Covering Disruptive Technology Powering Business in The Digital Age

Home > Archives > News > The Facial Recognition Showdown
The Facial Recognition Showdown
July 19, 2019 News


The combination of big data, AI and amazingly powerful CCTV cameras has opened up a pandora’s box of a debate when it comes to how it is used.

An article published by the BBC yesterday highlighted a group of members of parliament in the UK calling for British police to stop using live facial recognition technology. Amongst their concerns are built-in and learned bias of the AI and doubts about accuracy, causing many false positive identifications.

Freedom of rights groups are also questioning the ethics of wide scale video recognition, and that’s yet another pandora’s box.

In terms of the technology itself, DTA had the chance to put it to the test yesterday when we visited Dell’s AI experience centre in Singapore. One of the “exhibits” is a CCTV system with an edge-powered AI system. For demonstration purposes, this “fun” exhibit captures everyone that appears before its cameras and tells you three things, gender, age and mood.

In light of the concerns raised by these British politicians, we decided to run a test. My colleague and I stepped in front of the camera to see if accuracy and bias allegations could be substantiated. Ok, a test of two people hardly stands up to scrutiny, but the results were interesting nonetheless.

I am a caucasion male, aged ahem 40 something. The AI CCTV nailed me perfectly, got my gender and age spot-on, it told me my mood was neutral, which I wouldn’t have argued with.

My colleague is female and in her early 50s. The AI told her, yes she was female but put her age at 23. Compliments aside, and despite my colleague’s obvious joy at how young she must look, our tiny litmus test immediately substantiated the questions about accuracy and bias.

Leaving aside the moral debate around the rights and wrongs of us all being watched all the time. Another big question that this raises is about how do you “fool” the AI. Because AI works on data points, it’s highly likely that people wanting to avoid detection will use different techniques as they learn how to fool AI.

As an example, the false beard may be thrown into the bin in place of “naked to the human eye, reflective tape” that confuse camera analytics.

As is the case with much technology these days, we need to understand where it fits. Generally, AI and data helps find needles in the haystack, but it still requires people to work out if they are the right needles!