Wired spoke to a 15-year-old in Pakistan who said he earns money after school by logging into a site where he’s paid to help train data sets used for AI.
He earned $1 to $2 an hour on a site that uses crowdsourcing to train algorithms, Wired reports.
The story goes on:
Some tasks involve content moderation — helping AI distinguish between innocent content and that which contains violence, hate speech, or adult imagery … he was shown pages upon pages of partially naked bodies, including sexualized images, lingerie ads, an exposed sculpture, and even a nude body from a Renaissance-style painting. The task? Decipher the adult from the benign, to help the algorithm distinguish between salacious and permissible torsos.
This kind of gig work is often performed by workers in developing countries, including ones in East Africa, and in Pakistan, Venezuela, or the Philippines.
Many of the companies that use gig workers for training data reportedly can be lax with age verification — some simply ask the workers to check a box agreeing they’re over 18; others aren’t too stringent about rules being worked around. (Wired gave an example of one teenager who it said he used his grandmother’s face to pass an age-verification test.)
There’s been growing concern about the humans doing the labor behind training AI. Workers in Kenya who helped train ChatGPT said that they were exposed to horrific material, including text centering on sexual violence.
And Meta ended its contract with one outsourcing firm shortly after the reports.
The potential exploitation of labor in the global south to fuel the technical innovation of the West has been an ongoing theme for the last 20 years. Content moderation for social media has long been outsourced to India and the Philippines and elsewhere — potentially exposing workers to horrific images and traumatic material.
But Wired’s report that some of this work is apparently being done by young teenagers is particularly alarming. It’s worth a read.
Read the full article here