Explaining further to Ars, Reddit spokesperson Tim Rathschmidt said:
There was a small bug in a test we ran that mistakenly caused the English-to-English translation(s) you saw. That bug has been resolved. Unsurprisingly, English-to-English translations are not part of our strategy, as they aren’t necessary. English-to-English translations were not a desired or expected outcome of the test.
Reddit pulled the test it was running, but its machine learning-powered translations are still functioning, Rathschmidt said. The company plans to fix the bug and run its unspecified “test” again.
Reddit’s explanation differs from user theories floating around beforehand, which were mainly that Reddit was rewriting user-created summaries with generative AI, possibly to boost SEO. Some may still be perturbed by the problem persisting for weeks without explanation and the apparent lack of manual checks for the translation service. However, Redditors can now take comfort in knowing that Reddit is not currently using generative AI to alter user-generated content without notice.
Paige_Railstone, however, maintains frustration and wants to tell Reddit admins, “STOP. Hand off.” The translation bug, they noted, led to people posting on a subreddit for parents with autism that their child might be autistic, “and how terrible that would be for them,” Paige_Railstone recalled.
“These are the kind of unintentionally insulting posts that drive autistics into leaving a community, and it increases the workload of us moderators,” they said.
Paige_Railstone also sees the incident as a reason for moderators to be more cautious.
“This never used to be a concern, but this translation service was rolled out without any notification that I’m aware of, and no option to disable it within the mods’ control. That has the potential to cause problems, as we’ve seen over the past two weeks,” they said.
Disclosure: Advance Publications, which owns Ars Technica parent Condé Nast, is the largest shareholder in Reddit.
