Walmart’s anti-shoplifting tech slammed by staff as ‘fake AI’

A group of anonymous Walmart workers have raised concerns about the anti-shoplifting technology used to monitor the company’s self-checkout kiosks.

A group that calls themselves ‘Concerned Home Office Associates’ has circulated a video documenting the system’s flaws, including frequent failures to identify unscanned items, and incorrectly identifying personal items potentially shoplifted.

In an email sent to company management at Walmart’s headquarters in Bentonville, Arkansas, the group claims to be ‘past their breaking point,’ saying the system’s frequent false positives are irritating customers and putting workers at greater risk of COVID-19 exposure by unnecessarily having to verify customer’s purchases at unsafe distances.

An anonymous group of Walmart employees have raised concerns about anti-theft technology used at self-checkout kiosks, saying it’s ‘a fake AI that just pretends to safeguard’

‘It’s like a noisy tech, a fake AI that just pretends to safeguard,’ one of the Walmart employees, who asked to remain anonymous, told Wired.

The system was originally designed by Everseen–an artificial intelligence and technology firm based in Cork, Ireland–and relies on overhead cameras, or ‘digital eyes,’ that film customers as they scan objects into the register.

The cameras are connected to a cloud-based object detection AI that can identify individual items and flag customers who may not have properly scanned an item.

The system is so faulty, that many Walmart workers have taken to calling it ‘NeverSeen,’ and employees have registered continuous complaints since it was first implemented in 2017. 

The anonymous employees used actual footage from the Everseen to document several represent system failures, or easy ways to trick its AI.

The system uses overhead cameras to identify items that haven't been scanned into the self-checkout system but were still placed in the customer's bag on the other side, which momentarily locks the register and requires a Walmart worker to further investigate

The system uses overhead cameras to identify items that haven’t been scanned into the self-checkout system but were still placed in the customer’s bag on the other side, which momentarily locks the register and requires a Walmart worker to further investigate

In one example, a customer stacked two Reese’s White Peanut Butter Cups packages on top of each other and scanned the barcode for only one of them, while the Everseen system couldn’t detect there were actually two.

Another example showed a customer moving two one-gallon containers of milk across the scanner with one hand, with only one of the two barcodes registering because of the viewing angle.

The Everseen system appeared to think the two containers were actually a single object because the customer had used one hand to move both.

The cameras are connected to a cloud-driven AI system that was designed by the Irish company Everseen, which has been implemented in more than 2,000 stores around the US since 2017

The cameras are connected to a cloud-driven AI system that was designed by the Irish company Everseen, which has been implemented in more than 2,000 stores around the US since 2017

A third example showed the system flagging a person’s smartphone as an unscanned item after they placed it on the kiosk counter for a few seconds, locking the system and requiring a customer support representative to come over and go through the full list of scanned items.

A Walmart spokesperson told Wired significant improvements had been made to the Everseen system in early 2020, which had led to a drop in overall alert levels in the 2,000 stores that currently use the technology.

A Walmart spokesperson said that the Everseen system had recently undergone several updates to make it more accurate in early 2020

A Walmart spokesperson said that the Everseen system had recently undergone several updates to make it more accurate in early 2020

‘We assess our technology regularly, and as evident with the large scale implementation of Everseen across the chain, we have confidence it is currently meeting our standards,’ the spokesperson said.

The Everseen system replaced an earlier antitheft system that used sensors to compare the weight of the goods in a customer’s bag against a tally of what it should have weighed based on what had so far been scanned in. 

The company acknowledged that it’s also developing a number of other alternative technologies internally to help make self-checkout less cumbersome.

‘At an enterprise level, there are a number of tests happening at any given time across our footprint of nearly 5,000 stores,’ the Walmart spokesperson said.

‘The goal of IRL is to build AI capabilities that can be transferred to additional stores. We regularly test capabilities built internally in a small number of stores.’

In early 2020, the Wall Street Journal reported that one such self-checkout monitoring technology was being tested in 50 stores.