How does facial recognition work and is it safe? | WIRED Explains

There's one technology that's already being used

by millions of people around the world,

that has the potential to define the next decade

of technological progress, facial recognition.

Facial recognition systems work by matching an image

of a person's face to a previously held image of them.

The technology is powered by artificial intelligence,

and it can be used in a number of different ways,

from unlocking mobile phones to real world events.

It's been used at protests, sports matches, and at concerts.

Police say it can be used in public spaces

to improve security by helping

to locate dangerous people and suspects.

The rapid increase in the use

of facial recognition has proved to be hugely divisive.

There has been a lack of transparency

around how the technology is being used.

Critics claim the systems are unreliable

and erode people's privacy.

There are broadly two types

of facial recognition systems that are used.

The first doesn't work in real time,

it's static and uses individual images.

If you have a photo of someone, it's possible

to compare it against a database of tens, thousands,

or even millions of images to find a potential match.

This type of facial recognition technology could be used

on one individual computer or powered through the cloud.

Police in the UK have used this type

of facial recognition system

with images captured from CCTV cameras.

The second type of facial recognition system is more dynamic

and therefore more controversial.

It works in real time.

It's called Automated Facial Recognition.

It all starts with cameras

that can be positioned in public places,

such as outside shops or on high streets.

The images captured by these cameras are then processed

by artificial intelligence software.

It's able to pick out humans

and their faces from other objects around them,

such as cars, lampposts, and street signs.

The faces that analyze in a matter of seconds

and compared to images that are already held on a database.

The systems that are currently being trialed

can identify tens of faces in one image.

They can also scan huge crowds with very little effort.

When a match is made, police can receive a notification

telling them about a potential suspect on the ground.

They're then able to go and identify and locate them.

The facial recognition systems used

to unlock mobile phones work in a similar way.

They identify our likeness in real time,

but there are some differences.

The systems on our phones

are only trained to recognize one face.

They work at short distances,

and crucially, they have to be opted into

by the phone's owner.

Facial recognition systems are looking at your face

in the same way that a human would.

Instead, they process biometric markers on your face.

For instance, the gap between your nose and lips

or the width of your eyes.

An algorithm can essentially create a map of your face

by combining different measurements and personal traits.

Much like your fingerprints, your face is unique to you.

It can't easily be replicated.

Around the world, facial recognition

has been used in lots of different ways.

Airports are using the technology

to verify people's identities,

shopping malls have experimented with motion tracking.

In Taylor Swift's Reputation tour,

it was used to spot stalkers.

China has tracked Uyghur Muslims throughout the country,

and police in the UK and US have used the software

to arrest alleged criminals.

So what's the danger with facial recognition?

This is where it gets complicated.

Questions have been raised about bias in data sets,

accuracy, and the ethics of facial recognition deployments.

Essentially, a facial recognition system is only as good

as the data that has been used to train it.

Researchers in scientific labs can get the technology

to be highly accurate.

They use high quality images

of people showing their full faces in well-lit conditions.

When they do this,

they can get matches nearly 100% of the time.

But when the tech's being used in the real world,

this is when things can get messy.

Things don't work as well when CCTV cameras are low quality

and people are moving around,

cameras can't capture people's faces as well

in poor quality lighting conditions.

One of the first uses of facial recognition technology

in the UK was that the UEFA Champions League Final in 2017.

South Wales police received 2,400 alerts

of possible matches, less than 200 of these were right.

The risk is people who aren't criminals get identified

as potential suspects

and then receive unwanted police attention.

And then there's the issue of bias.

A study by MIT Media Lab has found problems

with Amazon's facial recognition technology.

The system could identify the gender of lighter skinned men,

but it mistook women for men 19% of the time.

On top of this, it mistook darker skinned women

for men 31% of the time.

The danger here is that AI systems

won't treat people fairly.

In May, 2019, lawmakers in San Francisco

banned facial recognition technology from being used.

This is massively significant

as Silicon Valley is the home of big tech.

Both Amazon and Microsoft have said there needs

to be rules put in place to control facial recognition.

The main argument in favor of the technology

is to help improve law enforcement and increase security.

Police say that facial recognition technology can be used

to help find suspects in large crowds.

They argue that this can be cheaper, more efficient,

and help to reduce crime in a certain area.

Is this gonna end up with us living in a dystopian society?

Well, it's too early to say, but one thing's for sure.

Facial recognition is gonna be debated for years to come.

Police, government and law enforcement are going

to increasingly use the technologies in live deployments.

The key issue is that the technology

is still being developed.

Without control, there's a danger

we will rely on facial recognition technology

when we don't know whether it's safe or accurate,

and at that point, things will be too late.

[suspenseful music]