Mohamad Dunggio earns extra money by recording himself doing household chores. The 23-year-old from Gorontalo, Indonesia, recently mounted a phone on his head and filmed himself washing dishes repeatedly for four hours. He said the work was worthwhile because he was exchanging his time for income.
Dunggio is among thousands of gig workers producing “egocentric training” videos, which capture tasks from the perspective of the person performing them. He said companies based in Argentina, India and the United States paid him between $US3 and $US10 an hour for videos of dishwashing, cooking, cleaning, ironing, folding clothes and sweeping. Some of his recordings were rejected, and he said a transfer fee of about $US1 could apply when withdrawing less than $US10.
Dana Kulić, a robotics professor at Monash University, said robots needed extensive examples to learn how to respond to the differing layouts, objects and routines found in people’s homes. She compared the use of egocentric footage to the way generative artificial intelligence systems learn patterns in language. Robot-generated data may be more directly relevant, she said, but recordings made by people are cheaper, easier to obtain and offer a wider variety of situations.
The contributors include house cleaners, delivery drivers and workers in factories, according to the report. Sergio Bruccoleri, vice-president of delivery at Appen, said the company can collect about 60,000 hours of video each month from as many as 5,000 contributors worldwide. The footage is gathered in homes, workplaces and controlled studios, sometimes using specialised cameras and sensor gloves.
Bruccoleri said Appen was placing greater emphasis on data from wealthier countries, including the United States and Canada, because robots are likely to be sold there initially and household equipment differs between markets. He said the company had fair-pay and privacy standards, including a system intended to allow contributors to request that their data be withdrawn after sale. He also acknowledged that payment fees were unfair, but said they were charged by payment platforms.
Kulić said the collection of human footage raised questions about whether contributors understood how recordings would be used, privacy inside homes and the possibility of data being sold to third parties. Dunggio said he did not feel exploited and viewed the work as an income option, although he was embarrassed filming himself in public and said the work had also taught him to cook.