84,181
edits
Template:Data as a self-fulfilling prophecy: Difference between revisions
Jump to navigation
Jump to search
Template:Data as a self-fulfilling prophecy (view source)
Revision as of 20:05, 23 October 2023
, 23 October 2023no edit summary
Amwelladmin (talk | contribs) No edit summary |
Amwelladmin (talk | contribs) No edit summary |
||
| Line 6: | Line 6: | ||
''[[Weisendämmerung]]'': the twilight of the wise. | ''[[Weisendämmerung]]'': the twilight of the wise. | ||
Wisdom only comes with time, experience and anecdotally-accumulated expertise. It is hard to acquire and expensive to buy. | Wisdom only comes with time, experience and anecdotally-accumulated expertise. It is hard to acquire and expensive to buy. Technology, by contrast requires ''no'' time and ''no'' expertise: only brute information processing capacity — which gets ever cheaper — and enough data to process — which gets ever more abundant. | ||
The more data there is, the more you can process, the more [[neural network|neural networks]] can crawl over it, [[pattern matching]] and framing and analysing and the more “insight” — unexpected, machine-generated insight — we can extract. | |||
====Data is theory-dependent=== | |||
But this insight is a function of the data: we can’t analyse or pattern-match data we don’t have — and what we do have we must select, filter, format, array and frame according to some pre-existing [[Narrative|theory of the game]]. Our data paints a picture from shadows: by blocking out “irrelevant” data we have collected but which doesn’t advance, bear upon or fit our theory. The more data we have, the more of it we must block to make a meaningful model. | |||
So, dilemma: the ''less'' data we have, the stronger the model, but the less reliable the insight, because we don’t know what we’re missing. The ''more'' data, the weaker the model, and the less reliable the insight, because we still don’t know what we’re missing, but the more of what we ''do'' know we have had to rule out to draw a single coherent model. | |||
So, dilemma: the ''less'' data we have, the stronger the model, but the less reliable the insight, because we don’t know what we’re missing. The ''more'' data, the weaker the model, and the less reliable the insight, because we still don’t know what we’re missing, but the more of what we do know we have had to rule out to draw a single coherent model. | |||
We don’t know how experts do what they do — that ineffability is their expertise — until they acquire tools to help them — ''digital'' tools — and they make us lazy, at the same pushing experts towards activities that generate metadata (the tools don’t help with ineffable stuff that doesn’t generate metadata) that the technocrats can collect. (A conversation across the desk is purely analogue; it contains no recordable data or metadata; a typed letter is an analogue artefact with no metadata; a facsimile is a digital graphic of an analogue artefact with limited extractable data or metadata; an electronically transmitted ASCII document is ''only'' data, and has no meaningful analogue existence at all) | We don’t know how experts do what they do — that ineffability is their expertise — until they acquire tools to help them — ''digital'' tools — and they make us lazy, at the same pushing experts towards activities that generate metadata (the tools don’t help with ineffable stuff that doesn’t generate metadata) that the technocrats can collect. (A conversation across the desk is purely analogue; it contains no recordable data or metadata; a typed letter is an analogue artefact with no metadata; a facsimile is a digital graphic of an analogue artefact with limited extractable data or metadata; an electronically transmitted ASCII document is ''only'' data, and has no meaningful analogue existence at all) | ||