Slashdot Mirror
Challenging the Ideas Behind the Semantic Web
mytrip writes to tell us that after a recent presentation to the American Association for Artificial Intelligence (AAAI) Tim Berners-Lee was challenged by fellow Google exec Peter Norvig citing some of the many problems behind the Semantic Web. From the article: "'What I get a lot is: "Why are you against the Semantic Web?" I am not against the Semantic Web. But from Google's point of view, there are a few things you need to overcome, incompetence being the first,' Norvig said. Norvig clarified that it was not Berners-Lee or his group that he was referring to as incompetent, but the general user."
is the users.
Not the ones searching but the ones creating the content.
They'll be some idiot out there (like there is now) that will code his data in a way that guarantees that he gets the most page views etc. So often searched terms will turn up on search indexes and other ilk.
It's a loosing proposition unless you come up with filters but then they have their own set of problems.
There's a gorilla from Manilla whose a fella that stinks of vanilla and has salmonella.
"Norvig clarified that it was not Berners-Lee or his group that he was referring to as incompetent, but the general user."
Here I was, thinking we were arguing over Semantics...
Help save the critically endangered Blue Iguana
I'm calling the Anti-Neutrality Web Designers of Amerika!
Demands of inequality such as this should be allowed!
(btw, the spelling doctor has "loosing" as in "loosing the hownds for the huhnt")
But you just gotta have another sigarette
The current semantic web seems to offer a technology too fragile to use on the global scale. The complexity of various classification and ontological schemes, work needed to provide the metadata etc. Also, semantic web seems to offer great opporturnities for spammers and other mischief makers. Now we already have comment and reference spamming, but semantic web (on the global scale) raises the possibilities enormously.
I dunno, I'm guessing he knew exactly what he was saying... But I do wonder if he was trying to tease Tim Berners-Lee a little. It would be interesting to see/hear audio/video of that exchange.
The biggest problem with the semantic web is spam. If you can trust the tags, it's a beautiful idea. If you can't, it's worse than useless - it's a waste of time. Google has the right idea, automatic extraction of semantics from content. If there's no real content, then (hopefully) that will be reflected in the semantic analysis.
Me, I estimate we're 5-10 years away from doing anything terribly useful with all of this stuff, but I can definitely envision the day when an internet without semantics seems as distant as an internet without Google.
Thanks for the illustration of what Norvig meant. How is "Google Director of Search and AAAI Fellow Peter Norvig" (original article) semantically equivalent to "fellow Google exec" (Slashdot summary)? The latter suggests that Tim Berners-Lee too is a Google exec, which would be news to him.
Semantic webs (emphasis on plural) produced by editors such as those at /. or in the consumer-rated style of Digg, Del.icio.us etc might actually work. Trusting authors to do it right is a disaster, as Norvig suggests.
It's really, really difficult to get people to follow rules. We're lazy, we're incompetent (yes), and some of us are evil. I still don't think I truly understand how RDF is supposed to work exactly, and it doesn't even seem like it will be fun to try.
On the other hand, it's really easy to release a million monkeys and let the create what they will. It's not so easy to sort through what they end up producing, but Google does a surprisingly good job of this.
It reminds me of the early days of the Web, when companies like CompuServe and AOL wanted to design and own all content. On the other hand, an internet server with httpd let anybody make a ~/public_html directory and put up whatever they wanted to. The million monkeys won that battle. I think they'll win this one, too.
From http://www.7nights.com/asterisk/archive/2004/03/do nt-blame-the-users
Maybe the Semantic Web should aim to be useful to people rather than require people to be useful to it. There has to be a better way than trying to educate droves of people to a problematic and vulnerable design.
I design user interfaces for a free network management application,
In one of the very first papers mentioning the Semantic Web, some paragraph was devoted to something then lost in the hype around the semantic web: the Web of trust, which had to be something like a certification of metadata. This is perhaps to be again regarded as important for the semantic web and the web in general (although not easy to manage).
By the way, Norvig is not only a Google exec, but also a well known AI researcher, author of one of most important books on that subject.
Slightly offtopic. Peter Norvig gave a talk at my university on similar topics, and there was a short Q&A afterwards.
:)
One of the students asked him what he did for his 20% project. He said that he was usually too busy keeping tabs on what the other employees were doing with their 20% time, so he didn't quite get around to working on his. He told us what he wanted to do, as motivation for himself.
The basic idea is that when he used to work for NASA, it'd always make him upset when people saw faces in random spots on the moon's terrain, and claimed it was aliens that NASA was covering up, or similar. So, he was planning on taking facial recognition software and running it on all of google earth. I think it'd be pretty awesome..
Any progress yet, Mr. Norvig? I'd love to see the results..
Powered by Web3.5 RC 2
The jig is up!
It's the business users too that are a problem. I'm currently trying to get a project on the rails based on semantic web technology, and I'm confronted with an IT department where some are even struggling with the difference between subtyping and instantiation- let alone more advanced modelling issues... It doesnt help ofcourse that most people never even heard of conceptual modelling languages such as ORM but instead were thought to use uml and ER where it's the modellers' responsibility to make a distinction between what is conceptual, logical and physical which ofcourse most never did.
In regards to the google issue I think the idea that you should crawl everything is faulty cause you need to be able to trust the source. Most ontologies will simply be restricted to a certain domain and corresponding user group, often in a b2b context. Integrating every man and his dog, the lawnmower and the kitchen sink with some kind of top level ontology is merely a nice-to-have philosophical issue that I dont expect to be solved in the near future, if only cause we havent seen much advances since Aristole started toying around with the idea. In other words, at google they are worried about an issue that's atleast a decade away from now, probably even more.
That anti-semantic bastard...
True, the web had a similar problem, however creating a webpage is a lot more interesting (you see the results directly, how terrible they might be you do see a result) than structuring data. The latter takes a lot more work, and the direct benefit just isn't there.
Sem-Web-like standards like RSS, XML and SOAP have become mainstream, but primarily because they fill a gap. The adoption of RDF or OWL simply doesn't solve anything. Yet. It would be cool to let agents loose onto the semantic web and retrieve them together with a summary on a certain subject using a multitude of sources, but as long as it's easier to Google I don't think it would generate any interest outside academia.
Feel free to prove me wrong though.
This sig is intentionally left blank
Especially if the rules appear to be an incomprehensible ad-hoc mix of principles taken from a dozen not-quite-fully-baked AI dissertations.
I still don't think I truly understand how RDF is supposed to work...
I don't think anyone does.
I'm not saying that the semantic web is bullshit, but it does trigger my bullshit detector. At least one of them must be broken.
Am I part of the core demographic for Swedish Fish?
Even if we are inherently lazy, and even though some people seem to be generally against the idea, it doesn't make any sense to me not to employ this and experiment with it. Norvig is an AI guru, and his ideas on the Semantic Web may be interesting, but Google is not against the idea. Google's GData looks to me like a primitive Semantic Web. Even if only 10% of web masters adopt the system, querying to find a set of results that have been tagged as certain meta-data can come up with some interesting results. If the results are interesting enough, more webpages will include meta data tags. Also, being inherently lazy argues for not spending time writing tags all over your code, so why would anyone take the time to sabotage the system. While I understand the difficulties of the spamming problem, there are plenty of cookies on the internet anyway. I think the same inherent problem in the Semantic Web exists with PageRank. In PageRank what happens is a web page will say the same words over and over to acheive a higher ranking in the semantic analysis of the page, and thus the page will be a top result when entering a query with related words. But I think PageRank works pretty well overall. Google's next step with PageRank is to filter all the spam sites that just say the same words. Security in the Semantic Web would also be to filter those sites with obviously spammy RDF or OWL tagging. Overall the Semantic Web is a cool project that could lead to really smart searches, with axioms involving how different meta-tags are related to each other. I'm in favor of the new technology.
The idea of RDF is applicable to much more than public innerweb content. I've spent the last 7 months researching and developing an RDF backed system for my company's core products. Everyone should think of the value of RDF beyond the scope of trust, and then it becomes easy to realise methods of simple non-web implementation. We can all spend the next 5 years pondering how we're going to figure trusted content providers for RDF web data, or we can just start developing apps for sources which understand themselves as trusted (ie. data input from an individual, employees of a company, and any group where the individual must be accountable for their actions). Whats more important than the blind trust of sources, is data verfication. There are ways to run data input from one user by another user, without doing it in an infringing, demanding way, for validation. I'd like to go into detail of exactly what I mean by all this, but I don't want to violate any portion of my NDA or tip off industry competition (I know that sounds retarded, sorry). If RDF does gain popularity, I can say it will from within the private sector, not the public. Genious implementation may bring RDF to the public sector, but thats not something I would say is guaranteed to happen.
Current technical obstacles to creating any RDF applcation: The matter of complexity of its integration into DB backed systems (popular methods), and instatiated class marshaling within not-so-object oriented languages. The technical design and implementation of a standards compliant RDF system has been extremely difficult for me. I don't think it would ever be possible to get RDF data represented nearly as minimally as you could with simple relational tables (although formally no more bloated than bloaty XML). RDF also creates many long linked relationships; this tends to create some serious performance issues in querying the data. Lastly, I hate XML, and you can't always correctly export from RDF to XML (capable type to incapable type) in a correct manner.
This remind me of the famous Semantic knigth parody...
Do not forget that the semantic Web is not a replacement of the existing technologies: HTML contents will always be there but, What if these little 'metadata' description where added to ALL the Web Pages? In this case, the pages could be categorized, analysed and searched much more easily, and the algorithms related to these operations would be better. In such an scenario, the use or one or another Web search engine would be irrelevant because all of them would have powerful and acurate algorithms. Maybe a threat to google's business model? These would be the perfect world, but we have to assume that Webs would certainly lie or made mistakes in their semantic descriptions. OK, but... would it produce an scenario worst than the actual?. Now, fake webs are quite common; irrelevant sites try to advertise them by using all the available means to attract the most visitors, misleading them. The best web search engine is this who best filters these sites in the searches. In a semantically described Web, the problem will be the same, but there would be another easy-to-use filtering criteria to enhance the results. the Web search engines' algorithm will be better for sure.
Bingo! You've just proven that the incompetence spreads beyond MySpace.
The problem with the semantic web movement is this: You have the web guys from the W3C who got famous by building kinda crappy, but effective technology (HTTP, HTML, etc...) going goo goo gah gah over PhD Ontologists from the AI community. They team up and build these great things that the average person (including the people who think they are really really smart, like the Slashdot editors), has no chance in hell of using effectively. What'll happen, is that eventually there will be useful Semantic content and Intelligent Agents doing great things, but that work will be done by a select few. The unwashed masses will still be the domain of Google.
Help me take back Slashdot. When did 'News for Nerds' become 'FUD and Conspiracy Theories for Extremist Nutjobs'?
As I pointed out in the previous comment authoring data on the semantic web is no more difficult than authoring RSS or XML.
Yes, figuring out for the first time how to represent your data in RDF (or XML for that matter) can be difficult. Imagine if everyone was trying to come up with an RSS standard on his own instead of using RSS export functionality of his content management tool. That's why we need good guidelines how to publish information on the semantic web. And RDF export functionality (plugins) similar to what RSS plugins are doing.
As for opportunities for spammers and mischief - don't think so.
Why? - If you look at the Semantic Web "layer cake" you will notice such technologies as digital signatures, encryption and trust being part of the scheme. They allow to identify the author of data and ensure he is what he claims to be. There is nothing wrong with your application if it only accepts signed and trusted data. And there is nothing preventing authors of the data from signing the contents. Since the semantic web is a new technology and we already know about problems that spam and misuse can present it is more not less prepared to fight spam.
Note1: Semantic web should be viewed as an integral part of the existing web, not its opposite. Might even be that it can provide an additional layer that will help to combat spam and other problems you mention here. Who knows.
Note2: Spammers will always try to come up with new exploits. We all have to be prepared for this and think how to close the holes they are using. But saying that newer (a further development of existing [web]) is necessarily more opportunities for spammers in wrong.
Here is a Tutorial on the Semantic Web.
Pay attention to the slide #22 which shows how data from different sources can be merged together. This is one of key differences between XML and RDF - to merge XML data from a number of different schemas one would need to create an application that processes data in these schemas and generate merged data (possibly inventing a new schema to represent the merged information).
In RDF that happens "magically" - in order to merge heterogenous data you don't need to do *anything* - just put all the information in an RDF store and it merges. If the data to be merged change no modifications to the store are necessary - it is like a bag that can hold anything.
Ugh, this is the major misconception of proper Semantic Web implementation.
There are two user types of Semantic Web materia: the individual user and the group.
The individual user only cares about context. It's like a Proustian adventure for him. If he tags Slashdot as "blatherscyte" because that's how he views it, then that's valid. If he tags it as "cmdrTaco" because he is stalking Rob, then that's valid, too. And if he tags it as "monkey" because one time he was petting a monkey while he viewed the site, then that's valid, too. It's like the old saying, "Whether you think you can or think you can't, you're right." There are no wrong semantics for the individual user, because it is his context alone which defines the usefulness of a tag.
For this reason, the individual user should be allowed to tag freely and without limits, and also be able to edit or remove tags later.
----
Now for the group, they have a different goal. Context does them no good, because they don't have the same context. Their goal then is consensus. Take your problem at FreeDB. The simple solution is to let people vote on the accuracy of disputed tags. Or flag ones they view as incorrect, and then review those that meet a certain threshold for flagging. Basically, you want the group to filter out things that don't apply to the group, WHILE maintaining individual context. You don't delete the tags that the group has rejected - you just hide them from the person who has come to view the group tags.
I think this dichotomy of group vs. individual is what has gotten us into trouble with the Semantic Web. To use one example, I think delicious' big mistake was to show you "popular" tags for a given link. What that does is encourages you not to create your own tags, but instead just piggyback on popularity. Over time, this creates homogeny, which is great for the group, but not for the individual user. Sure, they can probably find that link again in a minimal amount of time, but if an individual tag might help them find it faster, but they shunned individual tags for groupthink, so much the worse for them.
And on the flipside if you don't provide proper weighting and trust metrics into your tagging system, you are opening yourself up to not only abuse and inappropriate behavior, but also to the "incompetence" mentioned in the article, which is not so much incompetence as a zero-filter. It's like reading Slashdot at -1. It's kind of a touchy-feely way to look at it, but in Web 2.0 thinking, it's bad to delete content; just filter it out instead. It's bad to censor opinions from the software side; let each user do their own stifling. Give the users complete control over the content, and they will find models that work. It's that simple.
The main problem with the Google guy's point is that philosophically, Google is more groupthink than individual user, because they're a search engine. They value consensus over context. In the future, perhaps they will value context a little bit more than they do. Until then, they have to stand where they stand, because they can't let context into their system. They've tried some clunky mechanisms to do so (Personal Search, anyone?) but until they get it right, the Semantic Web won't have any value to them.
But what, exactly, is the definition of the 'Semantic Web'? How is it different from what has been done in the past? Is there any agreement of any sort as to what it means? If yes, please let me know. If not, then how can we achieve this goal if we do not know what it is?
I am confused, I really do not see too many differences in the web in the last few years. Nothing 'Earth Shattering' anyway.
putting the 'B' in LGBTQ+
Depends--if Norvig got Russell (co-author with him on Artificial Intelligence - a Modern Approach) to go in with him for a tag-team kind of thing, they'd probably win. On the other hand, Berners-Lee has the W3C on his side, a notoriously large and heavy organization, which could be hard to topple.
As a side note, I heard from a friend who was attending that Norvig's opening comment about people always asking him "Why are you against the Semantic Web?" was a response to Berners-Lee's opening, 'Poeple always ask me, "Why are you against Artificial Intelligence?"'
It's trying to impose structure on something that is not very structured--human thought. Even the use of the word "semantic" points out the futility of the exercise, as it indicates language and changes in meaning--not structure.
Semantics is a human discipline--it is focused inward, not outward. Likewise the proper place for semantic technology is in the client, not the content. Building "semantic web sites" makes no sense. Google is absolutely right on this one--Web sites should simply be what they are, and it is up to the client to assign meaning and remember connections. Google provides a variety of tools that help people do just that.
Why should I have to tag everything I read online? I don't tag things I see in real life. I just remember them and make connections in my mind. If we want computers to be actually useful to us as assistants and not just stupid tools then they will need to begin to operate the same way. That is a very tough problem, yes. But it is the way we are headed, and the "semantic web" is IMO just a bad hack until we get there.
Furthermore the idea of trustworthyness and authority online is ridiculously complicated. I can't think of a harder problem in all of AI. It's much harder to determine if someone knows what they're talking about, or if they are trustworthy, than it is to simply identify the topic of an article. And we're still struggling with the latter.
Build a man a fire, he's warm for one night. Set him on fire, and he's warm for the rest of his life.
That was the entire point of my post! The group benefits from standardization, but the individual suffers. The Semantic Web is an attempt to give power back to the individual user. Subjectivity is a crucial element of the system, and sanitized, standardized, NPOV systems deny the individual subjectivity.
Delicious is very smart in that it left the *option* for customised tags, but they are clearly saying by implication that the best tags are the ones everyone else is using. My point being that the idea of a "standardized vocabulary" is antithetical to the ideals of the Semantic Web. We don't want a democracy of ideas; we want a free market of ideas!
Think of the concept "funny." Let's say I asked you to go to 100 different random sites and tag them as funny or not funny. Let's say that of the sites you listed as funny, it was clear you enjoyed witty, New Yorker-style humor, and not fart jokes. But let's say 99 other people did the same thing, and they did the opposite: they clearly enjoyed the fart jokes, and hated the New Yorker wit.
Now if you asked this seeded engine for a recommendation of a new, 101st site that was funny, should it give you fart jokes, or New Yorker style? This is the power of the Semantic Web. What's funny to you, isn't funny to everyone else. Why should you be punished for that? And if a total n00b comes to our engine for a recommendation, they get the fart jokes page, because it assumes they're like everyone else. But if they start marking those sites as not funny, eventually it'll figure out they're more like you, and start giving them sites that you like.
Now, will delicious ever do that? Of course not, because it doesn't offer any discrimination to you on the word funny. You get the democratic version of funny. Fart Jokes for all. And that's what "standardization" has to offer. So, no, you can keep that; I want the Internet to understand who I am, and what I like, not what everyone else likes. And if they HAPPEN to coincide, that's fine, so much the better - things are popular because of the people, after all - but they shouldn't have to.
I won't be so sure (that it is not a general case).
A regular user won't be inventing his own ontologies the same way as he is not inventing a new RSS format. There is a set of well-define ontologies that you can use to describe your data. And a regular user won't be hand-crafting RDF data either. Instead RDF data will be exported from his applications the same way as RSS and Atom are exported from his weblog software or as Word saves users files.
RDF data will still merge together, provided there are "crystalisation points" that are common to data from different sources.
Regarding Luc Steels' research you are mentioning - could you give some pointers to his work?
Google seemed to be looking to the web for meaning, but they should be building their own Ontology of Everything, based on what they find in the content. Let Cyc loose on there caches perhaps would be a good start. Then integrate their Ontology of Everything with those Formal Ontologies that already exist. About Intelligent Searching, when a person asks me for advice, I tap into my Personal Ontology, which has overlap with other ontologies in a domain specific way. i.e. I read information, much of it structured, I then fit it into my Personal Ontology and if required expand my ontology to fit the new information. I may even face a paradigm shift that requires a major restructuring of my Ontology, i.e. I need to have a set of new transforms to link the old with the new in a way that lets me sanely access both. At this point I have acquire new Knowledge which I can now share with people that ask me questions. When I'm talking to a Knowledgeable Source I need to find the Transforms that allow me to incorporate "knowledge of shared knowledge" as well as knowledge of our unique knowledge. This is how we are able to communicate and learn from each other. If I am dealing with a Naive Searcher I need to Probe their Personal Ontology or World View until I am able to construct enough domain specific transforms to allow me to know what they are trying to learn and how best to find it and Teach it to them.