> "Do we throw out metadata, then?
Of course not. Metadata can be quite useful (...) often a good means of making rough assumptions about the information that floats through the Internet"
Could it be that over-confidence in metadata was common around this time (turn of the millennium), then everyone realized the limitations of meta-data as illustrated in the article, subsequently dropping meta-data approaches altogether, forgetting that it is also useful.
So now we have a situation where the pendulum is now at the other extreme, where machine-generated classification barely intersects with human-generated meta-data at all.
The sweet spot seems to lie somewhere in the middle, as Aristotle used to say.
> machine-generated classification barely intersects with human-generated meta-data at all
What do you mean by this? A big part of the reason that deep learning was a major advancement is the fact that it helps you extract data from metadata in a semi-automated (and increasingly automated) fashion.
Maybe the difference is that we gave up trying to "make sense of" the metadata in any kind of universal way.
We outsourced generation of high quality training data, resulting in Wikipedia being one of the few (and quite messy) people can play around with... garbage in, garbage out.
And to be fair, most people who lovingly curated special knowledge databases only to see them get used to train machines (without any acknowledgement) discouraged them to continue and resulted in a drop of good training sources (at least in my field and non-english language)...
>So now we have a situation where the pendulum is now at the other extreme, where machine-generated classification barely intersects with human-generated meta-data at all.
The problem is that human metadata has a pretty solid barrier, but the barriers of what machine metadata can do keeps shifting due to increases in technology, therefore people keep investing in machine processes because they do not yet seem to have reached their limit.
2.1 People lie
2.2 People are lazy
2.3 People are stupid
but there is also the obvious scalability issue regarding people adding metadata - if one person can add metadata to 100 files per hour and you have 1 million files a day you will need a large team of people adding metadata.
Nah. That line was a sop to the metadata advocates by pretending that some forms of useful not-metadata (or at least not capital-M Metadata) are actually metadata. Sometimes we overcorrect when an overhyped technology doesn't live up to expectations. But other times there really is no there there and the whole thing is just completely useless.
Was always my reaction when hearing about efforts like The Semantic Web.
"Yeah, we can get everyone to classify and tag their data with machine-readable metadata and then use that to build all sorts of amazing castles in the sky!"
"But what about when someone misuses this metadata to, for instance, get listed in categories they don't belong to, lie about content to reach more viewers, etc etc?"
Let the handwaving commence...
(Yes, various interesting things have fallen out of the research, but the overall drive always seemed naive)
That might be the case for the internet, but in a company there can be real dedicated curation resources and a policy-driven schema and naming convention. I think the viability is different depending on the use case.
That said, I'm not sure universal exchange formats like OWL are really necessary for any use case. Maybe RDF is even overkill, but the concept of subject-predicate-object triples is powerful and opens up a nice space for using rules engines to explore amorphous and multi-dimensional graphs of data.
Also I think the line between data and metadata is pretty blurry and not so easy to separate. At what point is a property of an entity considered "meta"? I'd argue that if the thing it is on can be considered an entity at all, that collecting data about it is just part of the concrete definition.
It's interesting to note that in the case of books and publishing, content creation (authorship/publishing) and metadata classification (libraries, largely) are two separate domains conducted by entirely separate entities.
The separation isn't total --- publishers will typically indicate the genre or category of a work, for example. But even here there's generally a divide between authorship and editorialiship / publishing.
A strength and weakenss of the blogs and other individual authored-and-published works is that there is no such divide, and much incentive to misrepresent, without internal checks.
It is not that the semantic community was not warned - philosophers, linguists, archivists, librarians, historians, we talked and talked to our IT counterparts but could not get through - too much love for technology, not enough study of humans... So it is interesting to read as someone who is interested in archival sciences and history...
the woes of the archive, the problems of wrongly categorized stuff, perhaps even due to political reasons (famously Diderot's encyclopedia in the 18th century linked from "eucharistie" to cannibal, showing how political linking can be) are as old as people trying to sort, use and abuse archives - would be great, if some dialogue might have created a middle ground between dumb full text vs semantic search... with some magic AI put over it ;)
I know, I know :) it was more as a reference on how the pendulum swings in this field as well...
But even there: Luhmann's "Zettelkasten" externalises the main data processing effort back to the users conscious and unconscious references of recorded information, giving back associations not conciously realised before - all with paper cards... would have been another good path not really followed...
The Zettelkasten in particular requires an incredible degree of conscientiousness.
I've found that having a whiteboard in my office helps me jot down quick mathematical ideas as soon as they come (but note how simpler this is due to the home office wave), but that doesn't work well for fleeting thoughts of a more philosophical or conceptual nature.
There's a lot of innovation going on for conscientious people to have a better UI to keep notes, but really all innovation can happen post hoc if you have a text editor and devise a quick text language for linking and metadata (maybe based off something like yaml to cut down on parser writing). The open problem is how to stop losing the fleeting thoughts of smart distracted people.
> that doesn't work well for fleeting thoughts of a more philosophical or conceptual nature
I don't understand. If you're at a keyboard then just type, and if not, just write. Don't include much (if any) refinement, as you can do that at a later stage. Just get down as many associations as you can.
I fear Tim Berners-Lee's promising Solid project is already DOA because of the team's shoehorning this type of semantic web stuff into it - articles like this won't sway the true believers.
It won't work if you demand general consensus, but it can work at low scale, and your friends group is probably pretty low scale. Once you start using friends of friends etc. you have a fairly powerful network. It's relatively easy to get rid of spammers and bad actors.
It's antithetical to what big tech wants to do though, which is control how you interact (e.g. compare the earlier Facebook to the ad-ridden mess now).
The bottom line is: if you're judging all possible social data interactions by the ad-driven approach now, you're bound to be narrow-minded.
2.5 - There's no need for a hierarchy. You can have optional matches: if a match doesn't have a tag, include it anyway, yet if it does have a tag and the value is not what you want, exclude it. The bigger problem nowadays is with people refusing to add '+' and '-' to their site's search function.
2.3 - When a user is adding tags they should be able to see that something has low matches and get a warning. Autocomplete should also help a lot.
2.1 You need to give more power to users to block people categorically and add reasons, and share their blocklists with their friends.
Anyone on a blocklist can be ignored or they can have a big warning placed next to them about what's wrong with them according to your friend.
That is what local communities already do. If Facebook will never include that functionality, then fine, but it is possible.
Also, if you are a liar (unwittingly or not), you should be directed to other liars just like yourself.
2.7 - You didn't even try. It's about as lazy as saying everything is political. You don't even have to use the word 'art'. You can say 'oil on canvas' etc. You don't have to say 'cartoon' when '2D animation' is also good (you can exclude 'non-fiction' etc.). I find it hard to believe that people are oblivious to loaded language.
24 comments
[ 0.25 ms ] story [ 59.9 ms ] threadMetacrap: Putting the torch to seven straw-men of the meta-utopia (2001) - https://news.ycombinator.com/item?id=8511553 - Oct 2014 (3 comments)
Could it be that over-confidence in metadata was common around this time (turn of the millennium), then everyone realized the limitations of meta-data as illustrated in the article, subsequently dropping meta-data approaches altogether, forgetting that it is also useful.
So now we have a situation where the pendulum is now at the other extreme, where machine-generated classification barely intersects with human-generated meta-data at all.
The sweet spot seems to lie somewhere in the middle, as Aristotle used to say.
What do you mean by this? A big part of the reason that deep learning was a major advancement is the fact that it helps you extract data from metadata in a semi-automated (and increasingly automated) fashion.
Maybe the difference is that we gave up trying to "make sense of" the metadata in any kind of universal way.
And to be fair, most people who lovingly curated special knowledge databases only to see them get used to train machines (without any acknowledgement) discouraged them to continue and resulted in a drop of good training sources (at least in my field and non-english language)...
The problem is that human metadata has a pretty solid barrier, but the barriers of what machine metadata can do keeps shifting due to increases in technology, therefore people keep investing in machine processes because they do not yet seem to have reached their limit.
If our metadata are to hold greater than archaeological value, they need to come with a maintenance plan.
Was always my reaction when hearing about efforts like The Semantic Web.
"Yeah, we can get everyone to classify and tag their data with machine-readable metadata and then use that to build all sorts of amazing castles in the sky!"
"But what about when someone misuses this metadata to, for instance, get listed in categories they don't belong to, lie about content to reach more viewers, etc etc?"
Let the handwaving commence...
(Yes, various interesting things have fallen out of the research, but the overall drive always seemed naive)
That said, I'm not sure universal exchange formats like OWL are really necessary for any use case. Maybe RDF is even overkill, but the concept of subject-predicate-object triples is powerful and opens up a nice space for using rules engines to explore amorphous and multi-dimensional graphs of data.
Also I think the line between data and metadata is pretty blurry and not so easy to separate. At what point is a property of an entity considered "meta"? I'd argue that if the thing it is on can be considered an entity at all, that collecting data about it is just part of the concrete definition.
The separation isn't total --- publishers will typically indicate the genre or category of a work, for example. But even here there's generally a divide between authorship and editorialiship / publishing.
A strength and weakenss of the blogs and other individual authored-and-published works is that there is no such divide, and much incentive to misrepresent, without internal checks.
Anyone on a blocklist can be ignored or they can have a big warning placed next to them about what's wrong with them according to your friend.
That is what local communities already do. If Facebook will never include that functionality, then fine, but it is possible.
Also, if you are a liar (unwittingly or not), you should be directed to other liars just like yourself.
the woes of the archive, the problems of wrongly categorized stuff, perhaps even due to political reasons (famously Diderot's encyclopedia in the 18th century linked from "eucharistie" to cannibal, showing how political linking can be) are as old as people trying to sort, use and abuse archives - would be great, if some dialogue might have created a middle ground between dumb full text vs semantic search... with some magic AI put over it ;)
augmented human intelligence ++
authority blankets with massive data backing -1000
But even there: Luhmann's "Zettelkasten" externalises the main data processing effort back to the users conscious and unconscious references of recorded information, giving back associations not conciously realised before - all with paper cards... would have been another good path not really followed...
I've found that having a whiteboard in my office helps me jot down quick mathematical ideas as soon as they come (but note how simpler this is due to the home office wave), but that doesn't work well for fleeting thoughts of a more philosophical or conceptual nature.
There's a lot of innovation going on for conscientious people to have a better UI to keep notes, but really all innovation can happen post hoc if you have a text editor and devise a quick text language for linking and metadata (maybe based off something like yaml to cut down on parser writing). The open problem is how to stop losing the fleeting thoughts of smart distracted people.
It's a big schlep, cue in Paul Graham essay.
I don't understand. If you're at a keyboard then just type, and if not, just write. Don't include much (if any) refinement, as you can do that at a later stage. Just get down as many associations as you can.
It's antithetical to what big tech wants to do though, which is control how you interact (e.g. compare the earlier Facebook to the ad-ridden mess now).
The bottom line is: if you're judging all possible social data interactions by the ad-driven approach now, you're bound to be narrow-minded.
2.3 - When a user is adding tags they should be able to see that something has low matches and get a warning. Autocomplete should also help a lot.
2.1 You need to give more power to users to block people categorically and add reasons, and share their blocklists with their friends. Anyone on a blocklist can be ignored or they can have a big warning placed next to them about what's wrong with them according to your friend.
That is what local communities already do. If Facebook will never include that functionality, then fine, but it is possible.
Also, if you are a liar (unwittingly or not), you should be directed to other liars just like yourself.
2.7 - You didn't even try. It's about as lazy as saying everything is political. You don't even have to use the word 'art'. You can say 'oil on canvas' etc. You don't have to say 'cartoon' when '2D animation' is also good (you can exclude 'non-fiction' etc.). I find it hard to believe that people are oblivious to loaded language.