Axon, creator of the Taser, did something unusual for a technology company last year. The Arizona corporation convened an ethics board of external experts to offer guidance on potential downsides of its technology.
Thursday, that group published a report recommending that the company not deploy facial recognition technology on its body cameras, widely used by US police departments. The report said the technology was too unreliable and could exacerbate existing inequities in policing, for example by penalizing black or LBGTQ communities.
Axon’s CEO and founder Rick Smith agrees. “This recommendation is quite reasonable,” he says in an interview. “Without this ethics board we may have moved forward before we really understood what could go wrong with this technology.”
The decision shows how facial recognition technology—while not new—has become highly controversial as it becomes more widely used. The power that software capable of recognizing people in public could give police and governments has struck a nerve with citizens and lawmakers seemingly inured to technology that redefines privacy. As a result, Axon and other technology companies are advancing more cautiously with the technology, a departure from the usual pattern of moving fast, breaking things, and leaving society to patch up the problems.
Civil rights groups, lawmakers and companies including Microsoft and Amazon have called for restrictions on facial recognition—although there is disagreement on how tight or absolute those should be. Their concerns have been amplified by researchers showing how facial analysis algorithms can suffer biases that make them less accurate for women, children, and people of color. San Francisco has banned city agencies from using facial recognition, and at a congressional hearing last month lawmakers on both sides of the aisle expressed support for federal rules on the technology.
“Often new technologies are just sold and our communities find there are negative impacts later,” says Mecole Jordan, one of 11 members of Axon’s ethics board, who include lawyers, technologists, and law enforcement veterans. Jordan is executive director of the United Congress of Community and Religious Organizations, which works on police accountability and other issues in Illinois. “I’m excited that Axon’s CEO has said he would adhere to our recommendations,” she said.
Axon formed its ethics board in April 2018, saying it would meet quarterly and publish one or more reports each year. That schedule proved optimistic: Thursday’s report was the group’s first and it said the members have met only three times. But the board is one of the most prominent examples of a tech company creating a new governance structure to keep its artificial intelligence projects within moral bounds.
“Overall this appears to be a very good effort,” says Don Heider, executive director of Santa Clara University’s Markkula Center for Applied Ethics. “These efforts to put together ethics boards are new [and] this company is on the cutting edge.”
Microsoft and Google both say they have internal review processes for AI projects that have led them to turn down certain contracts. Critics—including employees who protested Google’s work on a Pentagon drone project—say external oversight is necessary. In May, Google abandoned an attempt to create an external AI panel after opposition to one of its members, Kay Coles James, president of conservative think tank the Heritage Foundation.
Smith says Axon’s ethics board was inspired by a controversy of its own, after the company acquired two AI companies in 2017. Speculation began to spread that Axon would inevitably threaten privacy by adding facial recognition to its cameras; Smith says the company only planned to create tools to help police manage videos, and redact faces and other identifying information. “The idea was this board could bring us the perspectives we don’t normally hear,” Smith says. Axon also has a history of fighting having to fight claims its technology is dangerous. The company has been a defendant in more than 120 wrongful death lawsuits involving Tasers in the US, according to a 2017 Reuters analysis of legal filings. In some the company was found liable, and paid damages.
After Axon revealed its new ethics panel, some academics and community groups said it didn’t represent a sufficient diversity of views. The report issued Thursday describes how new members were added—including Jordan—to counter accusations including that the group did not include enough representatives of the people most affected by police technology.
Thursday’s report says facial recognition emerged as centerpiece of the board’s discussions. It notes evidence that the technology is less accurate for people with darker skin, and on fast-moving video footage, and concludes that Axon could not ethically integrate it into bodycams. “Face recognition technology is not currently reliable enough to ethically justify its use on bodyworn cameras,” the report says.
Smith agrees with that—but expects to change his mind later. He predicts that improved technology will eventually resolve the problems with accuracy and bias highlighted in the report. “I’m confident those will get solved over time,” he says.
Axon’s AI experts will keep evaluating facial recognition. Smith says that if the company does decide it’s time to deploy the technology, advice from its ethics board will help ensure the product is designed responsibly. “In our industry we will use ethical design as a competitive advantage the way Apple uses privacy as a competitive advantage,” he says.
Smith’s determination to keep his options open sets up the potential for conflict with his ethics board over when exactly the technology is ready, and if improved accuracy alone is enough. Heider, of Santa Clara, says that would be a major test of the ethics board format and the extent to which it constrains companies.
Smith says although he is free to ignore the ethics board, it is independent and public enough that doing so without presenting a good reason would be painful for Axon, a publicly traded company. He also expresses hope that Axon’s show of self-restraint can help convince regulators not to rush into legislation like that under consideration in California that would ban facial recognition on body cameras. “I get a little worried that we may over regulate something that’s not a practical problem and create challenges down the road,” he says.
Jordan, the board member, says the group will weigh the evidence carefully if Axon suggests facial recognition algorithms are ready for broader use. “If ever the technology catches up to a point where it is equitable across race, gender, and ethnicity that would be a conversation for a different day,” Jordan said. “But that’s a big if, and a big when.”
Alvaro Bedoya, director of Georgetown’s Center on Privacy & Technology, welcomes the Axon report but says that it shouldn’t distract from the need to rein in already deployed uses of the technology. The center has produced influential reports revealing broad use of facial recognition by the FBI, and how Detroit and Chicago bought systems capable of watching for specific faces in real-time.
“Face recognition is not nascent—in 2016, we figured out that FBI face recognition searches were more common than federal wiretaps,” he says. “If we regulate now, we would be regulating after the technology has spread and after seeing it be misused and abused.”