In an attempt to obtain more data to train robots to perform dexterous tasks, start-ups like Egolab.AI are focused on collecting “egocentric data” — footage from head mounted cameras. This data is not just used for training but also for surveillance, as the startup makes deals with factories to provide “AI powered efficiency reports” on the workers in exchange for the data obtained by the workers.
The workers in two textile factories in India spoke with openDemocracy about how they fought back against a data collection pilot program. The justifications for the cameras were suspiciously vague and workers investigated and concluded that, contrary to the management’s claims, they were used for AI training. They did not stay quiet about this and continued to complain and strategically remove the helmets to render the footage unusable. The data collection program was ended after only two weeks.
Key takeaways from this story
Collective pushing back–in the form of a formal union or not–can be effective.
The workers shared knowledge about the true purpose of the cameras and encouraged each other to take off the cameras and ruin the footage.
This was occurring during a larger worker movement in the region that involved work stoppages at one of the factories and resulted in a minimum wage increase, which may have impacted the shelving of the data collection program.
- Many industries are desperate for data as they attempt to take advantage of the current trend of scaling.
- Due to the apparent success of LLMs and similar AI technologies, quantity, rather than quality, is the goal. Specialized data, such as data for robotics, is especially hard to come by and they have to get creative with their theft.
- It is part of a larger trend of workers being exploited for their data, without proper compensation, as they train systems that their bosses hope will be able to replace them. This is happening to everyone from these factory workers to software engineers.
- Data often serves a dual purpose of AI training and worker surveillance.
- The idea of training robots that can replace workers is a long term project (if anything ever comes of it at all), but worker surveillance and “efficiency reports” offer an immediate tangible benefit to the factory owners.
- No benefits are offered to the workers who are creating this data and who have to wear an uncomfortable camera strapped to their heads.
- These start-ups target areas with poor worker and data protections to run these experiments for ease of exploitation.
- Piloting this project in Indian factories–where one less than 20% of workers have a written contract–was not a neutral decision.
- The lack of communication (and outright deception) about the purpose of the cameras, the absence of any incentive, and the refusal to listen to any complaints were choices the start-up and the factories made because they felt like they could get away with it.
Image: Kathryn Conrad & Rose Willis / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/