Facial-recognition technology like Rekognition analyses facial features – such as the distance between the nose and eyes – in images and live video, allowing for people captured in security camera footage to be compared against an existing database of faces (such as of people with criminal records, or missing people). Law-enforcement agencies in the US, UK and elsewhere have been experimenting with deploying live facial-recognition in public areas, such as at airports and sporting and music events.

This has received criticism as an infringement on privacy, as well as due to the fact that commercial facial-recognition software tends to perform poorly when identifying women, and people with darker skin.

In order to demonstrate the potential for commercial facial-recognition software to lead to incorrect matches, the American Civil Liberties Union (ACLU) conducted a test in which it applied Amazon’s Rekognition software to 120 images of Californian lawmakers. The ACLU found that it incorrectly identified 26 of the lawmakers (more than 1 in 5) as matches against a database of 25,000 mugshots.

The ACLU warned that more than half of the falsely matched lawmakers were ethnic minorities. These findings reflect a similar test run by the ACLU in 2018 which misidentified 28 members of the US Congress as criminals, and which disproportionately affected women and ethnic minorities. Amazon argued that the ACLU misused the software by running facial recognition with an 80 per cent confidence threshold. While 80 per cent confidence is the default setting, Amazon recommended in a blog post that Rekognition should be used with 99 per cent confidence threshold.

In February, California Assembly Member Phil Ting introduced a bill, the Body Camera Accountability Act or AB1215, that would ban use of the technology in police body cameras. Although no police forces in California require officers to wear cameras containing this software, this bill could prevent them deploying it until it is absolutely fault-proof. Axon, which makes police body cameras for the Los Angeles Police Department and many other police forces, has warned that it may never be appropriate to introduce facial-recognition software to these devices.

The bill was approved by the Calfiornia Assembly in May and is due for a Senate vote in the coming weeks. Ting ran the demonstration with ACLU to demonstrate that the technology could lead to dangerous mistakes, and “innocent Californians subjected to perpetual police line-ups because of false matches”.

Writing on Twitter, Ting described the findings as: “Another reason to pass my bill, [AB1215], which bans facial recognition software in police body cameras. [The ACLU] test found 26 of 120 state lawmakers were falsely matched with someone in a mugshot database, [including] me. Imagine the real world implications.”

The bill is opposed by the Riverside Sheriffs’ Association, which stated that: “By banning this technology, California will be announcing to the nation and world that it doesn’t want our law enforcement to have the necessary tools they need to properly protect the public and attendees of [large cultural] events.”

If the bill is passed, California will become the largest US state to ban facial-recognition technology in police body cameras. New Hampshire and Oregon passed legislation in 2017 prohibiting this use, while the city authorities Oakland, San Francisco, and Somerville have banned the use of the technology by law enforcement this year.

This week, Amazon announced that its Rekognition software could now detect fear as well as performing more accurately in identification of gender and other emotions such as happiness, sadness, anger, surprise, disgust, calmness and confusion.

Facial-recognition technology like Rekognition analyses facial features – such as the distance between the nose and eyes – in images and live video, allowing for people captured in security camera footage to be compared against an existing database of faces (such as of people with criminal records, or missing people). Law-enforcement agencies in the US, UK and elsewhere have been experimenting with deploying live facial-recognition in public areas, such as at airports and sporting and music events.

This has received criticism as an infringement on privacy, as well as due to the fact that commercial facial-recognition software tends to perform poorly when identifying women, and people with darker skin.

In order to demonstrate the potential for commercial facial-recognition software to lead to incorrect matches, the American Civil Liberties Union (ACLU) conducted a test in which it applied Amazon’s Rekognition software to 120 images of Californian lawmakers. The ACLU found that it incorrectly identified 26 of the lawmakers (more than 1 in 5) as matches against a database of 25,000 mugshots.

The ACLU warned that more than half of the falsely matched lawmakers were ethnic minorities. These findings reflect a similar test run by the ACLU in 2018 which misidentified 28 members of the US Congress as criminals, and which disproportionately affected women and ethnic minorities. Amazon argued that the ACLU misused the software by running facial recognition with an 80 per cent confidence threshold. While 80 per cent confidence is the default setting, Amazon recommended in a blog post that Rekognition should be used with 99 per cent confidence threshold.

In February, California Assembly Member Phil Ting introduced a bill, the Body Camera Accountability Act or AB1215, that would ban use of the technology in police body cameras. Although no police forces in California require officers to wear cameras containing this software, this bill could prevent them deploying it until it is absolutely fault-proof. Axon, which makes police body cameras for the Los Angeles Police Department and many other police forces, has warned that it may never be appropriate to introduce facial-recognition software to these devices.

The bill was approved by the Calfiornia Assembly in May and is due for a Senate vote in the coming weeks. Ting ran the demonstration with ACLU to demonstrate that the technology could lead to dangerous mistakes, and “innocent Californians subjected to perpetual police line-ups because of false matches”.

Writing on Twitter, Ting described the findings as: “Another reason to pass my bill, [AB1215], which bans facial recognition software in police body cameras. [The ACLU] test found 26 of 120 state lawmakers were falsely matched with someone in a mugshot database, [including] me. Imagine the real world implications.”

The bill is opposed by the Riverside Sheriffs’ Association, which stated that: “By banning this technology, California will be announcing to the nation and world that it doesn’t want our law enforcement to have the necessary tools they need to properly protect the public and attendees of [large cultural] events.”

If the bill is passed, California will become the largest US state to ban facial-recognition technology in police body cameras. New Hampshire and Oregon passed legislation in 2017 prohibiting this use, while the city authorities Oakland, San Francisco, and Somerville have banned the use of the technology by law enforcement this year.

This week, Amazon announced that its Rekognition software could now detect fear as well as performing more accurately in identification of gender and other emotions such as happiness, sadness, anger, surprise, disgust, calmness and confusion.