A Rotorua woman wrongly kicked out of a supermarket using facial recognition technology says she will keep on fighting its spread.
Foodstuffs says a trial in 25 New Worlds and Pak'nSaves has concluded the technology reduces harmful behaviour shoppers, and its use will carry on.
Te-Ani Solomon said she too would carry on with her case at the Human Rights Review Tribunal, claiming the technology was discriminatory.
"Because I really don't want this to happen to anybody else and for my own selfishness, I have to, because I've never in all my life felt so powerless till I did that night, where I was in front of my son, embarrassed, trying to justify how I wasn't a thief," she told RNZ on Friday.
Her son had to see her "telling them it's not OK".
An independent assessment of the six-month trial said nine people were misidentified and two were asked to leave stores. An estimated 100 "serious harm events" were prevented, it said.
"They have found that facial recognition has reduced serious harm in the trial stores by an estimated 16 percent and has strong public support," Foodstuffs North Island lawyer Julian Benefield told Morning Report.
But Solomon said 100 preventions for nine shoppers compromised was a high rate of harm.
"These are targeted results, because from where I'm standing, and I'm sure eight other people would agree, that was no success."
Benefield said Foodstuffs would keep the technology in the 25 stores but not roll it out further.
Solomon, a business analyst, said people should think again - "there's got to be a better way" to fight shop crime, she said.
She was pursuing the case partly because her 13-year-old son, who was with her that night in April, still did not understand fully what happened.
Two "burly" staff members came up to them at the local New World, misidentifying her while she was buying chops for a birthday tea, asking her to leave, then followed her around.
She said she was "mortified" as they took photos of the back of her and her son. She admitted she swore at them - but also said she showed them three forms of ID.
"Their piece of technology that they were using was worth more than my word and my ID and photo.
"It made them feel justified."
She had gone on to find out more about facial recognition trials and use, such as by New Zealand police, and its weaknesses, such as its historic track record of bias against people with browner skin, and women.
Her claim to the human rights tribunal argues that Foodstuffs was conducting a live trial of an emerging technology with known bias, at settings that invited her racialised discrimination and preventable harm.
Benefield said all nine misidentifications were due to human, not software, error, and there had been more training of staff.
Solomon rejected that, saying they had already had training before the trial.
Foodstuffs was arguing at the tribunal that it was not the right target as the store was owner-operated, she said.
But the store owners likely had little choice about joining the trial, she said - her local owners were "lovely people" who were mortified at what happened, and apologised - though she now shopped at a Pak'nSave without facial recognition.
The tribunal process was slow, and she had also tried the Privacy Commissioner.
"We have been trying to do everything ... to get the trial stopped."
Solomon said it harmed eight other people.
In 2019 testing, bias was found to be high in many facial recognition systems globally.
However, more recent testing shows it has been reduced a lot. The top global tester, the NIST in the US, has introduced a demographic accuracy metrics and routinely puts out the results about bias.
The technology is hampered by poor lighting, such as in some shops, and by poor quality photos to match against, reports show. Official systems such as police use, match against driver's licence and passport photos, which are high quality, but shops might not have quality photos of shoppers to compare against.
Woolworths supermarkets do not have facial recognition but its staff do wear bodycams.
Sign up for Ngā Pitopito Kōrero, a daily newsletter curated by our editors and delivered straight to your inbox every weekday.