Who trains the Artificial Intelligence of the future cars without a driver? A group of Venezuelans


An interesting phenomenon arises in Venezuela: it has been a fertile field for recruitment by technology companies that are looking for human drivers to complete the training of their artificial intelligences.

The story, here:


Desperate for work Venezuelans have discovered a new group of collective work platforms (or crowdworking) on-line. These companies, like Mighty AI, Playment, Hive and Scale, supply the driverless car industry and could be a new battleground in the debate over whether these freelancers should be considered employees.

Hundreds of thousands of Venezuelan workers signed up last year to work with these companies, and in some cases they represented as much as 75 % of the workforce of one of those companies. Even today, 75 % of search traffic to Mighty AI comes from a page that advertises jobs in Venezuela. Are companies don't pay more for data tagging than platforms like Amazon Mechanical Turk, but they offer a more stable source of income, thus granting a guarantee of safety to workers in a country where inflation recently hit 10 million percent. (Mighty AI did not respond to a request for comment for this article.)

Labeling pixels

It's no secret that artificial intelligence (AI) relies heavily on underpaid people to tag massive amounts of data. Those people do everything from transcribe voice recordings until identify NSFW images (not safe for work, not appropriate for the job). The new companies of crowdwork They are the result of increasing competition to develop autonomous cars and the high demand to train vehicles to see and navigate properly.

Other data labeling tasks, such as creating an algorithm for search results, have more room for error. "If you do a search engine query and three out of 10 results are bad, it doesn't really matter," Schmidt emphasizes. "But a level of 30 % of incorrect answers would be totally intolerable in the traffic conditions«. The job itself can be more demanding as well. In-car cameras record a large amount of visual information, and data taggers must describe each object in an image or video.

More information in the MIT Tech Review in Spanish

Previous What iPhone 11 and iPhone 11 Pro Reviews Say: Stick with the Basics
Next Basic tips for customer loyalty

No Comment

Leave a reply

Your email address will not be published. Required fields are marked *