The Internet mocks Amazon after their AI is exposed as using 1,000 Indian workers

earthpressnews

Amazon boasted of automated supermarkets equipped with state-of-the-art sensors and artificial intelligence. In reality, there should have been more than 1,000 human workers who kept the store running.

What kind of project was that? Under the name “Amazon Fresh”, the tech giant operates supermarkets where you can shop locally. A technology called “Just Walk Out” was used, which was supposed to process the purchase fully automatically:

  • The company advertised “checkout-free shopping”: cameras and sensors were supposed to closely track shoppers and automatically scan the goods
  • Customers had to scan a QR code when entering the store and could simply walk out of the store after their purchase, without the hassle of checking out
  • However, as The Information (Paywall) reported on April 2nd, it is now parting ways with the concept

    • However, it also turned out that the system only worked with more than 1,000 Indian workers watching and labeling the videos. So somehow there were still cashiers – just no longer on site.

    By 2022, 70% of purchases made via Just Walk Out will have required manual verification – significantly more than Amazon’s target of 5%.

    However, Amazon denies the number of manual checks. A spokesperson for the company told Gizmodo that employees simply annotated the videos to improve automatic recognition. They would only have had to validate a “small minority” of purchases if the AI ​​couldn’t determine a purchase.

    However, one thing in particular seems to stick with many Internet users: Amazon’s AI is actually 1,000 Indians.

    With news like this, the “deep dive” feels very far away again, doesn’t it?

    The “Deep Dive” is intended to immerse us deeper into MMORPGs than ever before – what is that actually?

    More videos

    Autoplay

    Users scoff: AI = Everything Indian

    How is this discussed? There are already numerous videos circulating online that deal with the topic in a more or less humorous way. A sketch depicting a theft in the supposedly fully automated Fresh supermarkets received more than 60,000 likes on TikTok within one day.

    In many places you can also see comments that make fun of reinterpreting the English abbreviation “AI” for “Artificial Intelligence”:

  • chocolateboomslang writes on Reddit: “It was AI… All Indian.”
  • Nor 2020 says on TikTok: “They use AI = Alternative Intelligence.”
  • Other users on TikTok talk about “Accurate Indians” or “Actually [Tatsächlich] In the”.

    • Still others, however, point out that the Indian labor force was merely doing quality control. This is how machine learning would work: people would still have to tell the AI ​​whether it came to the right conclusion.

    That’s true, but it can’t completely eliminate the bad aftertaste. After all, Amazon boasted about the technology on its website: There it talks about a combination of computer vision, object recognition, advanced sensors, deep machine learning and generative AI – but not about workers in a low-wage country like India.

    In fact, it’s not the first time something like this has come to light: AI-controlled drive-thrus for fast food are actually said to be Filipino workers (via The Verge).

    The startup x.ai advertised with AI assistants, but here too, people did the work. The company closed in 2021 because they were apparently unable to get the algorithms to a point where they could work independently.

    Another prominent example is Facebook, which switched off its AI assistant “M” in 2018 after around two and a half years. Here, too, human employees were called in if the AI ​​didn’t know what to do – and that probably happened more often than planned.

    Ironically, Amazon itself even has a department dedicated to helping other companies train and operate AI systems. “Amazon Turk” is named after the “Chess Turk,” a supposed chess robot from the 18th century. However, the mechanical chess player was actually operated by a human.

    Reports like these show that in the wondrous world of AI, everything is not always as it seems and that we are probably still a long way from “real” AI. According to Elon Musk, things won’t get any better in the foreseeable future, because the development of AI could slow down significantly next year:

    According to Elon Musk, artificial intelligence will be slowed down in 2025 – and his electric cars are to blame

    mmod-game