OpenAI used $2 African workers to improve ChatGPT

Artificial intelligence startup OpenAI is at the epicenter of the storm after the release of the magazine TIME to report the “abuse of labor” he carried out on African workers in order to improve his famous ChatGPT software.

According to a recent report, workers Kenya they were paid less than $2 an hour to try and polish the AI ​​language by removing hate speech.

OpenAI and complaints

Command Open AI already had success with GPT-3, artificial intelligence able to join offers and provide them to users. However, he also added violent, sexist, and racist speech, which caused controversy.

With this in mind, the company started a new project to avoid the same behavior with ChatGPT and remove biases and preconceptions from the internet content that you feed your knowledge.

For this, Open AI outsourced to a company in Kenya in November 2021. Thousands of text snippets with “terrible” content, such as child sexual abuse, bestiality, murder, suicide and torture, were given to this company, all in order to create a tool that detects these signs in texts created ChatGPT and filter the content before it arrives as a response to the user.

The company was called herselfwhich is also based in San Francisco, which used workers from KenyaUganda and India to tag data from tech giants like Google and Microsoft.

TIME indicates that he has reviewed internal documents between herself D Open AI to understand that the workers’ balances were between $1.32 and $2 an hour. Just to remind Open AI it is currently valued at $29 billion.

Inhuman conditions

The pay does not reflect the amount of obscene and traumatic content employees had to go through to filter out that content.

“Despite the critical role played by these data enrichment professionals, a growing body of research is revealing the hazardous working conditions these workers face,” the AI ​​Partnership, a coalition of artificial intelligence organizations to which it belongs, said in a statement. Open AI. “This may be the result of an effort to hide the reliance on AI on this large workforce, while noting the efficiency gains of the technology. Out of sight, too, out of mind.

Documents reviewed TIME show it Open AI signed three contracts totaling approximately $200,000 with herself at the end of 2021 to flag verbatim descriptions of sexual assault, hate speech and violence. About three dozen workers were divided into three teams, one for each question. Three employees told the publication that they had to read and mark 150 to 250 passages of text in a nine-hour shift. These fragments can be anywhere from about 100 words to over 1000. Workers could only attend psychological sessions on topics they had read (they clearly cite cases of bestiality, etc.) in groups, not in person.

The contracts stated that Open AI will pay an hourly rate of $12.50. herself because of work. The spokesperson noted that this price “also covers other costs such as infrastructure, salaries and benefits for employees and quality control analysts.” representative Open AI The statement said the company did not set any performance targets and that Sama was responsible for managing employee mental health and pay regulations.

The problem came to a head when, in February 2022, Open AI delivered images of the same content to herself classify them.

We recommend you METADATA, an RPP technology podcast. News, analytics, reviews, recommendations and everything you need to know about the tech world.