I believe Wiley's definition should be modified slightly: “any digital resource that can be reused AND supports learning." This adds Gibson's concept of an instructional event to the LO, eliminates single images as LOs (they are information objects not learning objects) and reduces the granularity "problem" while increasing utility (as Merrill recently observed, "If everything is a LO, nothing is."). While LOs are not necessarily objects in the software sense, in the OO world (of C++), objects require among other attributes discoverability (LOs metatags), runtime polymorphism (LOs adaptibility and the extremely attractive possibilities afforded by collaborative filtering and recommender systems--which could be treated as 3rd party meta-data), and protected encapsulation (with objects, all you can do is send a message, and all communication with objects occurs through its message interface). This latter attribute requires that LOs "talk" which necessitates an assessment feedback loop. Chuck Barritt's model at Cisco, while somewhat rigid, seems practical.

I too think the OO metaphor could be taken further with LOs. I'd like to add the concepts of classes and inheritance to Michael's list. In the programming world, what makes objects useful is the fact that they package both data and functionality together. So far, we've only used the data portion of the object paradigm for LOs. Abdul Motaleb el-Siddik's book on Interactive Multimedia Learning does a good job of exposing SCORM's weaknesses in this area, as well as workarounds.

The implications of the model for instructional design are profound. ID needs to let go of control, to some extent, and design in such a way that learners find natural and useful paths through the LOs available to them. Sequencing is going to be problematic in this sense if it's coming from the object/sender as you say. In any case, thanks for the coverage of this important topic.

In talking about learning objects, we have to separate presentation from content. Learning objects are presentational templates (content packages) that can be re-used. How often can content be re-used (especially if we agree that effective content is context-specific)?

When talking about content re-use it would be a good idea to maybe look at a slightly diffrent model ...as David Suggested --separate the content from the presentation.
This can be done by keeping the content in XML format, this would make it deployable in any template.
As far a context of use ...well content usually is context specefic, but there are areas especially software training scenarious where content can be resued effectively.

In the discussion of reuse, I believe one point is being missed. Although reusing finalized object is conceivable, it is quite unlikely, at least according to some of the discussions (and I tend to agree with that).
What we do find is "design reuse", in other words, certain recurring structures that have proven useful in some context. The OO community discusses Design Patterns as being a way to to foster reuse, at various levels, ranging from abstract design to implementation.
I believe Learning Object could be most beneficial if actually understood more as design patterns than as finalized components.

Learning objects are fundamentally flawed. I think we should consider the position of the learner in a more holistic way. Learning is about making connections. People are essentially creatures who like narrative, who like connections, who make sense of their lives from what happened before and by make predictions about what might happen next. They like the flow of learning, the start, the finish and the middle, not a disjointed flight from island to island of information whilst floating about in object land. Objects allow learners or their agents to only choose the learning that they want or need at the right level, which seems to come out of the car production school of education or just in time (JIT) manufacturing. Think back into the midsts of time and consider how you learned things, also consider how you like to learn things. Was it in bite sized manageable chunks or
was there more to it than that? Does anybody learn that way? It’s an extremely Fordist notion that knowledge can be broken down into just enough for a particular task, just enough for a particular skill. It defies logic in the knowledge age society that learning should be so focused that particular items can be segregated out to teach particular skills and that there is no connection between items. Surely the span of a learning object at 15 minutes doesn’t allow enough time for a combination of teaching and learning styles to be addressed so that students can become multi skilled, so that they can both interpret information and apply it so that they can acquire, develop and demonstrate skills and then transfer them to different situations.