Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ontology Is Overrated: Categories, Links, and Tags (2005)

shirky.com

Post reply on HN
Print view View original

81–90 of 94 posts

Re: Ontology Is Overrated: Categories, Links, and Tags (2005)

#81

Post by barrkel »

ChuckMcM wrote: post #7

I enjoyed this piece, mostly because I disagree so strongly with it :-). Clay takes the topic of Ontology, contextualizes it into an impractical domain, and then calls it overrated. Cars are overrated too because they can't cross oceans right? When done correctly, ontology combined with knowledge graphs can make an impractical problem much more practical. Consider for the moment the problem of 'fake news'. What chara…

Ontologies are problematic because they are almost invariably hierarchical and binary.

If your ontology is probabilistic and multi-dimensional, it's much more likely to represent truth.

I actually think ontologies are a trap for smart people. For people who like to systematize, they have an allure. Ontologies give you the illusion of putting the chaotic world into order. Even phrases like "true facts" - in reality, facts are hard to come by and change over time, and the closer you look at them the more woolly they get - predicates are far too abstract and simplistic, and when you dig in, the fractal messiness overwhelms.

Re: Ontology Is Overrated: Categories, Links, and Tags (2005)

#82

Post by astazangasta »

JetSpiegel wrote: post #67
heavenlyblue wrote: post #52

Earlier quoted context omitted.

I must absolutely disagree with you: in my opinion the problem of fake news stems from compartmentalisation of news sources (lack of their diversity), rather than lack of actual facts in the news sources. The fact that you can read about the same topic in several (more often opposing) ways is already a reason to red-flag the piece in the first place. This isn’t readily available through any of the current newspapers.…

> I would happily pay for an aggregator that simply ejects facts There is already the AP news ticker, or Reuters. Bloomberg has the same thing for financial matters. If you read only that, you will die of boredom.

The AP news ticker has its own bias and narrative, as does Reuters. There is no truth in any text, only narrative. So long as you are reading trying to find truth in a single text you will fall into error.

Re: Ontology Is Overrated: Categories, Links, and Tags (2005)

#83

Post by usrusr »

dithering wrote: post #68
ChuckMcM wrote: post #7

I enjoyed this piece, mostly because I disagree so strongly with it :-). Clay takes the topic of Ontology, contextualizes it into an impractical domain, and then calls it overrated. Cars are overrated too because they can't cross oceans right? When done correctly, ontology combined with knowledge graphs can make an impractical problem much more practical. Consider for the moment the problem of 'fake news'. What chara…

Putting aside the difficulty of building that database of true facts, and of parsing unknown facts out of English text, that solution feels like it's inviting an arms race. I can write a bunch of uncontroversial "true" facts into an article containing one or two false facts.

Diluting lies with verifiable but only tangentially related truths is a well-established tactic to fooling humans as well.

Someone with an agenda in the symbolic vs statistic AI question could take this parallel as an example of how close ontology-based AI approaches are to the way humans think. And then someone with the opposite agenda would point out that the example is all hypothetical.

Re: Ontology Is Overrated: Categories, Links, and Tags (2005)

#84

Post by mac01021 »

ThomPete wrote: post #78
cdbattags wrote: post #55

Earlier quoted context omitted.

I know this might be fuel to the fire on here, but I think Tanushree's work at Georgia Tech in her paper "A Parsimonious Language Model of Social Media Credibility Across Disparate Events" [1] is a good stab at the "fake news" problem with a lens of "credibility". As a disclaimer, we tried to get this model off the ground in YC summer 2017 batch but were rejected after the phone interview. I did not assist in the res…

The problem is that "Fake News" is a misleading category in itself. The primary issue with "fake news" isn't political it's advertising. Fake news is made mostly to lure you into clicking or sharing which in turn increases the advertising revenue. A lot of it is saying outrageous conspiratorial stuff but in essense it's not different than tabloid news which are often also "fake news" On top of that what might be cons…

> what might be considered fake news by some doesn't have to be by others because it might have utility

Do you have an example of this in mind?

Re: Ontology Is Overrated: Categories, Links, and Tags (2005)

#85

Post by JetSpiegel »

astazangasta wrote: post #82
JetSpiegel wrote: post #67

Earlier quoted context omitted.

> I would happily pay for an aggregator that simply ejects facts There is already the AP news ticker, or Reuters. Bloomberg has the same thing for financial matters. If you read only that, you will die of boredom.

The AP news ticker has its own bias and narrative, as does Reuters. There is no truth in any text, only narrative. So long as you are reading trying to find truth in a single text you will fall into error.

If they contain only facts, there is no narrative.

The only narrative is the selection of what facts to report. Without that you get lost in the deluge and get equally uninformed.

Re: Ontology Is Overrated: Categories, Links, and Tags (2005)

#86

Post by JetSpiegel »

heavenlyblue wrote: post #80
JetSpiegel wrote: post #67

Earlier quoted context omitted.

> I would happily pay for an aggregator that simply ejects facts There is already the AP news ticker, or Reuters. Bloomberg has the same thing for financial matters. If you read only that, you will die of boredom.

Well, truth is boring, isn't it? Would you otherwise prefer to read Russian propaganda speaking of German authorities policing Germans of Russian heritage? Because that's exactly what they are currently doing. 0 facts, amazing opinionated narrative in multi-paragraph form. Pictures on top of that. But that definitely sounds more fun. Moreover - that sort of narrative is way closer to the general sentiment of the work…

I can't be so boring it leads to complete apathy. Unless you are a computer, there's a limit to the boredom you can take.

Re: Ontology Is Overrated: Categories, Links, and Tags (2005)

#87

Post by tabtab »

twblalock wrote: post #27
tabtab wrote: post #12

Earlier quoted context omitted.

Indeed. This is also why I think OOP is missing something to scale better in terms of flexibility and domain size. The ideal code structure for a particular need is rather arbitrary. OOP "prefers" things in a hierarchy, both in terms of inheritance, and in terms of an object being less powerful than a class (in most languages). You can force or use OOP outside of these, but it's unnatural in my opinion. For example,…

You can do OOP without inheritance. I do it all the time, including in Java. It works extremely well alongside dependency injection. You still get all of the other parts of OOP, like encapsulation and polymorphism. The only real trick is to use delegation patterns instead of inheritance, i.e. dispatch requests to class members rather than super/subclasses. It doesn't feel awkward or unnatural.

It does feel awkward to me. I tend to view it more like RDBMS design and queries where the structures fit the domain and one can query for subsets or dispatching using something as flexible and simple as SQL (relatively speaking).

Then you could easily answer:

A) Show me all the GUI event handling code for all buttons. B) Show me all the event handling code for Form X. C) Show me all the event handling code for all buttons within forms having at least one drop-down list.

Similar goes for executing code (running business logic), not just code inspection. Then one doesn't have to crawl object graphs as often. Early databases did graph crawling, but then Dr. Codd showed a better way, and graph-DB's mostly withered. I'm waiting for the Dr. Codd of code and behavior dispatching management to come save the day.

I'd like to explore using relational modelling for behavior, not just attributes (data). I used to do such in xBase (dBASE, FoxPro, etc.) because of its dynamic nature and ease of editing tables, including code in tables. I thought that direction was the future, it looked bright to me. Then OOP came along and killed the seeming birth of table-oriented programming.

(It might seem like a security risk to put code in tables, but the difference between a file system and database are not different enough to say putting them in a database is "bad" while putting them in a file system is "good". I'd like to blur the distinction between a file system and database, but maybe that's a diff topic. Plus, one doesn't necessarily have to put code "in" the tables, just references to it in tables. One just needs a system/convention to track and map each to the other.)

Re: Ontology Is Overrated: Categories, Links, and Tags (2005)

#88

Post by tabtab »

narag wrote: post #56
tabtab wrote: post #12

Earlier quoted context omitted.

Indeed. This is also why I think OOP is missing something to scale better in terms of flexibility and domain size. The ideal code structure for a particular need is rather arbitrary. OOP "prefers" things in a hierarchy, both in terms of inheritance, and in terms of an object being less powerful than a class (in most languages). You can force or use OOP outside of these, but it's unnatural in my opinion. For example,…

For example, why is it not easy to add an "onClick" event handler method to a particular button in a Java GUI? Lack of pointers. An event reference is a double pointer: to the method and to the object that is "self" or "this" for the execution of the event handler. Since they refused to have pointers in the language, they devised that convoluted solution.

When I try to design an "ideal" GUI system on paper, I start with a relational model. Relational is quite adept at dealing with "pointers" in terms of foreign keys. A given snippet of behavior (such as an event-handler method) can be easily associated with multiple things; something OOP has a hard time with. Thus, we don't have to go back to direct address pointers to solve this issue, but rather learn from relational modelling, as I hinted at in a nearby reply.

(One problem with traditional RDBMS and GUI's is that different widgets need a variety of similar and often overlapping attributes {columns}. It's not realistic to have a table dedicated to each widget "type". Thus, I propose using "dynamic relational" instead, in which column existence is optionally situational.)

Re: Ontology Is Overrated: Categories, Links, and Tags (2005)

#89

Post by skybrian »

cheschire wrote: post #75
skybrian wrote: post #45

Earlier quoted context omitted.

This is true of names and borders, but the cities themselves remain where they are even when names and borders change, so they're more persistent than countries.

That might seem logical on the surface, but dig just a little deeper and the flaws become readily apparent. For example if you're looking for the population of New York between 1850 and 1950, you would see a potentially dramatic shift of the population around the turn of the century, which would be misleading if you didn't already know the caveat about the consolidation of the city in 1898. However that shouldn't dis…

Well, yes, few things are entirely immutable. Cities do grow and (rarely) shrink or disappear. But this is rare enough that it's useful to hang other data off of their names.

For example, take the locations used in time zone databases, where an official way to name a location is something like America/Los_Angeles. The assumption is that major cities don't get split across time zones and users know which cities they are nearby that they share a time zone with. Country names aren't used.

A lat/long pair would be more precise, but sometimes precision isn't needed.

Re: Ontology Is Overrated: Categories, Links, and Tags (2005)

#90

Post by james_s_tayler »

cheschire wrote: post #77
james_s_tayler wrote: post #65

"when people were offered search and categorization side by side fewer and fewer people used categorization to find things" Two thoughts pop into my mind which guide me gently to the conclusion we need to be careful here. 1) every configuration is a set of trade-offs 2) any parameter or paradigm pushed to extreme produces perversion On point number 1: categorization contains the benefit of discoveribility when you la…

As a solution developer, I find the most bang for my buck is when I produce solutions for people who come to me with a problem rather than a solution. Imagine if Google made you answer the question "Why?" five times before it showed you results. I wonder if a well trained network would produce such accurate results we would no longer need to see a list of results, and could simply hit the lucky button. But good luck…

Well theres no disagreement from me that what you describe is how it plays out.

My caution is that we have found the equivalent of cheap, fast, microwavable meals and we are now consuming them almost exclusively and in 15 ~ 20 years it's going to become painfully obvious what deleterious side-effects arise as a result of that. We can't see it now, so we don't feel the need not to do it, and that simply is what it is, but in time it will act as a control signal that nudges us to modify our behavior.

This is both a great and terrifying time to be alive. The 20th century was full of some pretty hefty social experiments and the thing with those were they were obvious and in your face. The 21st century versions are much less obvious and reveal themselves very slowly.

Post reply on HN