Veritone Inc (VERI)

[Paullee]

https://www.ft.com/content/c2300a60-5a2b-11e9-939a-341f5ada9d40?ftcamp=traffic/partner/feed_headline/us_yahoo/auddev

Taser stun gun maker has developed facial recognition software
Axon has filed patents for technology that can guess a person’s race and age

The maker of Taser stun guns has filed three patents for software with facial recognition capabilities, despite repeatedly claiming not to be working on the controversial technology.

Axon, which provides equipment and software to law enforcement agencies, has invented video analysis tools that recognise and blur out sensitive information in police footage, such as officers’ faces, and can identify and match individuals to police databases.

According to the US patents, two of which are still pending approval, Axon’s technology can guess a person’s race and age as it measures features including “nose length” and “spacing from cheeks to chin”.

As well as supplying Tasers, Axon is the biggest provider of police body cameras in the US, and bought its main competitor Vievu in 2018. According to Axon, 48 major cities in the US use its cameras and software.

Facial recognition technology, which analyses images to identify people, has come under fire from civil rights and privacy groups as an unreliable tool that could worsen existing inequalities and be used for mass surveillance.

Last week, the US Securities and Exchange Commission told Amazon it must allow shareholders to vote on a proposal to stop selling its Rekognition software to governments after concerns that it threatens civil and human rights.

Axon has denied developing similar software, though in April 2018 it founded an artificial intelligence ethics board to consider the implications of technologies including facial recognition.

Rick Smith, chief executive, last year said the technology was not yet accurate enough and technical failures could have “disastrous outcomes”.

He said that once accuracy and privacy concerns had abated, the company could begin developing facial recognition products and “move into commercialisation”.

But two patents for the company’s video redaction product filed in 2016 and 2018 said it could automatically find and blur out sensitive information by analysing footage recorded on Axon body cameras.

Once a user selects specific details to redact from video evidence, such as a face, the tool uses techniques including “facial recognition” and “licence plate detection” to find and bulk redact those details from large volumes of footage.


The company filed a third patent in 2017, which is pending approval, for real-time video analysis software that uses facial recognition to identify individuals captured on police body and in-vehicle cameras.

The system takes a photograph of an individual’s face and matches it to law enforcement databases, such as mugshot files. It then tells the officer the person’s name, and “may include any law enforcement information related to the person”.

It could determine factors including a person’s race, gender and age, and generate a “mathematical description of the person’s facial geometries (eg. eye size, eye shape, eye spacing, nose shape, nose length, spacing from nose to eyes, spacing from cheeks to chin)”, the application said.

It could also detect threats such as gunshot sounds or weapons in the video, which may “trigger an alert” and “transmit a request for additional back up”.

Axon said in a statement: “Patents are awarded to the first person to file the idea?.?.?.?Filing a patent application is by no means a declaration of taking action or considering technology or product development.”

It added: “Our redaction product is currently limited to detection of faces. We do not match the detected faces against a centralised database which would be identification of the individual’s faces.”

Moji Solgi, Axon’s director of AI and machine learning, denied the company was actively developing facial recognition products, but did not dispute it could. “Someone with some level of expertise in their garage could do it,” he said. Ensuring a product was accurate and had “more benefits than risks” was more difficult, he added.

Mr Solgi said Axon would not pursue the development of this software until after the ethics board had outlined its findings, which it is expected to do within the next year. Even with approval, Axon would not necessarily begin development, he said, adding that the company was not bound by the board’s recommendations.

Michael Connor, executive director at campaign group Open Mic, said it was dangerous for companies to be “paving the way” for the rollout of facial recognition products “without having thought it through”.

“AI experts say we’re not there yet,” he said, pointing to a study by MIT Media Lab researcher Joy Buolamwini, which found that two leading facial recognition programmes misidentified black women up to 35 per cent of the time.

Recommended

FT Magazine Aliya Ram
The perils of AI bias
Axon is looking to rising software sales for growth. Vievu, the company it acquired in May 2018, had previously partnered with the AI company Veritone to improve its video and audio analysis.

Chad Steelberg, Veritone’s chief executive, said the company’s tools, which included “face” and “object recognition”, would help police officers “derive comprehensive, actionable insights from their body camera footage in near real-time”.

Veritone’s IDentify application also uses facial recognition to analyse “massive data sets at incredible speed” to identify “potential suspects of persons of interest”.

Axon said it was not using Veritone technology on any Axon or Vievu products. Mr Solgi said a report last year that Axon had sought to buy AI company Kairos or license its facial recognition software was false, although engineers from the two companies had met.

Axon plans to unveil a new AI-powered video analysis product at its annual conference.