View Single Post

Thanks for breathing some fresh insights into the conversation. I'm happy to read something besides "me too" on the subject. (BTW, I wasn't guessing philosophy professor. I was guessing that you might be a fellow Computer Science professor, perhaps one studying semantic web issues.)

I think its useful to separate your idea of associating actions with multiple identifiers (park, exercise, dog) from the idea of "contexts". A context has a very specific meaning in the GTD parlance and in OF, so using that term brings in lots of ideas about how contexts should work in a GTD system. OF strives to be both a GTD system and more flexible than just a GTD system. Using a term besides contexts helps to separate the issues. You've objected to the term "tags", so I'm going to call your proposal "semantically meaningful identifiers", or SMIs.

Logically, SMIs are at least as expressive as the current implementation of contexts in OF. This is because we could implement OF's contexts using SMIs. Just associate a unique SMI with each context in the context hierarchy. An action in a particular context would get the SMI for that context, plus the SMI for every parent context up the hierarchy. For example, my "Call Tree Service" action might get the "Phone" SMI and the "Phone:BusinessHours" SMI. Viewing a single context in the current implementation would be accomplished by showing all actions with the corresponding SMI. Viewing several contexts in the current implementation would be accomplished by showing actions that have at least one of the corresponding SMIs.

But there's more. SMIs are strictly more expressive than the current implementation of contexts. Actions could be assigned multiple SMIs unrelated hierarchically. For example, I could have separate groups of SMIs corresponding to different people, different places, or different times of day. Views in a SMI-based system could use boolean combinations of SMIs, like (Alice AND Bob AND Carol) OR StaffMeeting.

So, if SMIs have so much going for them, why not dump contexts and just use SMIs? A few possible reasons come to mind:
  • Consistently and accurately assigning SMIs to every item is a daunting task. Tagging works for very large data sets, like Flickr, where an infinite number of monkeys are assigning tags and approximate results are sufficient. (Our brains also return approximate search results.) However, when the goal is to show the complete set of actions that could be done at the moment, much greater accuracy and consistency is needed when assigning the SMIs. This makes capturing actions and organizing them a more costly endeavor.

    Bringing GTD back into the discussion for a moment, the Mind Like Water state comes from having everything in a trusted system. If the system instead replicates our brain's capacity for storing lots of stuff but providing approximate search results, then we're back to having incomplete actions and projects intruding on our consciousness.

    Of course, this argument against SMIs is mostly just a straw horse. One doesn't have to create a complex system of SMIs. So, it really comes down to whether the system should guide users or give them all the rope they need to hang themselves. In this case, I'd argue that OF is already in the rope dispensing business. Massive power and a steeper learning curve seems to be OF's niche.

  • Could we get there from here? Substantial portions of the OF architecture seem to be designed around having a single context for each item. (Think quick entry, context mode, context column in planning mode, group by context, and context navigation on the iPhone, just to name a few areas.) Would the additional benefits of SMIs be sufficient to justify the redesign required to replace contexts with SMIs?

    Leaving contexts as is (as are?) and treating SMIs as additional meta-data works around these issues to some extent, but still requires substantial UI and database work on both the Mac and iPhone applications. It sounds like the Omnicians are working on this. I can imagine how this might look on the Mac. Clean querying of meta-data on the iPhone is going to require some innovative UI design. I look forward to seeing the results. (And never mind the computational complexity of querying set-based fields as opposed to singletons.)

  • While philosophically fascinating, are the additional benefits of SMIs practically compelling? Many of the people arguing against multiple contexts on this thread are arguing this point. It's hard for me to say without doing the experiment, but then I'm not paying (much of) the developers salaries. I do know that I mostly gave up on complex tagging in Mail, but I still use three tags there (Personal, Work, and Quicken) and often assign multiple tags to a single message.

  • And finally, SMI is a really dumb name, but I'm sure the Marketing Weasels could come up with something better. Perhaps "tag" or "context". ;-)

(Sorry for the post length. I really should start that blog I've been thinking about.)