Employees walk past Tesla's electric car plant in Gruenheide near Berlin, eastern Germany, on March 13, 2024, as employees resumed work after production had to be halted due to a suspected arson attack that caused a power outage. Far-left activists from the "Vulkangruppe" (Volcano Group) have claimed responsibility for the sabotage, saying they aimed to achieve "the biggest possible blackout of the gigafactory", a reference to the Tesla plant. (Photo by Odd ANDERSEN / AFP) ODD ANDERSEN/AFP via Getty Images
For Tesla's Autopilot software to be at its best, there should be an army of annotators tasked with labeling thousands and thousands of video clips and images taken from cameras on Teslas.
According to Mindy Support, they work largely from facilities in Buffalo, NY; Palo Alto, CA; and Draper, UT. These annotators help train the AI driving Tesla's driver-assist features.
How Tesla Autopilot Annotators Work
The process, however, is grueling and monotonous. As Business Insider reported, employees slave away for hours labeling lane lines, curbs, and other crucial features that allow the system to do precisely what human driving was meant to do.
In the same report, it was found that data annotators are subjected to strict surveillance and rigorous performance monitoring.
At the Buffalo facility, employees are watched by an array of surveillance cameras and tracked through software systems that monitor productivity. One software, called "HuMans," imposes strict time limits on how long workers are supposed to spend on any given clip. Another, called "Flide Time," monitors active time spent with the labeling software.
Per Futurism, falling short from these standards can either result to disciplinary measures or worse-termination. An employee has already walked out after allegedly failing to accurately label a highway exit sign.
However, it's not just the work condition that makes Tesla annotator worse.
According to Business Insider, the footage that annotators tend to work on often reveals very personal moments about Tesla drivers' lives; a big concern to privacy.
In one task, employees labeled videos that were shot inside customers' garages using Tesla's Sentry Mode. In other projects, the workers were tasked with data observation that included monitoring whether drivers were paying attention through in-cabin cameras while using the Autopilot technology.
These tasks have made some employees uncomfortable because they feel like they are intruding on people's privacy.
Reuters shared that Tesla has also faced criticism for how it handles sensitive footage, especially after an incident in 2023 when workers shared a video of a bicycle accident involving a Tesla vehicle. In response, the company tightened access to certain clips and added watermarks to make it easier to track and prevent unauthorized sharing.
Despite these measures, the company still pressures annotators to meet strict performance goals, leaving them balancing between maintaining efficiency and protecting privacy.
Response from Tesla Annotators
Tesla's strict monitoring practices prompted workers at its Buffalo facility to attempt unionizing in early 2023, citing frustration with productivity metrics like keystroke tracking.
As a result, the New York Times learned that Tesla laid off several workers, attributing the cuts to poor performance rather than union activity. The National Labor Relations Board (NLRB) has filed a complaint, though Tesla denies any wrongdoing.
Despite these challenges, Tesla's workforce remains integral to CEO Elon Musk's vision of achieving fully autonomous driving. While the company aims to eventually automate much of the labeling process through AI, human annotators continue to play an important role in training the system.
Join the Conversation