Behind the success of ChatGPT, OpenAI used an army of underpaid little hands in Kenya

A survey by Time magazine reminds us that behind the scenes of great adventures, less happy stories can be brewing.

The American company and its star chatbot are talked about a lot, as the latter fascinates or frightens, if not both. As was the case with its cousin DALL-E, ChatGPT is powered by a massive database of content on which it relies. However, as advanced as artificial intelligence is, these data often still have to be identified by very human beings.

Workers paid below minimum wage

OpenAI called on an American company, Sama, to annotate thousands of inappropriate textual and pictorial data. If the general objective is to continue improving ChatGPT, the developers want to provide it with a tool allowing it to moderate its results. They thus have less chance of making comments on sexual abuse, hate speech, or other forms of violence, for example.

This is how Kenyan workers, hired by Sama, found themselves sorting through a few hundred pieces of explicit content every day, some of which was even illegal in the United States. Paid less than two dollars an hour and depending on their performance, some of them did not know how to classify what they saw, so vague were the guidelines and some confusing and disturbing content.

In addition, Sama, which describes itself as an “ethical” company, had to guarantee access to sessions with “wellness” counselors, but several employees explained that these sessions were often unnecessary and rare due to the requirements. very high productivity. In addition, requests for individual sessions were, most of the time, refused by the management.

An essential task for AI

The partnership between the two companies, which began in early 2022, was broken a few months before the end of their contract by Sama. This justifies itself by stating that it refuses to receive more content of this type, especially since it had already experienced a very similar situation a few months earlier with Meta, the parent company of Facebook.

Although the company wants to continue working with big companies like Google and Microsoft by offering ” visual data annotation solutions “, she says she wants to withdraw from the moderation market. But, the end of the lucrative contracts with Meta and OpenIA means the closure of its offices in Nairobi and the dismissal of many workers, who deplore this situation despite the working conditions to which they were subjected. ” It was a way to support our families “says one of them.

This story is not an isolated case in the AI ​​industry, as this kind of cheap labor is essential for developers. An OpenIA spokesperson explains that ” classification and filtering of [textes et images] harmful are a necessary step to reduce the amount of violent and sexual content included in the training data and to create tools capable of detecting harmful content “. With the growing importance and interest of systems such as ChatGPT, as evidenced by Microsoft’s significant investment in OpenAI, it is very likely that more such examples will multiply.

To conclude, know that I have asked ChatGPT to write this article for me. But, he seems to have a very biased opinion on the matter:

Source : Time