MIT AI: Haystack

[nfotxn]

Haystack

1. Genericity. Haystack incorporates and exposes all types of information in a single, coherent manner. It provides a single, uniform interface for manipulation of e-mail, instant messages, addresses, web pages, documents, news, bibliographies, annotations, music, images, etc.
2. Flexibility. The data types Haystack understands are not hard-wired; any additional types of information that a user wants to work with can be easily incorporated. The user can readily define new object attributes that help them categorize and retrieve information, and new relationships between objects. Rather than being tacked-on afterthoughts, user-defined attributes and relationships are given the same centrality in the interface as built-in relationships such as “author” and “date.”
3. User-Object Orientedness. Haystack breaks down the artificial barriers created by giving distinct applications responsibility for different data types. Instead, Haystack attempts to match a user’s own focus on objects in view and what can be done with them. An operation (such as spellchecking, sending an e-mail message, or rotating an image) can be invoked at any time on any object for which the operation “makes sense” (i.e. a blob of text, a person, or an image respectively). New operations can easily be downloaded into the system and immediately become visible in all contexts where they apply. Operations are themselves first-class information objects in Haystack; like all other objects they can be categorized, annotated, and searched for.

I have to admit I find this concept fascinating.

Comments

[tcbear.livejournal.com]

This is the proposal for a project that will happen at MIT..or is it currently under development..and possibly for release soon?

[theotherqpc.livejournal.com]

there's an early release of it (see the upper-right corner of the page for a downloads link), but it seems pretty heavy - it requires a *minimum* or Pentium III, for example

Re:

[tcbear.livejournal.com]

ahhh...wasn't paying attention...thanks :0)

My computer is way way too slow to run that sucker...it makes AIM look like its churning out data.

[dizzi-d.livejournal.com]

It's kind of what Apple has been trying to implement for years with their "system services" initiatives. They wanted applications to make their capabilities available to all other applications that shared a compatible context for usage.

Unfortunately, all the application manufacturers only wanted to enhance the capabilities of their own products.

*glares at adobe*

[soul-spider.livejournal.com]

I was just on the Haystack site. I'm just wondering if it's the best way for *me* to manage my various data streams. Hrm.

Thanks /.

[bigrock.livejournal.com]

Yet another reason to come to Boston :)

three observations

[ranger1.livejournal.com]

1. The project appears to be associated with the Lab for Computer Science (LCS), not the AI Lab.

2. Jeez, that's a confusing name. At first I thought it was the homepage for the MIT Haystack Observatory. Damn LCS twits. If they're so brilliant why can't they come up with an institutionally-unique moniker for their project?

3. As for the software, the theory sounds akin to what Apple was trying to do with OpenDoc (http://www.everything2.com/index.pl?node_id=511958) in the mid-90's. That is, before they learned the hard way that despite the enthusaism of black-turtleneck-clad, Perrier-sipping Apple employees and their syncophants, everyone else (and especially developers (http://developer.apple.com/techpubs/macos8/Legacy/OpenDoc/opendoc.html)) would rather have a paint-thinner enema than use OpenDoc. Hopefully these people will have better luck with their approach. They couldn't do much worse than Apple.

Genericity and Usability

[that-dang-otter.livejournal.com]

I have spent much of my life trying to create "generic" applications for managing data, and unfortunately I have come to the conclusion that the idea is fundamentally flawed. Real-world data types have far too much "nuance" in everyday use for generic interfaces to ever become broadly practical (i.e. in consumer applications). Abstraction at the level of Haystack invariably loses the nuance, making the applications too awkward to use.

Furthermore, any app that requires users to understand the relational structure of their data is pretty much doomed. 98% of computer users don't actually understand the conceptual model behind the applications they use; they merely muddle through. As unbelievable as it may seem to software engineers, even simple notions like hierarchical file systems are beyond the grasp of the majority of users. An app that does not support muddling is therefore not really deployable.

Am I just a bitter pessimist? Nah... I'm still trying. But it's going to be a long, long time before the principles espoused in this product come to fruition. It's a vastly harder problem than most people realize.