In 2020, Amazon announced that its grocery delivery service Amazon Fresh was opening physical stores featuring its revolutionary AI-powered frictionless checkout system.

But just four years later, the company is dropping its Just Walk Out technology with reports claiming that it lacked the automation it claimed to have and relied on more than 1,000 information technology technicians in India.

So, are we saying goodbye to this dystopian retail concept - or is this just yet another case of the right technology at the wrong time?

The not-so-automated AI technology

When it was first introduced, Amazon Fresh’s Just Walk Out technology claimed to use a host of cameras and sensors to track shoppers throughout the store recording what items each took from the store shelves. AI-powered computer vision software would then compile and analyse the store data and charge the customer as they left the store. In practice, it took hours for customers to be charged for their purchases and receive their invoices.

Recently, it was revealed that while the technology seemed automated, in reality, it relied on more than 1,000 IT technicians in India who not only trained the AI model but also worked to verify video footage and confirm the accuracy of checkouts. Amazon has justified this practice as “necessary for continuously improving the underlying machine learning model powering Just Walk Out technology.”

A marketing ploy or a common AI practice?

Reactions to this news have varied, with many customers feeling deceived by Amazon and calling the practice dystopian and immoral. Most of the criticism focuses on how Amazon purposely falsely advertised the Just Walk Out technology while relying on low-income workers in another country and undercutting the local job market in Amazon Fresh store locations.

However, these criticisms misunderstand how AI models are made and trained. Human-led data labelling is how tech companies train their AI models. It is the only way to verify the accuracy of AI models and encode human values and goals to make them as helpful, safe and reliable as possible. This training is often located in India because of the country’s advanced IT industry and large-scale cheap workforce. Outsourcing business operations overseas is a common practice among for-profit organisations to reduce production costs and ensure competitive pricing.

When the technology was first adopted, Amazon used human labour as a back-stop, hoping that eventually the AI model would take over. It then decided to drop the technology because the data labelling became too labour-intensive without significant model improvement.

According to Theo Wyat, the reporter who broke the story at The Information: “As of mid-May 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.”

Amazon disputed the numbers, but clearly, the company had difficulties scaling the technology without significant human input, leading to its abandonment. Frictionless shopping has become relatively popular across single stores in sports stadiums and arenas, but it is still unrealistic for a retail conglomerate to adopt the technology on a large scale and at a reasonable price.

One step back, two steps forward

Amazon is not giving up on frictionless! To replace Just Walk Out, the e-commerce giant launched Dash Carts. Cameras and sensors will now be installed on shopping carts as opposed to across a store’s layout. We are yet to see whether the new iteration of the technology will stick, but it is evident that Amazon will not let one failure dent its ambition to digitalise the human retail experience.