Jump to content

Amazon kills “Just Walk Out” shopping tech—it never really worked


Recommended Posts

ARSTECHNICA.COM

"AI" checkout was actually powered by 1,000 human video reviewers in India.
Quote

Amazon is giving up on the cashier-less "Just Walk Out" technology at its "Amazon Go" convenience stores. The Information reports that new stores will be built without computer-vision-powered surveillance technology, and "the majority" of existing stores will have the tech removed. In the early days, Amazon's ambitions included selling Just Walk Out to other brick-and-mortar stores. The problem was that the technology never really worked.

As it says on the tin, Just Walk Out was supposed to let customers grab what they wanted from a store and just leave, skipping any kind of checkout process. Amazon wanted to track what customers took with them purely via AI-powered video surveillance; the system just took a phone scan at the door, and shoppers would be billed later via their Amazon accounts.

When the technology was announced in 2016, Amazon's sales pitch asked, "What if we could weave the most advanced machine learning, computer vision, and AI into the very fabric of a store so you never had to wait in line?" The store was filled with 100-plus cameras and rigid item locations, all designed to try to make AI-powered computer vision checkout possible.

A May 2023 report from The Information revealed the myriad tech problems Amazon was still having with the idea six years after the initial announcement. The report said that "Amazon had more than 1,000 people in India working on Just Walk Out as of mid-2022 whose jobs included manually reviewing transactions and labeling images from videos to train Just Walk Out’s machine learning model."

Training is part of any AI project, but it sounds like Amazon wasn't making much progress, even after years of working on the project. "As of mid-2022, Just Walk Out required about 700 human reviews per 1,000 sales, far above an internal target of reducing the number of reviews to between 20 and 50 per 1,000 sales," the report said.

The report said Amazon's team "repeatedly missed goals" to cut down on human reviews, and "the reliance on backup humans explains in part why it can take hours for customers to receive receipts."

According to The Information report, the Just Walk Out tech was deployed in 20 Amazon Go stores, 40 Amazon Fresh grocery stores, and two Whole Foods stores. It had even gotten some third-party outlets on board, including "30 stores operated by other companies in US sports stadiums, 12 airports, and one university in Arlington, Virginia," the report said.

Amazon will be switching to a more reasonable cashier-less format: shopping carts with built-in checkout screens and scanners. Customers can leisurely scan items as they throw them in the "Amazon Dash Cart," and the screen will let them instantly see a running total of their purchases. The cart scanners also mean that Amazon no longer needs a small army of store employees to constantly tidy up the shelves so the camera-vision system can work.

While the concept was that AI would track which items were purchased , "Walk Out required about 700 human reviews per 1,000 sales".  In essence, purchases were mostly tracked by workers watching you on cameras in India.

Link to comment
Share on other sites

14 minutes ago, AbsolutSurgen said:
ARSTECHNICA.COM

"AI" checkout was actually powered by 1,000 human video reviewers in India.

While the concept was that AI would track which items were purchased , "Walk Out required about 700 human reviews per 1,000 sales".  In essence, purchases were mostly tracked by workers watching you on cameras in India.

 

It wasn't a mechanical Turk situation, they were using a lot of human training. 

Link to comment
Share on other sites

1 hour ago, AbsolutSurgen said:

When 70% of the transactions are manually reviewed several years in, its not far off.

 

Seems the issue of scale. That is, their AI having trouble with larger, complicated purchases with multiples of multiple items. Smaller purchases were fine, larger ones weren't as accurate. Amazon seems to still be using the tech in their smaller convenience stores with greater luck/accuracy.

Link to comment
Share on other sites

Wou;dn't it be funny if it turns out that chat GPT and all this shit AI artwork was just being hastily made by tens of thousands of Indians in a warehouse somewhere?

  • Haha 1
Link to comment
Share on other sites

2 hours ago, Brian said:

I have an Amazon Go in my lobby and use it everyday.  Never had an issue

 

The reporting on this story has been really bad. Those Amazon Go-sized shops are using their shopper tracking AI thing just fine and that product is live and being shopped to other convenience stores and airports around the world. Seems Amazon has been unsuccessful at scaling that up to a full size grocery store where someone might purchase dozens or even over a hundred items. They've had a team of people in India doing some QA to make sure the AI is getting things right and found that 70% of the time the AI loses track of items and needs a human review to go in and double-check the numbers. Amazon was shooting for a 90% accuracy rate and couldn't get it past 30% after a couple of years, so they're scrapping the grocery-scale project and sticking with the Amazon Go-sized project.

 

I've read some awful headlines about how Amazon's AI was really just a thousand people in India, but that's really not what's going on here.

  • Halal 1
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...