Week 11: Machines in the queue

Automated contribution, automated moderation, and generated text arriving faster than volunteers can assess it

17 May 2027

Bots have been the highest-volume editors on Wikipedia since the mid-2000s and nobody objects, because the bots do work no one wants: fixing link rot, standardising date formats, reverting obvious vandalism within seconds. They are approved through a process, they have operators with names, and they can be switched off.

OSM’s version of the argument is about imports. Bulk-loading a government road dataset adds in an afternoon what surveyors would take years to map, at the cost of data nobody on the ground has checked and a local community who were not consulted. The project has been arguing about this, with unusual politeness, since the TIGER road import landed across the United States in 2007.

The new problem is different in kind. Generated prose is cheap, plausible, correctly formatted, and sometimes cites sources that do not exist. Assessment is the scarce resource in every volunteer project, and generated content attacks exactly that: it takes an editor twenty minutes to check what took four seconds to produce.

Which raises a question you should sit with, given how this course is taught and what you will use to draft your own submissions: the asymmetry is not between humans and machines. It is between producing and checking, and it was already there.

Outline#

• approved bots, bot policy, and why high-volume automation is uncontroversial
• imports in OSM: speed against local knowledge
• the assessment bottleneck, quantified
• detection, disclosure and why detection is the weaker of the two
• drafting your Governance Redesign’s position on machine contribution