November 8, 2003

"The logic is consistent, but incorrect".

Elwooodwiles, the reasoning you show is actually inconsistent. The result is incorrect viewed from outside the system too, but that's a whole other kettle of faulty premises.

That's another point which Shirky doesn't make very well, which is that correct, internally consistent reasoning doesn't save us from polluted data.
posted by i_am_joe's_spleen at 3:28 PM on November 8, 2003

Nicwolff, I think we are crossing wires here. You don't seem to be following the rules of the thought experiment Shirky is using to draw out his point. A syllogism is a system where a conclusion is drawn from given statements. You seem to be looking at the set of terms without drawing any conclusions from them.
This set of terms says absolutely nothing about whether all or any Nikes are people; that's not a conclusion you can draw, and in fact the only conclusion you can draw from this system is that if any US citizens exist, then some people are protected by the First Amendment.
It would follow from (that's what I mean by "therefore" - "For that reason") the terms that Nike is a person. It doesn't say Nike is a person within the set, but that is the next logical step taken when analyzing the set. There are no "if's" involved here. To make this all work, one must assume the validity of the statements under discussion, otherwise it would be impossible to conclude anything. Your conclusion isn't a conclusion at all, but the pieces of data we've decided to use to create the set.

And about Vampires:
Your still not looking at the statements as a set. To say "Vampires are not real" conflicts with the first statement and is therefore not a relevant or coherent conclusion to draw from the given data. The only statement that uses all the data in the set without conflicting (contradicting) with any of the data is the correct conclusion of the system. "Romania does not exist."

One could argue the sets of data Shirky gives are not complex enough to illustrate his point, but they are logically consistent syllogisms.
posted by elwoodwiles at 3:30 PM on November 8, 2003

"your argument would be more likely to be taken seriously if you demonstrated some understanding of logic."

Yours would be more likely to be taken seriously if you weren't a ridiculous prick. Good job failing at polite social interaction even on the internet. I hope you're a prick in meatspace too, there are few people I have less respect for than those who are meek in the real world and jackasses online. Especially since you haven't actually produced any formal logic in support of your claims, despite the great quantity of blustering you've engaged in over the subject.

Additionally, I'd like to remark that at the very least, Shirky points out that conclusions like this are intuitive, which indicates problems ahoy if you ask me.

That said, it's pretty fucking clear that the following statements form a consistent set:
- US citizens are people
- The First Amendment covers the rights of US citizens
- Nike is protected by the First Amendment
- Nike is a person

It is possible for all these statements to be true simultaneously. Thus, under a system of sentential logic, it is a valid argument, and since all the semantic web is checking for is validity, then yes it's in a bad place. Your statement that it's not sound is true, but that hardly matters for Shirky's purposes, since the semantic web as he describes it as only checking for validity.

So, yeah, maybe if you've forgotten about everything but quantifier logic, it's you who needs to take Philosophy 101 again.
posted by kavasa at 3:32 PM on November 8, 2003

Gah, I was going to reword the second to last paragraph because soundness is less the issue than the difference between sentential/quantifier logic, but oh well.
posted by kavasa at 3:33 PM on November 8, 2003

Also, if you read Tim Berners-Lee's biography, you'll see that he's dreamed for years about a sort of connection engine that would discover new relationships between known data, and that before the web, he tried to build one, with little success. The web was a diversion for him. What you can see is that he's reinventing his old dream on the back of the web's success. Poor guy. In his own way, he's on a hiding to nothing just like Ted Nelson. (Hi, xian!).
posted by i_am_joe's_spleen at 3:34 PM on November 8, 2003

kavasa, there is nothing wrong with the conclusion that Nike is a person. The error arises when you equate "person at law" with "human being".

It is an error, and a big one, that would bite constructors of metadata on the arse good and hard. But there's nothing wrong with the premises, or the conclusion, as stated.

Conclusions like this are intuitive. Mate, when my computer develops intuition, I'm heading for the hills.
posted by i_am_joe's_spleen at 3:36 PM on November 8, 2003

i_am_joe's_spleen: Couldn't we just set it up as
•US citizens are human beings
•The First Amendment protects US citizens
•Nike is protected by the First Amendment
==============================
•Nike is a human being

I realize we're getting away from the point of the example, but the conversation is fun. Right?
posted by elwoodwiles at 3:46 PM on November 8, 2003

but that is the next logical step

No, it's not. That's why I said you need to say "only US citizens are protected". You're falling into a classic fallacy right there (fallacy of the undistributed middle, in fact).

Try this.

All ducks can swim.
This fish can swim.
Therefore, this fish is a duck.

That is essentially what you are arguing.

You're right that it's a seductively obvious conclusion, but it's not one you can arrive at in formal logic, and it's not one that a semantic web engine could arrive at either.

If we do say "only", then we can draw the conclusion that Nike is a person, and that is true in the real world. Shirky's actual point is that "person" has more than one meaning, but he puts it very poorly. (kavasa, my apologies: it is actually erroneous to conclude that Nike is a person at all; the confusion about the meaning of "person" is secondary).

And why can't unreal persons live somewhere? You have an implicit assumption about what "lives" means that clearly conflicts with my understanding of the term (and nicwolf's too). Another good illustration of the pitfalls of metadata, I will agree.

On preview, yes, you can set the propositions up like that, but the conclusion is false, unless you think that the second statement is implicitly "... protects US citizens and noone else"
posted by i_am_joe's_spleen at 3:54 PM on November 8, 2003

oh, so i'm bored, and pictures are fun.



posted by rhyax at 4:01 PM on November 8, 2003

rhyax - I like how it's a smileyface.
Also, the whole point is that Nike could be placed in the set of people with no contradiction, and the fact that there IS no contradiction is why the semantic web is flawed. Shirky and everyone else already understood that the fact of the matter is that Nike is not in the set of people, the problem is that the web has no means by which to decide what set to put Nike in.

"it is actually erroneous to conclude that Nike is a person at all; the confusion about the meaning of "person" is secondary"

Well ok, now soundness is an issue. ;P And of course the real problem is that presented with these premises, the semantic web would go "um" and that's about it. But then you've already said that.
posted by kavasa at 4:14 PM on November 8, 2003

That is essentially what you are arguing.

Well, yes and no. I don't believe that Nike is a person, Romania doesn't exist or that ducks are fish, but the pattern to reach that conclusion is part of a logical process. What I'm failing to see is how the two example statements we have been using are illogical, as you and that other not so polite guy seem to be saying. You write that my reasoning is inconsistent, but I'm not sure how.

It seems that nicwolff wants to subtract terms from the sets and you want to add terms, but that creates different sets and doesn't invalidate the two original examples.

There are some assumptions I'm making when looking through the statements. In the Nike example it seems implied that the First Amendment only covers US citizens since no other groups of people are mentioned; in terms of the statements, other people don't even exist. In a way (almost completely,) I see why you want to insert "only" into the statement in order to give it definition. This, I'll still argue, doesn't invalidate the overall logic of the set to conclusion.

I feel similarly about the vampire example. I think it is implied that "Not-Real-Things" as a set are too contradictory to be useful within the example. Perhaps there are logical ways to talk about the "not-real" interacting with the "real" but this example isn't trying to examine that issue.

So, in the end, I'm not arguing anything other then the two sets and their conclusions are logical. It's very clear that the conclusions are false, but the point is the most logical conclusion is not the correct one when applied outside of a given system.

On preview: What Kavasa said
posted by elwoodwiles at 4:24 PM on November 8, 2003

Oh, and out of the two examples I feel the Vampire is more flawed and the Nike set is more useful in the discussion, that's just me though.
posted by elwoodwiles at 4:27 PM on November 8, 2003

Let us assume that the Semantic Web contained only the following information:

• US citizens are people
• The First Amendment covers the rights of US citizens
• Nike is protected by the First Amendment

Suppose that you ask the Semantic Web the following questions:

Is Nike a person?

It is neither required nor impossible.

Does the First Amendment cover people?

It is required.

Are you a ham sandwich?

It is neither required nor impossible.

So... what's the problem here?

To answer my own question, it's here:

I am a US citizen. Am I covered by the First Amendment?

It is neither required nor impossible.

Nobody ever said anything about the First Amendment covering all US citizens...
posted by Ptrin at 5:17 PM on November 8, 2003

I'm having a Prolog moment...
posted by inpHilltr8r at 5:39 PM on November 8, 2003

Look!

Fuck.
posted by Satapher at 5:49 PM on November 8, 2003

elwood and kavasa: am I talking to myself here?! I feel like I'm taking crazy pills!

elwood: it does not follow from those terms that Nike is a person! I'm not subtracting anything from the set of premises with which Shirky claims you can prove that Nike is a person; there just really isn't any way to get from his premises to that conclusion. You're not allowed to make any additional assumptions - this is supposed to be logic!

Look: Shirky is trying to show that a computer might derive valid but clearly false conclusions by too-rigorously applying modus ponens to resolve syllogisms in a set of flawed premises. But in each of his examples, he fails to derive a valid conclusion at all! He reaches a false conclusion, not because of flaws in the premises but simply by logical error. And so he forfeits any point he's trying to make about what errors a computer might make.

kavasa: oh, kavasa. Maybe I don't have a whole lot of book larnin', but I sure can Google like a motherfucker. In sentential logic, an argument is a "finite sequence of formulas, the last of which is designated as the conclusion" (d6) [italics added]. And an argument is valid "if and only if one cannot 'make' its premises true without also 'making' its conclusion true" (d7).

So, when you smugly say "It is possible for all these statements to be true simultaneously. Thus, under a system of sentential logic, it is a valid argument" you are just wrong. Wrongitty wrongitty super-wrong with wrong on top. Because an argument isn't just a set of non-contradictory assertions.

And once again, the premises "US citizens are people", "The First Amendment covers the rights of US citizens", and "Nike is protected by the First Amendment" could all be true while "Nike is a person" remained false. So, neener neener.
posted by nicwolff at 7:43 PM on November 8, 2003

"the web has no means by which to decide what set to put Nike in"

Sure it does. You're just imagining that the semantic web would use English, whereas any real metadata scheme uses a controlled vocabulary containing standard terms. Such a scheme should be able to distinguish "person" and "human being" easily, and know that while all humans are persons, not all persons are human.

Where these schemes fall down in the real world is that mastering the metadata for a small domain (say cataloguing library books) is hard. As Cory Doctorow says in his essay, most people won't be arsed applying accurate metadata. He also worries about classifications organised on different lines - namespaces would take care of that. (Shirky would say namespaces for different vocabularies is solving the trivial bit, whereas the hard thing, eliminating ambiguity, is left unsolved).

What you might find in practice is that business applications will be accurate, because no one wants to get the wrong-sized sprocket, or a widget instead of a gadget - there's a commercial incentive to be accurate. I imagine little else would be accurately labelled - Cory again, with "worse-is-better", what we have now is good enough.

Ptrin, rhyax - thank you. You explained the logic issues much better than me.
posted by i_am_joe's_spleen at 8:16 PM on November 8, 2003

You've lost me here. Is Shirky saying that the semantic web, left to its own devices, will always commit the fallacy of the undistributed middle?
posted by Sidhedevil at 8:33 PM on November 8, 2003

Shirky's (flawed) examples obscure his argument: He's saying that the real world doesn't lend itself to syllogistic derivations, and he's right. This whole thing reminds me of the CYC project, wherein they are trying to formally encode a huge knowledge base. They have trained people doing it, and they end up with bugs in the ontology; they have the advantage, though, of controlling everything that gets into their knowledge base. Now, imagine trying to get that to work in a distributed system. Any reasonable semantic web has to be able to robustly deal with contradictory information. This whole discussion on whether it can be deduced that Nike is a person (answer: no, not logically) illustrates the point: encoding knowledge is hard and not intuitively obvious.
posted by brool at 9:06 PM on November 8, 2003

I believe Shirky's saying this:

- Deduction does not yield enough to be useful to us
- The semantic web relies on deduction
- Therefore the semantic web is not useful to us.

:-)

He also beats up the semantic web guys for treating what they've done as clever (creating a markup language etc for premises, and engines that read it) and treating what they've omitted as easy (creating an, unambiguous, reliable set of labels for everything in the universe).

In his first example, ambiguity in the term "lives" is the actual reason there's an apparent error, however Shiry attributes it to the logic - thus his example doesn't show what he thinks it does.

In his second, he does commit the fallacy of the excluded middle, but then he talks about the resulting error as though it were to do with confusion about "person". Again, the example doesn't relate to the text that follows. This is why I'm annoyed, because I really actually agree with him in the main, and by botching his examples he loses credibility.

Shirky goes on about "context" a lot, but all he's really showing is that natural language is full of ambiguity. There is no theoretical reason why metadata need be so ambiguous as that though - this is why Cory is so much more convincing, because he shows why we have lousy metadata in practise.
posted by i_am_joe's_spleen at 9:12 PM on November 8, 2003

The Semantic Web's philosophical argument—the world should make more sense than it does—is hard to argue with.

In fact I'd say it's astonishingly easy to argue with. There is no requirement whatsoever, from any source other than our own desires for order and pattern-sensing, that anything make sense.

The idea of 'making sense' in and of itself is a slippery and subjective bastard.

Interesting as the logical disputation above is, it reminds me of the three wise men and the elephant a bit.
posted by stavrosthewonderchicken at 9:14 PM on November 8, 2003

In case this hasn't been mentioned on this thread - Wittgenstein's (apocryphal?) Italian rude hand gesture satori on the train
posted by troutfishing at 12:04 AM on November 9, 2003

I don't think Shirky's US/Amendments statement is flawed, I think that the wording he used has an ambigious meaning.

I think he should have said...

- US citizens are people
- The First Amendment only covers the rights of US citizens
- Nike is protected by the First Amendment
- Nike is a person

Which is logically correct.
With his Medical Analogy, he's confusing the same term (person) as applied to two different fields because it sounds the same.

e.g.
A.I. is used to breed pigs.
Computers are needed for A.I.
Computers are used to breed pigs.
posted by seanyboy at 5:51 AM on November 9, 2003

I believe the "right" conclusion is: "The Transylvania where Count Dracula lives is not real." (I would have used "imaginary".)

I think the right question to be asking is: "do we really need a mark-up language for this?"
posted by wobh at 11:46 AM on November 9, 2003

Seanyboy, I love your A. I. example. However, your first example is still fallacious, formal logic-wise, because you didn't say "All US citizens are people". That's a minor point.

However, the more important point is that it's fallacious, fact-wise, because we know that the First Amendment doesn't "only cover the rights of US citizens".

What you've come up with, there, is the equivalent of: Men are mortal. Socrates is immortal. Therefore, Socrates is not a man.
posted by Sidhedevil at 12:21 PM on November 9, 2003

"Yours would be more likely to be taken seriously if you weren't a ridiculous prick."

Dude. You either correctly use propositional logic or you do not. You do not.
posted by shabrem at 12:31 PM on November 9, 2003

What I want to know is, who is Charles Dewey, and is he by chance related to Melvil[le] Dewey, the inventor of the Dewey Decimal Classification?
posted by IshmaelGraves at 1:41 PM on November 9, 2003

I'd like to say that taking the sort of relish necessary to write something like "Wrongitty wrongitty super-wrong with wrong on top." is, again, pretty much the sign of a ridiculous prick - even if you're right .It's true! To form an argument from a set you need to have an inconsistent set, take one member of that set and set its negation as the conclusion and have the remainder of the set be consistent. Curses, I forgot a technicality of set-theoretical logic (which I never use for anything), you sure showed me! - kavasa, hiding in MetaTalk cause he's too embarassed to comment here

kavasa, I took tremendous relish in pointing out how stupidly wrong you were because you were so stupidly, pedantically, smugly wrong in the same comment in which you called me a ridiculous prick!

That's right, you forgot a "technicality of set-theoretical logic" - one that is pretty much the basis for all logic! - in the same comment in which you told me "it's you who needs to take Philosophy 101 again"!

Hence provoked, I have bitch-slapped you repeatedly, and I suggest you learn to like it.
posted by nicwolff at 3:50 PM on November 9, 2003

Um, nicwolff, things in MeTatalk are best kept in Metatalk, that's what it's for.

As to the discussion at hand I have found i_am_joes_spleen more helpful then your comments and I'll concede that we must add "only" to the Nike example in order for the conclusion to work with the premises.
posted by elwoodwiles at 4:46 PM on November 9, 2003

Yeah, it was hard to decide which thread that was less irrelevant to. In which case, I should have taken it to e-mail... Sorry, all.

Glad you saw the light, though. Now about the Dracula mess: it's a lot clearer if we resolve the two pairs of premises that share a middle term, which leaves us:

- Count Dracula is not real
- Count Dracula lives in Romania

Shirky's claim about these premises is that "[t]here is simply no way to cleanly separate fact from fiction" by which I think he means that if you are going to allow fictional things to have real attributes in the metadata, then your syllogisms may yield false conclusions.

But his example, "Romania is not real", has two problems: 1. it doesn't follow from those premises in any logically valid way, and 2. it isn't false! The Romania being discussed is in fact not real - it's the fictional one that Dracula lives in. In fact, there are an infinitude of Romanias: one for every set of assertions or beliefs you can make about the country.

Now, is representing that multiplicity of Romanias a challenge for AI? You bet - look upon CYC's microtheories and despair. But if you're going to write off so many peoples' work, then at least your examples had better hold together.
posted by nicwolff at 7:41 PM on November 9, 2003

Gasp! ".tr" - isn't that... Transylvania?!?!

Yes I know it's Turkey, no I didn't have to look it up at all, not me.
posted by i_am_joe's_spleen at 7:58 PM on November 9, 2003

.When nicwolff snaps his suspenders, does it make any noise?
.Does Clay Shirky know the difference between metadata and data?
.Does everyone realize that metadata is related to AI research in exactly the same way as foghorns are related to sombrero galaxies?
.Does anyone care at this point, or are we all backing toward the exits, safely enshrouded by familiar clouds of metapixiedust?
.Me - I'm late - I reiterate, I'm later than late - for a problem I really must qualificate, and I just have time to get to the lab in a fuzzy wuzzy ramified cab.
posted by Opus Dark at 3:23 AM on November 10, 2003

The transactional part of the article seemed a lot more interesting to me... pretty broke-ass way to do things, if you ask me.

As for this bickering about Shirky's (in)ability to put together any sort of coherent argument (in the logic sense), I believe that is the point he was trying to make. If you really want to create enough metadata that will allow an intelligent entity to make lots of solid connections, you're going to need a ridiculous amount of metadata. In his examples, we'd need many qualifiers (all, only, any, some, fictional/real) to draw further conclusions. No one is going to bother to set all these flags, so you end up with holes.

If I plugged a NY zip code into a metadata search engine and asked for celebrities living in the area, I might end up with characters from Seinfeld and Spider-Man.
posted by mikeh at 8:01 AM on November 10, 2003

Just found this: Deconstructing the Syllogistic Shirky [via DECAFBAD]
posted by brool at 10:13 AM on November 10, 2003

Because no one's mentioned this yet: Jorn Barger. Also, I believe the classification issues brought up here to be related. Just a hunch though.
posted by wobh at 12:39 AM on November 11, 2003