This analysis seems to be suffering from fixed world fallacy. "AIs have been created; they have some capabilities; now they are going to find their place in the world of work."
We see this fallacy also in discussions of adaptation to warming-induced sea level rise. Sea level rise is not "two feet by 2100", with "stopping there, so we can adapt", left unsaid but implied by that phrasing. It is 20 mm a year by 2080, rising rapidly to 50 mm a year, forever, practically, on human scales. There is no fixed state to adapt to.
So with AI. There is no fixed state to adapt to, and there is little reason to think there will be one any time soon. At some point AI will be changing work faster than we can re-organise bundles of tasks, and it will *keep on* changing.
Soon, all of work will be restructuring org charts, and the restructuring will not be able to keep up.
1. The employee can focus on the 10%, thus, maybe, better quality.
2. Only 10% of employees are needed, thus unemployment.
3. The employee, and maybe the company, can 10x their output, making the employee and/or the company much more valuable. Thus, more profits and profit-sharing.
I enjoyed this by there's one tension running through the argument that I'd like to highlight.
The arguement needs 'task' to be a stable, countable unit for exposure scores to mean anything, but the mechanisms you describe (bundling, focus effects, shifting expertise thresholds) all suggest task boundaries are drawn by the surrounding organization of work, not intrinsic to the activity.
Hutchins's distributed cognition and Sennett's The Craftsman both make this case from different angles—worth a look if you push this further, since it might change what an 'exposure measure' can claim to measure.
This analysis seems to be suffering from fixed world fallacy. "AIs have been created; they have some capabilities; now they are going to find their place in the world of work."
We see this fallacy also in discussions of adaptation to warming-induced sea level rise. Sea level rise is not "two feet by 2100", with "stopping there, so we can adapt", left unsaid but implied by that phrasing. It is 20 mm a year by 2080, rising rapidly to 50 mm a year, forever, practically, on human scales. There is no fixed state to adapt to.
So with AI. There is no fixed state to adapt to, and there is little reason to think there will be one any time soon. At some point AI will be changing work faster than we can re-organise bundles of tasks, and it will *keep on* changing.
Soon, all of work will be restructuring org charts, and the restructuring will not be able to keep up.
If you automate 90% of a job:
1. The employee can focus on the 10%, thus, maybe, better quality.
2. Only 10% of employees are needed, thus unemployment.
3. The employee, and maybe the company, can 10x their output, making the employee and/or the company much more valuable. Thus, more profits and profit-sharing.
Did you miss the last one?
I enjoyed this by there's one tension running through the argument that I'd like to highlight.
The arguement needs 'task' to be a stable, countable unit for exposure scores to mean anything, but the mechanisms you describe (bundling, focus effects, shifting expertise thresholds) all suggest task boundaries are drawn by the surrounding organization of work, not intrinsic to the activity.
Hutchins's distributed cognition and Sennett's The Craftsman both make this case from different angles—worth a look if you push this further, since it might change what an 'exposure measure' can claim to measure.