«The New York Times» published an article reporting that these cars «often» needed to be assisted by remote human workers. Offended by this "misinformation", Cruise CEO Kyle Vogt turned to Hacker News, a forum hosted by Y Combinator, a venture-capital-funded incubator, to set the record straight. These cars were not supposed to be remotely driven "often", but between 2 and 4 percent of the time in "difficult situations", which is itself an admission: these cars certainly should not be described as autonomous.
Most AI tools require a large amount of hidden labor to actually function. This enormous amount of work goes far beyond monitoring systems operating in real time and also includes the work of creating the data used to train the systems.
Workers who handle these tasks perform numerous activities. They are required to:
- highlight, with green boxes, the objects present in images from the cameras of self-driving cars;
- assess how illogical, useful, or offensive the responses produced by language models are;
- label social media posts that contain hate speech or threats of violence;
- determine whether people in sexually explicit videos are minors.
These workers deal with a large amount of toxic content. Since machines for the synthetic production of media recombine internet content into plausible-sounding text and understandable images, companies need a screening process to prevent their users from seeing the worst that the internet has to offer.
«Time» reported that OpenAI had outsourced to Kenyan workers paid less than two dollars an hour the removal of graphic content, hate speech, child sexual abuse material, and pornographic images from ChatGPT and OpenAI's image-generation tool, DALL-E. These workers had been attracted by the prospect of gaining access to the lucrative world of computing, but ended up suffering from post-traumatic stress disorder and having their personal relationships destroyed because of mental health problems.