Thursday, June 07, 2007

A Simple Definition of Knowledge?

Responding to George Siemens's A Simple Definition of Knowledge. It is currently pending approval over there.

Um... no.

I don't want to be antagonistic, but this account is not satisfactory.

> information is a node which can be connected

So what, then, a neuron is information? No, that makes no sense - because then we would have the same information, unchanged, day in and day out, in our brains.

At the very least - information has to refer to a neural *state*. A nodal *state*. At its simplest, a neuron can be 'off' or 'on' (actual neurons have more complex states, of course). A given neural state might be a bit of information - a sequence of neural states or a collection of neural states 'information'.

Even then, we may want to rstrict our attention to certain states, and not all states. Taking an information-theoretic approach, for example (cf Dretske) we might want to limit our attention to neural states that are reflective of (caused by, representative of) states of affairs in the world. This is the distinction between 'signal' and 'noise'.

There's a lot more to be said here. because now we night want to say that the information isn't the actual state, but rather, it is the (description of, proposition describing, etc) state of affairs represented by the neural state. Because the actual neuron doesn't matter, does it? If we switched the current neuron out for a different one, it would still be the same infomration, wouldn't it?

> When connected, it becomes knowledge (i.e. it possesses some type of context and is situated in relation to other elements).

The traditional definition of knowledge is 'justified true belief'. There are many problems with that definition, but it does point to the fact that we think of 'knowledge' as being something broadly mental and propositional. Knowledge, in other words, is a macro phenomenon, like an entire set of connections, and not a micro phenomenon, like a single connection.

But there's also more at work here. Is knowledge the actual physical se of connections? Is it the pattern represented by the connections, that could be instantiated physically by any number of systems? Is it tantemount to the state of affairs that caused the set of connections to exist? Is the connective state representational? Referential?

Simply saying 'knowledge is a connection' answers none of these. It offers no account of the relation between the brain and the world, if any. It doesn't account for the relation between, say, 'knowledge' and 'belief'. I am sympathetic to the non-representational picture of knowledge suggested by the definition - but if knowledge is non-representational, then what is it? Saying that it's some physical thing, like a connection, is about as useful as saying that it is a brick.

> understanding is an emergent property of the network

Which means... what?

To put this bluntly: is understanding an epistemological state - that is, it it some kind of super-knowledge, perhaps context-aware knowledge? Kind of like wisdom?

Or is it a perceptual state? Is 'understanding' what it *feels* like to know?

Is *any* emergent property of a network an 'understanding'? We could imagine a digital video camera that records the 'face on Mars'. So we have some emergent property of the networks of sensors. Is this emergent property 'understanding'?

One would assume that there would be, at a minimum, some requirement of recognition. That is, it doesn't get to be 'understanding' unless it is 'recognized' as being the face of Jesus on Mars. But this means it's not just the emergent property - it's a relation between some emergent property and some perceptual system.

Additionally - there is not really a face of Jesus on Mars. It's just an illusion. Does it count as 'understanding' if it's an illusion, a mirage, or some other misperception? If not, what process distinguishes some recognitions of emergent properties from others?

I don't mean to be antagonistic here. I am sympathetic with the intent of this post. But it is so far from being an adequate account of these terms it was almost a duty, a responsibility, that I post this correction.

I understand that I owe an alternative account of these phenomena. I have attempted a beginning of such an account in my Connective Knowledge paper. But it si clear to me that I need to offer something that is both significantly clearer and significantly more detailed.


  1. Stephen,

    I have been fascinated and invigorated by a number of your posts over the last few months that have been delving quite deeply into the epistemological elements of learning more generally, but distributed computer mediated networks of learning more specifically. I have been following these conversations closely and they really resonate for me just how much of the educational technology realm needs to be focused precisely on the issues you have been hammering away at -much more so than the notion of tools per se (I am more than guilty of that). Your impassioned commitment to re-thinking education though a methodology inspired by and grounded in many of the core philosophical tenets of knowing is a invaluable rigor that drives this community.

    Thanks for modeling the importance for keeping much of the conversation about educational technology in close proximity to the larger philosophical issues we are constantly struggling with.

  2. It comes to mind that one approach to this definition is to start with the notion of modeling.

    Grey Walter I think made a cogent case for operational models of cognition in "The Living Brain" - where he talks about analogy, metaphor, and physical analogs as both simplifications and modifications of "perceived reality".

    The bridging of the semantic/syntactic dichotomy we inherit from Greek thinking is a big step in doing this kind of meta work. It is ironic that we may need to make a "knowledge object" model to do this.

    It is possible that successively more complicated models may allow us to step out and do the necessary analysis. I suspect that that is why the "node" definition you mention and reject is attractive, it seems manageable and can be modelled and put into equations, akin to the ones defining "information" by Shannon.

    I would propose that if by successive approximation we can build models of "knowledge" and refine them, we would obtain a a workable and useful definition.

    Unfortunately, I have an (nagging) hunch that the model cannot be extricated from its world - knowledge is a mechanism for making as well as interpreting "reality"; it seems to be part of "reality".

    Breaking that tautological feedback loop is a big part of the challenge - we need a way to step out.


I welcome your comments - I'm really sorry about the moderation, but Google's filters are basically ineffective.