How writing creates value
alexkrupp.typepad.com
alexkrupp.typepad.com
For example, you could look at the top 100 hacker news links each month and get a large amount of insight/interesting stories/. I think people enjoy browsing, discovering and voting -- I'm not sure I'd really want a machine that just gave me the most interesting links (hn services as that, with the added gem of perhaps discovering something off the beaten path).
I did like the part about focusing on insight, and calling that out as a specific trait. I find that is one of the few transferrable skills -- facts, while interesting, are fragile; ways of thinking about the world can be more permanent.
Being right 100% of the time would require some AI that's way more advanced than anything existing today, but the thing is that you really only need to be right 25% of the time to create something that's an order of magnitude better than what exists today. For example, let's say you have this article: http://www.alternet.org/drugs/21673/
Being able to identify a set of people, 25% of whom would find that article worth reading, shouldn't be very difficult. In addition to all of the rating systems you mention, you can also do some basic things like tracking their reading history, asking them some questions before and/or after presenting them with article recommendations, etc.
In fact, the two things I would most go out of my way to avoid would be A) trying to figure out what people will like based on what other people with similar taste like and B) pulling specific facts out of articles and trying to analyze them or compare them with other articles. These two approaches seem like incredibly hard problems to solve, and they don't seem at all necessary to create something that's an order of magnitude better than what we have today.
The trick is getting humans to do as much of the work as possible. For example, one could create a delicious-like tagging system whereby people would tag articles with other articles they should read if they liked/disliked or agreed with or disagreed with the original article. Once you have a shared vocabulary to talk about the problem, getting a good solution isn't that hard even if not all the tools are 100% formalized.
A is a solved problem. Heck, there's an API for it even: http://directededge.com/
To me, part of the appeal to these sites are that links and content are driven by the users, and not by what someone else wants me to read.
I'm sure this could highlight interesting stories, but I think it starts to break apart the community these social networking sites provide.
Overall, I agree with you. It should be possible to define some objective criteria about what is interesting, and then some algorithm could check all the new articles from mainstream dailies, weeklies and monthlies, and return let's say up to 150 articles monthly which should be interesting. That I think could be an useful service.
For example, Vanity Fair has 1 or 2 articles monthly which are interesting, but I don't have time to read VF every month to see if I discover something good...if some algo can do that pre-selection, that would help.
Anyway, CDT attempts to extract structured information from text and I thought that if CDT worked, then you could measure something Alex talked about in the article: rating text as interesting if it introduced novel concepts.
I like the emphasis behind the article, but we need to get off this idea that writing is the thing, when really its ideas that matter.