I enjoyed the overall stories on their actual experience.
While the quality control paradigm in mass production aimed replaceability of workers, their knowledge management seems trying to make use of asymmetry and unreplaceability of workers and to provide a focal point namely organizational memory.
Nothing was not intersting, but the presence and roles of AI technology is not yet very clear although the authors often claims "AI technologies must work for..."
I understand that the technological pieces in their roadmap, such as agents and natural language processing, will require further AI researches especially on retrieving user's contexts, but do they really require the "Artificial Intelligence"?
What real AI outcomes did contribute to their current "Knowledge Hub"?
They presented two points.
The first one is about the operations to make the current organizational memory properly work in business process management sense.
The second is technological future extents of knowledge management systems to support continuous learning activities.
Knowledge management includes communication tools beyond geological and temporal constraints and business process to make use of the communication tools.
It virtually provides aggrigation of labors in actually disaggrigated workplaces.
Proper practice of knowledge management requires a lot of information management efforts.
First of all, I don't think a pure "push" technology really exists.
I would rather say "push" and "pull" is not a binary choice, but has to be mixed and evaluated with other axes including subscription, context-sensitivity and representation.
A user of "push" technology anyway has to "subscribe" information delivery, and it can be considered "pull" technology in longer term.
It will be harder to choose right criterion to filter/subscribe information for "push" technology because interaction to adjust the criterion will be looser than "pull" technology.
I think context-sensitivity is interesting aspect of information delivery; "Biff" is one of the most successful application of "push" technology and it does not watch user's activities but just polls mailbox.
The point is that "Biff" notifies a user existence of newly arrived mails even when the user is doing nothing, while CodeBroker requires user's inputs into emacs regardless whether the inputs are intended to CodeBroker or not.
Representation includes how much attraction for user is intended and how much information is provided, for example, existence of information, an access point to information with a brief explanation or a full description.
I think this is very important because representation of information can define interchanging crosspoint between push and pull accessibility.
For example, CodeBroker pushes access points to documentation of functions in a repository, according to user model collected in pull-styled interaction with users on the access points.
I think this is a very core part of knowledge management with temporally distributed cognition.
Writing personal memos and reading them have similar difficulties in maintenance efforts with corporation knowledge managements.
One thing is that potential "innovation" is not pushed.
The second thing is representation issue; it is difficult to distinguish the "innovation" from junks.
In other words, because everything can be "innovation", the difficulty is finding out an interesting phenomenon and evaluating the phenomenon to be "innovative" or junk in my current concerns.