Your Web News in One Place

Help Webnuz

Referal links:

Sign up for GreenGeeks web hosting
January 24, 2019 04:56 pm PST

Sort by Controversial: training machine learning to sow irreparable divisions

Scott Alexander continues to delight with his works of short, sharp science fiction (previously): this time, it's "Sort by Controversial," a teachnolovecraftian story of training a machine learning system to recognize (and then produce) "controversial" stories by exploiting Reddit's "sort by controversial" feature to obtain training data.

Alexander's mcguffin is something called "Shiri's Scissor," a machine learning system that produces polarizing statements whose deceptive obvious rightness (or, alternately, wrongness) pits people against one another so violently that once you've been scissored, your peace is forever fractured.

It's a lovely tale in the tradition of Lexicon and Snow Crash, turning on the use of algorithms to locate "spells" whose utterances destroy our ability to think clearly -- and as such, it's a wonderful metaphor for the engagement-maximized political climate we find ourselves imprisoned by.

Shiris English wasnt great, so I thought this was a communication problem. I corrected her. The program was spitting out obviously false statements. She stuck to her guns. I still thought she was confused. I walked her through the meanings of the English words true and false. She looked offended. I tried to confirm. She thought this abysmal programming decision, this plan of combining every bad design technique together and making it impossible to ever fix, was the right way to build our codebase? She said it was. Worse, she was confused I didnt think so. She thought this was more or less what we were already doing; it wasnt. She thought that moving away from this would take a total rewrite and make the code much worse.

Read the rest


Original Link: http://feeds.boingboing.net/~r/boingboing/iBag/~3/kBGt8e7Cm1Y/controversy-for-its-own-sak.html

Share this article:    Share on Facebook
View Full Article