How Facial Recognition Software Needs to Improve, Fast

The future of facial recognition promises so much. Security, direct marketing, improvements in AI, to name but a few.

But all is not well in the world of facial recognition. Issues that should have been resolved long ago are causing controversies that threaten to derail facial recognition.

What Facial Recognition Should Deliver

From science fiction to reality, facial recognition technology can detect facial characteristics and use them for identification. Various hardware manufacturers are using this technology for user authentication, for example.

This single piece of technology has so many uses, from targeted advertising to detecting missing persons. School attendance and gambling house cheats can also be monitored and dealt with thanks to facial recognition technology.

As advanced as the technology appears to be, facial recognition needs to improve, fast. From accusations of inadequate testing to issues with racial detection, facial recognition has serious problems.

Inadequate Testing of Facial Recognition Tech

Police in the UK have been trialling facial recognition technology in law enforcement for several years, as reported by the BBC. Unfortunately, it continues to undergo testing, following problems with duplicated and unlawfully held (of individuals cleared of offence) images in its facial recognition system provided by German company Cognitec.

And despite face match searches increasing from 3,360 in 2014 to 12,504 in 2017, assessors have repeatedly overlooked opportunities to check for racial bias in the software, probably caused by an unrepresentative dataset. Worse still, a Home Office assessment determined that the facial recognition system still has some way to go, describing it as “only half as good as the human eye.”

Describing the situation as “wilful blindness to the risk of racism,” Silkie Carlo of Big Brother Watch insisted the facial recognition technology be “dropped immediately.”

The UK police aren’t alone in struggling with facial recognition and race. Amazon too has been drawn into the challenges facing the technology, with shareholders attempting to limit the sale of the controversial Rekognition.

Amazon’s own facial detection software has been described as “automating mass surveillance” by the ACLU, who also claim that Rekognition suffers from shortfalls in facial recognition accuracy, including race.

The Risk of Ridicule

Elements of the technology clearly work. But when it comes to mass facial detection and recognition there is clearly some way to go before it is ready to be rolled out. The formation of a consortium of key tech players has worked for other technology, so may save facial recognition from long-term failure and ridicule.

Until inherent issues with racial differences, and even standard variations between people of the same ethnicity, can be overcome, facial recognition is at risk of becoming a joke.

When the punchline concerns liberty, improvements in facial recognition need to be made fast.