meta meta

February 3, 2005 4:31 PM

"Conversation will improve,language barriers will fall, artificial intelligence will begin to emerge, and, hopefully, people will be more honest about what they want and who they want to have sex with." If you could meta-tag a tag itself, would the resulting "tagweb" mirror how we collectively organize thoughts in our [collective] brain? (via del.icio.us)
posted by acid freaking on the kitty (17 comments total)

Extensively discussed here.
posted by vacapinta at 4:51 PM on February 3, 2005

I'm basically a real big fan of your moniker: acid freaking on the kitty.

Interesting links. I hope this thread turns into something crazy enough to warrant [inside joke picture], while still remaining educational.
posted by Kleptophoria! at 4:59 PM on February 3, 2005

Wuh? Did I skim over the Dramatically New Insight^TM or is this essay just the musings of someone who thinks he's (re)discovered graph theory and naively tried to explain how it might apply to machine learning? Zing! I'm not the only one...

I think you've run into the problematic dichotomy between enthusiastic amateurs and highly trained specialists. I'd like to nurse the romantic notion that someone with little background can stumble, by luck or genius, upon simple truths that all the experts have blinded themselves to, but I think it's a myth, and probably always has been.

Though it would be cool to combine an image processing library with the Flickr tag database to create some sort of "What the hell am I looking at?" tool. It'd be like funny-spell-check-suggestions^10²⁰
posted by Loser at 5:14 PM on February 3, 2005

By looking at how we tag photos on Flickr, we can understand how humans process information. Once we understand that, we can understand how to model it with computers, thereby creating better information retrieval systems.

Yum, the sweet smell of pseudo-science!

Yet another analogy that bills itself as a breakthrough understanding of how minds work.

You may recall such previous "breakthroughs" as:

• brain as the arena of competing memes;
• brain as a Darwinian competition or even ecosystem;
• the "neural net": brain as a connectionist weighted graph;
• brain as digital computer;
• the Freudian brain as a steam engine, where "pressures" must be "regulated" by machinery or "blown off";
• or Thomas Hobbes's strings pulled by sensory experience.

It seems every time a new technology comes along, somebody announces that the brain probably more or less works like that.

No doubt there's some truth in all of them (and I wouldn't be surprised if there's much truth in William Calvin's guesses about Darwinian competitions).

But analogy can only take you so far. The usefulness of analogies lies in their looseness: any analogy fits many sets of facts. Once the skin of analogy is draped over the skeleton of facts, it obscures as much as it serves to unite; it provides no falsifiable hypothesis, and so the analogy is more an aid to conjecture than a way to test theories.

We should be particularly suspicious of ebulliently optimistic theories like this which promise to eradicate barriers and even to make people "more honest" (!). That's not science or even disciplined philosophizing, that's Pollyanna gazing at her own navel.
posted by orthogonality at 5:48 PM on February 3, 2005

tags are cool; perhaps we should tag tags for the hell of it.

but tags are not neurons.

I totally do believe in the Flickr effect, though. I love to tag.
posted by blacklite at 6:06 PM on February 3, 2005

You guys totally just ruined pseudo-science for everyone.
posted by Kleptophoria! at 6:07 PM on February 3, 2005

ugh
posted by nj_subgenius at 6:23 PM on February 3, 2005

* the "neural net": brain as a connectionist weighted graph...

...It seems every time a new technology comes along, somebody announces that the brain probably more or less works like that.


I'm surprised you would put neural nets in that list for two reasons: first, the AI uses of neural nets did not precede the actual connectionist model of the brain, but rather grew out of it (see information here). Second, there is a tremendous amount of research on connectionist models of various subparts of the brain, most of which makes claims that are falsifiable (and many that have been falsified). Say what you will about connectionism but I've never seen it treated as just an analogy or pseudoscience, and it's still a live (though debated) option for analyzing lots of things (I only know about a few, and these include many aspects of speech production and language processing).

Sure, many people in the mainstream press talked about neural nets in a fairly ill-informed way back in the early AI days, and many computer scientists were fairly overoptimistic about their success in AI, but connectionism in practice in cognitive science has been much more fruitful than the other things you list, and I really don't see how it can be called a pseudoscience.
posted by advil at 6:24 PM on February 3, 2005

the "neural net": brain as a connectionist weighted graph;

The brain is a neural net, not figuratively, but literaly. A lot of people don't actualy know what neural networks actualy are. Nodes in a neural network definetly do not represent 'concepts' or anything like that, just parts of a formula.

Anyway, this article was just silly. He basicaly goes on how you can have a computer make analogies based on graphs.

He dosn't even have that basic part right, because he dosn't spesify what those relationships actualy mean.

For example, he relates "male" and "female" to person. But he could also relate "me" to person as well, and then you would end up with an analogy like

"this photograph of me is to 'person' as victoria is to person", in other words, his confused little matrix now says that his photograph is a person.

Okay, you say, let's just modify this to include the name of the relationship, thats great. Can't see anything too wrong with this idea, but. it won't do you much good. First of all, what do you want to do with it? Unforunetly, most of the intresting things you can do with such data are NP-complete.
posted by delmoi at 6:56 PM on February 3, 2005

Waitaminit... these tags that you tag things with... you're talking about tagging them? Tagging tags? Crazy idea, man.
posted by soyjoy at 6:56 PM on February 3, 2005

advil replies to my post calling Flickr graphs pseudo-science: "I'm surprised you would put neural nets in that list.... and I really don't see how [connectionism] can be called a pseudoscience."

I may have been somewhat careless in how I wrote my comment: I'm calling the link Flickr Tag as theory of mind business pseudo-science, but I'm not calling all the other theories of mind I listed pseudo-science.

As I wrote about those theories, "No doubt there's some truth in all of them", and that of course applies to connectionism.

My point was not that connectionism was pseudo-science, but that various enthusiasms -- fads -- in the theory of mind come along with regularity, and (as you note) overblown and over-optimistic claims are invariably made.

This is not to say that connectionism doesn't have some real uses, it surely does -- only that it does not form a complete theory of mind, as some of its less careful enthusiasts once claimed.
posted by orthogonality at 6:59 PM on February 3, 2005

In thinking about a bit more, all he's done all he's done is come up with a way to represent knowlage. And it isn't even a very good system.
posted by delmoi at 7:01 PM on February 3, 2005

I just decided that I'm not interested in this, or anything like it.
posted by newton at 9:15 PM on February 3, 2005

No, no, I'm still pretty sure folksonomies are the solution to all our problems.
posted by Hildago at 10:48 PM on February 3, 2005

hopefully, people will be more honest about what they want and who they want to have sex with

I think the Internet has already given us this.
posted by anthill at 11:47 PM on February 3, 2005

Yikes. Didn't he do any background web searches when writing this? How could have missed the immense quantity of scholarly work on graph theory, ontologies, concept maps, semantic networks, information storage and retrieval, memory, AI, neural nets, deduction and inference engines, data-mining, cognition, neuroscience, and so on, and so on, and so on.

And the awful thing is that thread over on flickr. People are telling him all of this stuff, but he doesn't seem to believe them. You don't have to be a professor in the field before we'll look at your idea, but man, you have to do your literature work first...

A skeptical mind starts at home, applied to your own brilliant new ideas. We've been doing this science and math thing for a while; we've sketched out most of the basics.

A handy rule of thumb: any idea you come up with has already been done, and after you read up on that work, your newly refined idea has also been done, and this process loops until you are finally out of papers to read -- then if you're still managing to hold onto a novel idea, you have a little time window before it also will be done by someone else. The size of that time window is some mysterious function of luck, funding, a dash of brilliance, and well-executed academic sabotage.
posted by justin at 2:37 AM on February 4, 2005

He lost me at the second sentence:

You may also notice that tagging photos on Flickr is the first time that organizing something has made perfect sense.

No, it doesn't make perfect sense. Tagging is pretty cool, but it has significant limitations as well. It may make more sense than other forms of metadata, for some collections of documents, but it's not good enough to say it makes perfect sense.

I like tagging, but apparently I missed the day they were handing out the tagging-is-the-best-thing-since-sliced-bread Kool-Aid.
posted by DevilsAdvocate at 7:03 AM on February 4, 2005