♪ ♪ ♪.

This program is brought to you by the members of WCNY.

Thank you.

With Pomeroy Markers, history is based on the facts embedded by professional historians.

Over 3,000 installed nationwide, from Maine to Florida, to California, to Minnesota, to Texas, and Alaska.

Since 2006, our markers have helped people celebrate community history.

Marker grant program information available at WGPFoundation.org.

Stickley Furniture is a proud underwriter of "Balancing Act" with John Katko.

Building on more than 125 years of expertise, Stickley is an enduring American brand that uses thoughtful design, authentic materials, and time-honored techniques to craft heirloom quality furniture.

cherished for generations.

♪ ♪ ♪ ♪ ♪ ♪ KATKO: Welcome, America, to "Balancing Act" the show that aims to tame the political circus of two-party politics.

I'm John Katko.

This week, facial recognition technology.

Is it innovative or invasive?

In the center ring, we'll ask biometrics expert, Dr.

Michael King, on the trapeze, Nathan Fried Wessler of the ACLU, and Jake Parker of the Security Industry Association, will take swings at the issue.

Plus, I'll give you my take.

And, in Did You Know, facial recognition can lead to some serious paperwork?

But first, let's walk the tight rope.

♪ ♪ Smile, you're on Candid Camera!

In the 1970s, that phrase meant a hidden camera prank on TV.

Today, the joke may be on all of us.

A 2020 analysis estimated that the average American is captured by security cameras 238 times a week, or about 34 times a day.

The number varies widely, but many Americans are not smiling.

Facial recognition is a form of biometrics, a technology that verifies or identifies people through physical characteristics, such as fingerprints, irises, or facial features.

Research began in the 1960s, but artificial intelligence has made the technology dramatically more powerful.

Today, it performs two very different jobs.

The first one is called one-to-one verification.

That's facial recognition, initiated by you.

With a glance, you can unlock a phone, sign into an app, authorize a purchase, or speed through an airport identity check.

The second one is called one-to-many verification, which means, Who is this person?

It searches an unknown face against a database containing millions, or even billions of photographs, that facial recognition used on you potential without your knowledge Supporters say it can catch travelers using false documents and help investigators find suspects, victims, missing people, or even terrorists, much faster than humans could alone.

Customs and Border Protection says facial biometrics have processed hundreds of millions of travelers and stops thousands of imposters from unlawfully entering the country.

Americans may opt out of facial comparison at an airport, but their photographs may already exist in passport files, some state motor vehicle databases, or commercial systems capable searching billions of images.

Critics warn that cameras could do more than record anonymous people.

They could attach names to faces and reveal where Americans travel, shop, worship, or protest, things that are inherently private.

In fact, the Federal Trade Commission alleged that the Rite Aid pharmacy chain's, facial recognition program, from 2012 to 2020, generated thousands of false alerts and disproportionately harm people of color.

Innocent customers were followed, confronted, or ordered to leave after being wrongly identified as suspected shoplifters.

Rite Aid agreed to a five-year ban on using this technology for security or surveillance.

And a few survey found that nearly two thirds of Americans accepted police facial recognition at large events, but more than two thirds also opposed scanning people simply walking down the street.

So facial recognition can open a phone, but it can also open a can of worms.

Is it innovative, is it invasive, or is it both?

Let's face that question in the center ring.

♪ ♪ Joining me in the center of ring is Dr.

Michael King, Biometric expert and professor at the Florida Institute of Technology.

Welcome, Dr.

King, and let's get right into it.

What kind of information?

Where does information come from?

That fuels the facial recognition software and tech Noll by?

Dr.

KING: Thank you for having me on the show, John.

Well, the information can come from a number of places, and it really depends on the application.

So when you're thinking about travel, for instance, that could actually come from someone's passport or some other identification, such as a driver's license, or whatever is being used to identify someone as they're passing through airports.

Others applications account like access control applications on your in your work environment.

Those could be controlled images that are that are taken, at the time when you're onboarding into an organization.

And the organization maintains control of those.

There are other applications, though, that deal primarily with law enforcement and public safety where those images can come from many databases.

Mug shot databases.

They could also come from driver's license databases in some instances and more broadly, in kind of a growing application area, is when these images are also being scraped from the web and associated with individuals.

KATKO: So, Doctor, I'm going to talk about the last thing you said, and that is, things that are posted in the public domain, on the Internet, for example.

I don't think people realize, if you post stuff on Facebook, for example, picture of yourself, picture of your family, or someone posts a picture.

Is that fair game to be swept up and used for the facial recognition technology?

Dr.

KING: Well, it really depends on your privacy settings that you have on the social media profiles.

And so, if you post an image and and widely available to the public, then it's an area where it's considered to be publicly available information at that point, because we live in a society where a lot of people, kind of want to be seen.

And so therefore, they will open up their profiles broadly and when you do that, you're also susceptible to, other computers basically being able to download those images and store them in some database.

KATKO: So, Doctor, we know kind of the basics now what feeds the system, but let's talk about the upsides and downsides of it, if we can briefly talk about the upsides for a moment, for law enforcement, in general, with facial recognition.

Dr.

KING: Sure.

I mean, I always tell people that you have to think about facial recognition not as one single thing, not as one single algorithm.

So there are face recognition systems that are designed specifically for ushering people conveniently through airports.

They're face recognition systems that are configured for the purpose of access control or people being able to access their computers.

And then there are other systems that are more loosely configured, if you will, to allow variable ranges of quality for law enforcement applications.

And so there have been several instances of face recognition technologies being used by law enforcement for good purposes.

I mean, I tell people all the time that face recognition is an invaluable tool when the only thing you have is an image of the person you're looking for, and what you're looking is any type of lead that you can get or just a person that you can investigate.

You also asked about some of the some of the downsides a little earlier.

And pretty much the downsides is when there's not a fully complete investigation that has actually occurred before law enforcement chooses to engage with someone.

Those could have not so good consequence, if you will, in terms of people being arrested for crimes they didn't commit and for being in areas they were not in.

KATKO: So let's take a look at a couple examples, and we can riff off of that a little bit.

A terrorist is in this country, obviously, he's in here illegally or snuck in, however, he got in here.

We have his picture from the Interpol database, for example.

And he's recognized in a crowd, and he's picked out of that crowd, and he's arrested.

I'm sure that's happened before.

In fact, I'm quite certain it's happened based on my time in Congress and the Homeland Security Committee.

So, just so people understand, there is the upside and downside in one little thing there.

They're scanning a crowd, and their facial recognition software recognizes that guy's face and he's picked out of the crowd.

So, is there... That happens, first of all, right?

And second of all, how is there any protections on how that information is used when you're scanning crowds?

Dr.

KING: Well, it depends on who's actually using it and who's actually scanning the crowd.

So one of the images that I typically use is from the 911 attacks where you see the terrorist coming through the Boston Logan Airport.

And there's an image of this person of course they are looking for him.

But this was in 2001, when the technology wasn't as mature as it is today, and that is, you know, kind of a perfect example of where face recognition technologies can actually play a role.

Because if I gave a TSA person or some security personnel at the airport, some book of people to look for, they have 1000s of people that they need to be looking at and and trying to make some decisions as to whether or not they approach a person or not.

But these are things where, you know, computer technologies are actually ideal for these types of applications because if you're looking for someone, you have the appropriate oversights in place and also the appropriate rules of engagement.

Then these systems don't necessarily sleep.

They're awake all the time.

They're going to apply the same analysis to every face that comes across this camera's view.

And you don't necessarily have to worry about instances of fatigue or distractions setting in.

But the law enforcement personnel and those who secure our airports have Do a superb job, and they continue to do so.

But these tools can actually help them in that instance.

KATKO: So, Doctor, let me have about a minute and a half left, but I want to talk about two things that are potential downsides with it, and that's the accuracy of the technology and the privacy issues.

could you address those briefly?

Dr.

KING: Sure.

I mean, it's widely accepted that these technologies are roughly 99%, 99.9% accurate.

>> KATKO: 99% accurate.

That's important to know.

>> But there are important caveats that actually go along with that.

When you're dealing with high quality images, and and reasonably controlled environments, these system is going to be extremely accurate.

When you're dealing with uncontrolled -- surveillancetype scenarios, you don't actually know what the accuracy is.

And the other thing is the database that you're comparing to, sometimes the person that you're looking for from surveillance perspective, is not even in the database.

And so these systems are geared for saying that this is the, this particular configuration is for saying that this is the person in the database that I was able to find with the most similar characteristics.

It does not say it is the person.

And so we have to keep in mind that there are different configurations for these for these systems.

And in terms of privacy, privacy is a big deal, really depends on who's actually dealing with this data.

I wear more hats than I have ever worn before, and particularly those that actually cover the upper parts of my face.

Privacy is an issue because if the more we're operating in public environments, cameras are basically proliferating all over the place, and we have to do Better with place regulatory restrictions on how this data is being used.

KATKO: Dr.

King, from the Florida Institute of Technology, thank you so much for a great conversation.

You set the table perfectly for the next segment.

Thank you very much.

♪ ♪ Joining me now is Nathan Freed Wessler, deputy director of the American Civil Liberties Union's Speech, Privacy, and Technology Project.

Also Jake Parker, senior director of government relations with the Security Industry Association.

Gentlemen, thanks both for being here.

Mr.

Wessler, I'm going to start with you real quickly.

Obviously, the ACLU has some concerns about facial recognition technology.

So could you just summarize those for us shortly?

WESSLER: We often say that this is technology that's dangerous when it doesn't work and dangerous when it does.

What I mean by that is that this is an algorithmic technology, basically a variant of AI, that's making a best guess about whether an input image might be a match to images in a database.

But it often gets it wrong.

And when it gets it wrong, we've seen it turn people's lives inside out over and over.

We now know of at least 15 publicly known cases of wrongful arrest in states across the country, after police relied on incorrect results from this technology.

Technology also has been shown in audit testing over and over to get it wrong more often with darker skinned people, so people of color and black people, which can accentuate biases in policing.

But even technology that gets it right relatively often poses some very serious threats to privacy and civil liberties.

And we are also attuned to the dangers of police departments in cities, hooking this technology up to surveillance camera networks, which will give the government a really unprecedented power to follow us, identify us, track us, everywhere we go.

Something that we really have never accepted in a free society.

KATKO: Mr.

Parker, your take?

PARKER: Well, really, facial recognition software is simple.

It compares and matches facial images within a database or on a device.

And this technology is actually, it permeates our daily lives, it has become something people use to unlock their phones.

Also to verify their IDs of the airport, and many other, many other benefits to consumers in society.

And there is many different uses of the technology that does this, and you've got to remember that, in all cases, this is making a process that already existed, where ID verification or matching images was required, and done manually, it just makes this process faster and more accurate.

And so, when law enforcement investigations was mentioned, it's actually become a critical piece of investigative tools that are used, and really , importantcongressmen for investigations into things like human trafficking and child sex crimes due to its ability to match images online, for example.

And it's something that's used daily across the nation and that is where you need to compare a photo that's part of evidence to some other data source, whether it's arrest records or other databases.

And so, without the technology, this would have to be done manually, much slower, less accurately.

And what you don't often hear about are the success stories that have resulted of being able to have this technology in a law enforcement context, create leads where all you may have to go on in a case is just a photo from a security camera, for example.

That's all you have to go on.

But the technology creates a potential matches for that identity to be further investigated.

>> KATKO: Like a force multiplier.

So, Mr.

Wessler, I want to come back to one thing you said, and then ask you about a possible balance here.

First, I want to come back to ask you is, there's been a lot of data out there, which tends to indicate that facial recognition software, on a general basis, is about 99% accurate, and I know there's some subsets where it's not as accurate, and you mentioned the African American community.

First of all, if it's that accurate, then can't you fix the algorithms to tweak things, to try and fix the anomalies, which are obviously very important to do Number one, number two, all this in about a minute, you got to respond.

Number two is, what about finding a balance between law enforcement and the goals of keeping the public safe and protecting society?

Can we find that balance?

WESSLER: So on your first question the companies and law enforcement agencies love to trot out these accuracy numbers.

But those are figures in test conditions.

There is a federal agency, that's part of the commerce department that has done rounds of tests of facial recognition algorithms over the years, in controlled test conditions, and those are really designed to show whether the algorithms are getting better over time as a way to support industry in selling us products.

What those tests don't even try to measure is to accurate the whole processes when used out in the world.

When we have tremendous variability in photo quality going in, which makes a humongous difference, the biggest difference, in how reliable the search is, right?

When police are actually using this, they're taking photos of suspects, often from security cameras, say, in a convenience store, where there's irregular lighting, it's from an angle, there's shadows that are irregularly across the face.

There might be a baseball cap, or sunglasses, or a surgical mask, lots of other features.

And then you put on top of that, the demonstrated higher error rates when you've done darker skinned people.

And the algorithms are going to get it wrong a lot of the time.

On top of that, these algorithms do not purport to make matches.

The manufacturers would never say they make a match.

What they do is spit out a candidate list of leads, of candidates who might be matches, and then someone in a police department is supposed to look at that list and decide, do we think a true match is anywhere in there?

Often a true match won't be in there.

Maybe the suspect isn't in the matching database.

They don't have a mug shot in that jurisdiction, or a driver's license in that jurisdiction.

Or maybe the quality of the input photo is just so bad that the system just totally fails.

But the person reviewing it won't know that to a certainty.

And so what we've often seen is police choosing the wrong person out of that list, and then it just taints the investigation.

sends police off in a totally wrong direction, and innocent people get arrested as a result.

KATKO: I'll get back to the second question of time permits, but, Mr.

Parker, I want to give you an opportunity to respond to that as far as the accuracy issues.

PARKER: Well, I completely disagree.

So, we've heard about supposed facial recognition bias is based on information that's very old or misconstrued.

So, typically, 7 or 8 years ago is when a lot of this this research came out that's being described.

If you look at the U.S.

government's test data today, you can see that the leading technologies are extremely accurate, even across demographic variables, over 70 different demographic variables that are tested.

And it's also true that the testing that is done actually is very similar to real world conditions the law enforcement agencies would be encountering because they're actually using images and searcing actual real mugshots.

They're part of arrest records, which is exactly what they do in the field.

And there we can see that over the top 40 algorithms, over 99.5% accurate in, matching photos in a database of many millions of photos.

And so today's information just shows that there's really no, should be no accuracy concerns at this point in technology.

WESSLER: I mean, even if those numbers bore out in real world conditions, which they do not, when you take even a 1% or a half a percent failure rate across 100s of thousands of searches across the country, you're going to have a large, raw number of false matches.

And you put that into police's hands who are not being adequately trained, who have grossly deficient police department policies that are not putting guardrails up, and we see the result.

I have three clients who have been wrongfully arrested because of police reliance on incorrect results from this technology.

There are more people around the country.

This is just a technology that has proved to be dangerous when it's actually operated in the world.

KATKO: Mr.

Wessler, I want to give Mr.

Parker an opportunity to have the final word, about 30 seconds.

PARKER: To put this in perspective, unfortunately, wrongful arrest is something that occurs every day in our country, every single day.

And it is often a result of incomplete or failure in police work to take leads and actually find independently verified evidence, to act on.

In every single one of the only 15 cases that were mentioned, that's exactly what happened.

There was a breakdown in police procedure.

This can happen when leads come in by other means as well.

KATKO: Well, gentlemen, we could talk about this for hours, and appreciate the intelligent conversation here.

Nathan Freed Wessler, of the ACLU, and Jake Parker of SIA.

Thank you so much for a great conversation.

♪ ♪ Facial recognition technology is both a sword and a shield.

It can protect the public from dangerous people and make us safer, but it can also be exploited by government and private companies, in ways we may never have imagined.

The cold hard reality though is this technology is here to stay.

We already have the pieces of oversight in place.

The federal government test facial recognition systems, agencies maintain their own policies, and regulators can act when companies misuse biometric information.

But those responsibilities are scattered, and no single national framework governs how facial recognition may be used across government and private industry.

Lord knows we don't need another sprawling bureaucracy.

Instead, we need Congress, technical experts, law enforcement, industry, and civil liberties advocates to work together on clear national standards, and determine which existing regulators should enforce them.

That process must be transparent.

Scientifically grounded and insulated as much as possible from partisan pressure.

It's time to face the facts of facial recognition technology.

And that's my take.

♪.

♪.

Did you know, in 2017, officials at Beijing's Temple of Heaven Park installed facial recognition toilet paper dispensers in public restrooms.

I'm not kidding.

Why?

Because visitors were taking large amounts of free toilet paper home.

The machine scanned each visitor's face before dispensing a strip of paper roughly two feet long.

Anyone needing more had to wait nine minutes before the machine would recognize them again and release another serving.

That's right.

Even the TP dispenser was keeping a paper trail.

And now, you do know.

♪ ♪ That's all for this week, folks.

To send in your comments for the show, or to see "Balancing Act", extras and exclusives.

you can follow us on on social media or go to BalancingActwithJohnKatko.com.

Thank you for joining us.

Remember, in the circus that is politics, there's always a Balancing Act.

I'm John Katko.

We'll see you next week, America.

♪ ♪ ♪ With Pomeroy Markers, history is based on the facts embedded by professional historians, over 3,000 installed nationwide, from Maine to Florida, to California, to Minnesota, Texas, and Alaska.

Since 2006, our markers have helped people celebrate community history.

Marker grant program information available at wgpfoundation.org.

It takes time to craft a piece of furniture and to craft a 125 year legacy.

It takes patience to execute every detail the right way.

And it takes family, working to bring out the best in each other over generations.

This is Stickley, the furniture that serves generations of family.

Maturing with time, rewarding patience, and celebrating every detail.

This is what excellence means to us.

♪ ♪