Reacting to the Meta data labeling reassignment story as a data bargain problem rather than only a status story.
The story about Meta engineers being reassigned to data labelling
deserves less of a jokey/dismissive response and more seriousness. On
the whole, the AI field and society really benefit from the creation of
the kind of data this program seems to have been set up to produce.
Most people in the AI space would agree that this kind of data
labelling is indeed very valuable for AI progress. If given an
option to wave a magic wand and conjure such a dataset for their
organization (or even into a commons), many would do so.
Similarly, purely from the perspective of a "data pilled"
capabilities researcher, if one could magically turn on trace logging
across an organization of knowledge workers and get access to the
traces, or add cameras to physical workplaces, etc., they would almost
always do so. A very committed AI researcher who was asked by their CEO
to support a decision to reassign some SWEs to do data labelling might
do so.
And of course, even though it seems Meta had to roll this back, much
of the AI progress that so many people remain very excited about and
committed to is still being driven in large part by this exact type of
labor happening opaquely. This story caught attention because of the
theme of potential loss of status for tech workers, but people are
taking the wrong lesson from it.
We actually do want the kind of data being ostensibly produced by the
reassigned Meta SWEs. It's just the incentives, labor conditions, and
messaging that were off base. Ideally, we all stand to benefit from
finding a way, whether that's policy, new types of organizations, or
other approaches, so people -- including Meta SWEs and people doing this
right now as gig work -- can do this with appropriate reward, credit,
and conditions.
This is why I continue to think that one of the grand challenges for
computing and AI is to figure out the balance for a "new bargain" that
produces more data along these lines but does so with greater reward and
concrete power for those doing the generation.
In this case, it might have boiled down to just having better
messaging of the rollout, engaging with worker representatives first,
etc.
We should find a way to make this work dignified and sustainable,
because barring serious scaling roadblocks (quite possible: politics,
physical constraints, and many more), it's likely that more
organizations will be pressured to adopt an "assigned to the data
labeling team" strategy. Basically, expect to see more of this.
Source and AT Protocol record
Source path
content/writing/posts/2026-06-20-data-labeling-work-should-be-dignified-not-dismissed.md
Local AT Protocol-shaped preview used to inspect the record before an exact public cache is refreshed.
{
"note": "Local AT Protocol-shaped preview. Run `make garden-refresh-atproto` to cache exact public records where available.",
"sourcePath": "content/writing/posts/2026-06-20-data-labeling-work-should-be-dignified-not-dismissed.md",
"value": {
"$type": "site.standard.document",
"title": "Data labeling work should be dignified, not dismissed",
"description": "Reacting to the Meta data labeling reassignment story as a data bargain problem rather than only a status story.",
"publishedAt": "2026-06-20",
"site": "at://did:plc:doxvahqvyhyqf32v7wz7p5xk/site.standard.publication/3lzrsw2kvwc2m",
"content": {
"$type": "at.markpub.markdown",
"text": "The story about Meta engineers being reassigned to data labelling deserves less of a jokey/dismissive response and more seriousness. On the whole, the AI field and society really benefit from the creation of the kind of data this program seems to have been set up to produce.\n\nMost people in the AI space would agree that this kind of data labelling is indeed *very* valuable for AI progress. If given an option to wave a magic wand and conjure such a dataset for their organization (or even into a commons), many would do so.\n\nSimilarly, purely from the perspective of a \"data pilled\" capabilities researcher, if one could magically turn on trace logging across an organization of knowledge workers and get access to the traces, or add cameras to physical workplaces, etc., they would almost always do so. A very committed AI researcher who was asked by their CEO to support a decision to reassign some SWEs to do data labelling might do so.\n\nAnd of course, even though it seems Meta had to roll this back, much of the AI progress that so many people remain very excited about and committed to is still being driven in large part by this exact type of labor happening opaquely. This story caught attention because of the theme of potential loss of status for tech workers, but people are taking the wrong lesson from it.\n\nWe actually do want the kind of data being ostensibly produced by the reassigned Meta SWEs. It's just the incentives, labor conditions, and messaging that were off base. Ideally, we all stand to benefit from finding a way, whether that's policy, new types of organizations, or other approaches, so people -- including Meta SWEs and people doing this right now as gig work -- can do this with appropriate reward, credit, and conditions.\n\nThis is why I continue to think that one of the grand challenges for computing and AI is to figure out the balance for a \"new bargain\" that produces more data along these lines but does so with greater reward and concrete power for those doing the generation.\n\nIn this case, it might have boiled down to just having better messaging of the rollout, engaging with worker representatives first, etc.\n\nWe should find a way to make this work dignified and sustainable, because barring serious scaling roadblocks (quite possible: politics, physical constraints, and many more), it's likely that more organizations will be pressured to adopt an \"assigned to the data labeling team\" strategy. Basically, expect to see more of this.\n"
}
}
}