Skip to main content

tv   House Hearing on Facial Recognition Technology Uses at Homeland Security...  CSPAN  February 18, 2020 3:21pm-5:31pm EST

3:21 pm
and even though they came at it from a different place, a both new that they cared about the country and they loved it. they just developed a bond and they found this little nuggets of common ground where they could work together. >> tonight on c-span senator kaine the winter of senator john mccain, in vicki kennedy, the widow of senator edward kennedy discussed their husbands careers and legacies in a conversation about public service at the kennedy school of government at harvard university. >> next a look at how the customs and border protection agency uses facial recognition in biometrics technology to identify people coming to u.s. pirg house homeland security committee heard from john wagner the customs and border protections deputy executive assistant commissioner of field operations. [inaudible conversations]
3:22 pm
the committee on homeland to get he will come t to order. let me set the outset number of our members are still en route from the prayer breakfast this morning. they will join us accordingly. the ranking member being one of them. the committee is meeting today to receive testimony on the department of homeland security use of facial recognition and other biometric technologies. without objection the chair is authorized to declare a committee in recess at any point. good morning. the committee is meeting today to continue examining the department of homeland security is use of facial recognition technology. the committee held part one of this hearing in july of last
3:23 pm
year, after news that thet department was expanding its use of facial recognition forth varying purposes, such as confirming the identities of travelers, including u.s. citizens. as facial recognition technology has advanced, it has become the chosen form of biometric technology used by the government and industry. i want to reiterate that i am not wholly opposed to the use of facial recognition technology, as i recognize that it can be valuable to homeland security and serve as a facilitation tool for the department's varying missions. but i remain deeply concerneded about privacy, transparency, data security, and the accuracy of this technology and want to ensure these concerns are addressed before the department deploys it further. last july, i along with other members of this committee, shared these concerns at our
3:24 pm
hearing and left this room with more questions than answers. in december 2019, the national institute for standards and technology published a report that confirmed age, gender, and racial bias in some facial recognition algorithms. nist, for example, found that depending on the algorithm,io african-american and asian-american faces were misidentified 10 to 100 times more than white faces. although cbp touts that the match rate for its facial recognition systems is over 98%, it is my understanding that nist did not test cbp's current algorithm for its december 2019 report. moreover, cbp's figure does not account for images of travelers who could not be captured due a variety of factors such as
3:25 pm
lighting or skin tone, likely making the actual match rate significantly lower. these findings continue to suggest that some of this technology is not ready for prime time and requires further testing before widespreadme deployment. misidentifying even a relatively small percentage of the traveling public could affectoy thousands of passengers annually, and likely would haveu a disproportionate effect on certain individuals. this is unacceptable. data security also remains an important concern. last year, a cbp subcontractor experienced a significant data breach, which included traveler images being stolen. we look forward to hearing more about the lessons cbp learned from this incident and the steps that it has taken to ensure that biometric data is kept safe. transparency continues to be
3:26 pm
key. the american people deserve to know how the department is collecting facial recognition data, and whether the department is in fact safeguarding their rights when deploying such technology. that is why we are here seven months later to continue our oversight. i am pleased that we again have witnesses from cbp and nist before us to provide us with an update and answer our questions. we will also have testimony from dhs's office for civil rightsn and civil liberties. this office is charged with ensuring the protection of our civil rights and civil liberties as it relates to the department's activities, no easy task, especially these days. be assured that under my leadership, this committee will continue to hold the department accountable for treating all americans equitably and ensuring that our rights are protected.
3:27 pm
i look forward to a robust discussion forum all the witnesses at a thank the members for joining us today. i welcome our panel of witnesses here our first witness, john wagner, , currently serves as te deputy executive assistant commissioner for the office of field operations, u.s. customs and border protection. in his current role he oversees nearly 30,000 federal employees and manages programs related to immigration, custom, and commercial trade related to cbp missions. mr. peter minute is a deputy officer for program and compliance at the office of civil rights and civil liberties here he previously served as chief of the labor and employmentpr law division for u. immigration and customs enforcement. dr. charles romine is the
3:28 pm
director of information technology laboratories at the national institute of standards and technology. in this position he oversees a research program that focuses on testing and interoperability, security, usability, and reliability of information systems. without objection, the witnesses full statement will be inserted in the record did i not ask eah witness to summarize his statement a for five minutes beginning with mr. wagner. >> good morning. chairman thompson, ranking member rogers, members of the committee, thank you for the opportunity to testify before you today on behalf of u.s. customs and border protection. i am looking for to the opportunity to discuss the recent nist report with you. since cbp is using an algorithm for one of the highest performing vendors identified in the report, we are confident our results are corroborate with the findings of this report. more specifically the report indicates while there's wide range of performance of the 189 differentfo algorithms that nist
3:29 pm
reviewed, the highest performing algorithms had minimal to undetectable levels of demographic-based error rates. the report also highlights some of the operational variables that impact their rate the chess gala precise, photo page, photo quality, numbers of voters of each subject in the gallery, camera quality, lighting, human behavior factors, all influence the accuracy of an algorithm. that's why cbp is carefully constructed operational variables in the deployment of the technology to ensure we can't attain the highest levels of match rates which remain in the night seven-98% range. one important note is nist did not test the specific cbp operational construct to measure the additional effect these fables may have which is why we've entered into an m.o.u. with nist three by which are specific data. as we build out the congressionally mandated biometric-based entry-exit system, we are creating a system that not only meets the security mandate but also inly a way that is cost effective, feasible and
3:30 pm
facilitated for international travelers. identity requirements are not new when crossing the border or taking international flight. several existing laws and regulations require travelers to establish their identity and citizenship whenla entering and departing the trendy pick cbp employs biographic and biometric these procedures to inspect the travel documents to verify the authenticity of the document and determine if it belongs to the actual person presenting it. again these are not new requirements. the use of facial comparison technology center automates the process that is often done manually today. ..
3:31 pm
passport, it makes sense that the technology may be useful in determination of the d rightful document holder. it's more difficult today to forge a legitimate passport insecurity features are much stronger than they were ten -- 15 years ago but we are stuck vulnerable to a person using legitimate documents that is real and belongs to someone else. using facial comparison technology to date, we identified 252 imposters to include people using 75 genuine u.s. travel documents. privacy continues to be our mission.
3:32 pm
it's compliant with the terms of privacy act of 1974 as amendment, the 2062 departmental policies and government collection and maintenance. personally identifiable information. cbp recently published updates and the privacy impact assessment covering this program. systems of record notice system publish on the database in storing the information. we've met three times with representatives of the community as well as discussions with privacy liberties oversight board and privately privacy committee. the office of management and budget, a rulemaking that would solicit public comments on the regulatory updates and amendments to the federal regulations. one final note, our private sector partners, the airlines and airports must agree that documented specific business requirements they are submitting to photographs as part of this
3:33 pm
process. these requirements include a provision the images must be deleted and transmitted to cbp and may not be retained by the private stakeholder. after the attacks of september 11, we as a nation asked how we can make sure it never happens again. part of the answer, the report commended the dhs should complete as quickly as possible a biometric entry exit screening system and it was an essential investment in national security. gbp is internet volunteering at the duty by strengthening biometric efforts verifying that people are who they say they are. thank you for the opportunity to appear today and look forwardtio your questions. >> thank you. i recognize you to summarize your statements for five minutes. >> good morning.
3:34 pm
thank you for the opportunity to appear before your directors discuss the home and security use of facial recognition technology. commitment to not determination and activities remain important cornerstone of our daily work. i'd like to make three points in my testimony today.ra first, the civil rights and liberties has been and continues to be engaged with dhs operational components to ensure use of facial recognition technology consistent with civil rights and liberties. law and policy. second from operators researchers and civil rights policymakers must work together to prevent algorithms from leading to biases and facial recognition technology. third, facial recognition technology can serve as an important tool to increase the effectiveness of the department mission as well as a facilitation of travel but it is vital that these programs utilize technology in a way that safeguards are rights and values. to achieve these, one,
3:35 pm
influenced vhs policies and programs throughout the cycle. engage with components in the development of new policies in the program to assure protection of civil rights and so forth liberties are fully integrated into the foundation. three, monitor operational execution and engage with stakeholders to provide feedback regarding consequences of policies and programs.ce fourth, we investigate complaints and make recommendations to the components, including allegations of racial profiling or other visible bias. they recognize the potential risk for bias and algorithms. they support rigorous testing and evaluation of algorithms used in the system to identify and mitigate visible bias. they will continue to support the relationship between the institute of standards and technology, the director, dhs office of biometric and
3:36 pm
components including border protection. in carrying out the mission, they advise the components of the offices by participating in enterprise level groups working on facial recognition pictures. further, we are directly engaged with the components read for example, they regularly engage them on implementation on the facial recognition technology. they advise on policy for appropriate accommodations for individuals with their objection of being photographed and for individuals who may have visible injuries, prevent charges. the facial recognition program, they will collaborate directly with them to address potential civil rights and liberties
3:37 pm
impact. further, to engage communities to both inform the public regarding facial recognition program and address potential concerns. finally, we will continue to evaluate any potential violation of civils rights and liberties o further inform our policy advice and strengthen the recognition program. successful appropriate facial recognition technology requires ongoing oversight and quality assurance. initial validation and regular revalidation and relationship between users and oversight office, this way it can be developed to work properly and without bias, achieved operating capability and continuing throughout the lifecycle. at the same time, we need to work with the operational components to ensure practices from the human part of the equation, users are focused on the technology, working in a manner that prevents bias,
3:38 pm
engaging in the activity. again, thank you for the opportunity to appear before you today and i look forward to answering your questions. >> thank you for your testimony. i recognize you to summarize your remarks. >> thank you. i am chuck, the director of the laboratory of the standard technology. nist. thank you for the opportunity to before you today for standard of testing of facial recognition technology. this has been working with public and private sectors since the 1960s. biometric technology provides a mean to establish or verify the identity of humans based upon one or more physical or
3:39 pm
behavioral characteristic. for self recognition current technology compares facial features for verification or identification purposes. this work improves the quality, usability, and consistency ofus systems and ensures that u.s. interests are represented in the international arena. this research has provided state-of-the-art technology benchmarks and guidance to the industry and u.s. government agencies that depend upon biometrics recognition technologies. the face recognition testing program provides technical guidance and scientific support for analysis and recommendations for utilization of face recognition technologies to various law enforcement agencies including the fbi, vhs, and i are both. this report released in december 2019 quantify the accuracy of face recognition algorithms or demographic groups defined by sex, age and race or controversy or both want one and one too
3:40 pm
many identifications algorithms. they found evidence for the existence of demographic facial recognition algorithm that nist evaluated. it is temperatures between false positive and negative errors and note the impact of errors are application dependent. nist conducted test to quantify the differences for 189 face recognition algorithms from 99 developers using four collections of photographs with 18.27 million images of 8.49 million people. these images came from operational databases provided by the statete department, department of homeland security and fbi. i will firstan address 11 verification application. false positive differentials are much smarter than those related to false negatives. they exist acrossch many of the algorithms tested. false positives might present
3:41 pm
our concern to the owner as they may allow access to imposters. other findings of false positives are higher in women than men and higher in the elderly and young compared to middle-aged adults. regarding race, we major higher false positives in asian and a african americans relative to those caucasians. the ohio false positive native americans, american indians, alaskan indians and pacific. these effective apply to those developed europe and united states but it was algorithms developed in asian countries. there is no such dramatic differences in false positives and one to one faction between asian and caucasian faces or the algorithms developed in asia. while the study did not explore the relationship between cause and effect from one possible connection area for research is the relationship between algorithms performance and for data used to train the algorithm itself. how comments on one too many algorithms.
3:42 pm
false positives in one too many search are particularly important because the consequences could include false accusations. for most algorithms, the study measured higher false positives in women, african americans in particular african-american women. some algorithms gave similar false positives across these demographics. it was accurate algorithms fell into this group. this underscores one overall message of the report. different algorithms perform differently. indeed, all of our reports note variations in recognition across algorithms and important results from a demographic study is a demographic effects are smaller with more accurate algorithms. this is the positive impact it's had in the last 60 years inve te evolution of biometric capabilities with nist extensive experience and expertise in laboratories and successful collaborations in the private
3:43 pm
sector and other government agencies, this is actively pursuing the standards and measurement research necessary to deploy interoperable secure reliable and usable identity systems.oy thank you for the opportunity to testify. i'd be happy to answer any questions you may have. >> thank you. i remind each member that he or she will have five minutes to question the panel. i now recognize myself for questions. part of your report was like next-generation technology in the cbp used in your existing technology. we are not certain about, we intend to continue our investigations though the
3:44 pm
existence of a specific algorithms that we have, they were submitted by the offenders, we have no independent way to correlate whether the algorithms being used in the field. >> part of what you said is how the technology is deployed depends on the application of the technology. explain that a little more to the community. >> certainly. our c approach is the significat thing to be cognizant of is the risk associated with deployment and the studies we do help inform policymakers such as members of congress as well as heoperators of these technologis about how to quantify those risks at least for the algorithms themselves.
3:45 pm
the deployed systems have other characteristics associated with them that we don't pass, only the algorithms currently. the that risk that comes from the error rates associated with algorithms as part of a much larger risk management the operators have to undertake. for example, access to critical infrastructures is different than access to a phone you might have. the risks are different. >> thank you. mr. wagner, if you could share with the committee the extent cbp goes to protect the information collected in this process. >> .the photograph taken by one of our stakeholders cameras, they are encrypted, transmitted
3:46 pm
securely to the cbp cloud infrastructure, the gallery is positioned, the pictures are alphabetized, they are turned into some type of mathematical structure that cannot be reversed in fairtu matchup with the photos we have in the gallery and response goes back yes or no with a unique identifier. >> so the comment that two to 3% of people who are misidentified, what is cbp doing to try to get that? >> is not that they are misidentified, it just means we didn't match them to a picture in the gallery that we did have of them.
3:47 pm
we should have matched him. you're right, i should be at zero. that's where we look at the operational variables. the camera, picture quality, the human behaviors when the photo was taken, the lighting, those different types, from the age of the photo and what we've seen ii the report, gallery size impacts it.y the 12 million from we are comparing against a few thousand here at most. the numbers of photos we have of particular individual can impact which one we match against the age of the photo. if you have your best photo taken at 20 and you are now 29, if your faith has changed, we are going to have a match against that which is compounded mp poor fighting of the conditions or the movement when the photo was taken or poor quality of the photo. >> listen to what you just heard, how have you dealt with
3:48 pm
complaints from citizens about this technology? >> we've received one complaint however, we have not seen a trend, that's when we would open an investigation. we are working on the policy side of the house, the other way in which we also hear from the community through our round table throughout the country and we have a forum about facial recognition technology and those are concerns and reason to inform our advice. >> can you provide the committee with work you have the forms around the country? >> absolutely. we have, we do round tables and about 18 cities. certainly in some. we will continue to have those
3:49 pm
discussions with cbp in the future at future roundtables. >> thank you. lastly, mr. wagner, i'm not sure you have information on this but last month the lebanese nationals and individuals who traveled to lebanon, most of whom were u.s. citizens, a green card holder were targeted, detained and subjected to prolonged questioning of up toat 12 hours at the port of entry. internal cbp memo, people are questioned based on their religion. which is completely unacceptable. they admitted to enormous mistakes in this incident. if you know how the situation happened, what is cbp doing to ensure it never happens again?
3:50 pm
>> there's no national directive or guidance that went out other than because of the things taken place in iran concerns about retaliation, we put our field managers on to be more vigilant about current events happening and work with your local state counterparts and really just be vigilant. there was more prescriptive guidance that went out at the local level in washington, which we are reviewing right now because there's a lot of concerning things that we saw in the interpretation of that guidance and management oversight of the weekend was unfolding and people are being worked been referred and for questioning and concerning.about the management engage mental or lack thereof of what transpired. internalof investigation the cdp
3:51 pm
is inducting so liberties is conducting an investigation and would make it a result of that, we'll proceed depending on what the results say. >> are you aware of that? >> yes. we do have an open investigation in this matter. >> thank you. >> yield to ranking member for opening statement. >> sorry for being late. we just got back. thank you. after the tragic event, congress recognized they are essential to homeland security. congress charged the creation of automated metrics system. translation security mission already showed the capability for security, travel and tour
3:52 pm
the existing immigration laws. government and private sectors, government and the private sector has made enormous strides in the biometric systems. the technologies of all times has seen improvements. these advantage in facial recognition algorithms in particular are transformational. the institute of standards and technology is the leader in testing evaluation technologies. they have done incredible work to help congress, dhs and industries understand the capability of currently available algorithms. i'm concerned some of my colleagues have had the misleading conclusion of the report called facial recognition, hours after being released from the majority belief the report shows facial recognition racially biased and
3:53 pm
we feared if the majority had taken the time to read the full report, they would have found the real headline missed determined the facial recognition algorithm being adopted as no great gender bias. in other words, there's no potential evidence that facial recognition algorithms in racial bias. i hope my colleagues will listen as hewr explains the most accure algorithms. the reality is that facial recognition technologyes can improve by reducing human error. the technology cannot and will not replace the cdc or tsa officers. concerns regarding the civil rights but these concerns can be fullyy addressed how systems implement it dhs. to hearing the steps rico is taking to ordinate and predict the civil rights of americans. as i have said before, all
3:54 pm
government programs is not a solution. technology is not violating their rights. most important, it does protect the home and. i appreciate the chairman calling this today, they want to further implement these issues. i yield back. >> thank you. there is some testimony contra contrary. >> my statement is wrong. anybody can jump at it. >> i would never tell congress they areer wrong. >> you are one of the few people. [laughter]
3:55 pm
literally, to my understanding, there's no statistic circle evidence that those racial bias. is that an inaccurate statement? >> in the highest performing algorithms for one too many matches, the highest performing algorithms we saw undetectable, the bias in the demographic differentials that we were measuring are undetectable. >> what you mean by undetectable? >> in the testing we undertook, there was no way to determine, the idea of having absolutely zero false positives is a big challenge. >> the algorithm being used? >> we tested algorithms, we have
3:56 pm
no way to verify that is the specific algorithms using by cbp. that would be something cbp and an ec would have to attest to. from our perspective, they provide algorithms, black boxes that we test just the performance of the algorithm submitted by the offender. >> cbp currently working to implement any algorithms? and easy algorithms? >> we are using an earlier version and we are testing an ec three, the version tested and the plan is to use it next month in march to switch to that one. >> who can participate in facial recognition? is accurate to r say some algorithms are accurate, more so than others? >> that's correct. anyone can participate. we have participants>> from industries biometrics industries
3:57 pm
around the country but also from universities and some experimental systems as well. >> thank you. you have operational airway in the country, right? technology goes into being you said -- >> the nec three is next month. the earlier version is operational now. >> the one we are talking about is not? >> correct. >> you mentioned african-americans and asians get misidentified? >> in the highest performing algorithms, we don't see that to a specific physical level or one
3:58 pm
to many algorithms. we do see one-to-one algorithms, we do see evidence of demographic for african americans or asians. >> chair recognizes miss -- >> thank you for clarifying that. it was hard to understand from your testimony so just to be clear, can you pronounce your name? so in aso certain segment of the algorithms, there is some evidence that they have higher rates of mistakes for african americans and asians. asian americans. >> it is correct that most of the algorithms in the one too many submitted, do have the
3:59 pm
differentials. the highest performing's and the one too many do not. >> spoke some do and some do not. i'm just trying to clarify. thank you all for t being here. i'm from michigan so we have a long history of meeting our cbp officers to protect us in all of orour bridges and crossing. you help me understand, is six technology used in any way at our bridge crossings in the northern border? >> no, not at the bridge crossings. >> but at the airport. so while i recognize it seems te be a small number of times of these programs were theirs, they've detected more problems with particularly african-american women in asian americans, what's the process where it would be, you are an
4:00 pm
average citizen, you are an african-american woman. let's say we employ this technology and it shows a positive. think. walk me through that process and how he would deal with that at the actual border. >> you would choke your passport, which is what you do today in a person would manually review it if you did not match. >> if they show the passport affect technology still showing a match, what does the officer do in that situation? >> if the machine is saying one thing and the passport sayingnt another. >> we would go on the faces of the document presented in which photographs we've identified or which identity we've identified you with. >> that would go on -- i'm just asking for the average person understand? >> the passport photo, we have an electronic copy of that. >> excuse me.
4:01 pm
that's disrespectful to the hearing. please. >> we are flying into the country, that's what we match you again. the photograph which should be printed on your passport and should be on the electronic passport, they make sure you are fat same person. if it doesn't match, then we have to figure out why. >> when you figure out why, is the individual allowed to progress? we go to see a concert. >> it could be as simple as looking at your passport document. we'll figure out later what happened. >> what happens with that data? essay a woman has gone to her doctor in canada, what happens to her data where it flagged that she matched against, falsely flagged against someone who did something wrong. what do they do with that information? >> if you're a u.s. citizen, the
4:02 pm
new photograph we've taken is discarded after 12 hours. there's no reason to keep the photograph. as a record of the transaction across the border. a if there's some type of error and they will credit and corrected basically. if you matched, your name, date of birth, to the wrong person, your biographic matches identical to someone else, that's where we can use the facial recognition to distinguish between the people with common names. we put notes in the system and for the officers to suppress the information we get your passport the next time. >> how long have you been y implementing it at airports, tell me what the results are. how many people have you identified in a positive way that needed to be identified? tommy's statistics, demonstrate
4:03 pm
the value of these programs. >> 43.7 million people we run through all the different locations in cruise ships, pedestrians, we have caught 252 imposters, people with legitimate travel documents, i think 75 of those were u.s. travel documents, travel documents, the only biometric we have is a photo that state department put on the electronic ship. there's no fingerprint record or requirement to get a u.s. passport. there isn't one there so the only biometric we have the travel document is that digitized photograph. as a worldwide standard. that's not allowed to be opened by any country participating in the scheme can access it and pull off the digital photograph and do a comparison to that. >> in my remaining time, tens of
4:04 pm
millions of people you views have gone through the technology, tell me more about your stats. how many positives, how many negatives hit? >> our factory is about 97 -- 98%. two to 3%,en that generally meas we can find that person in that preassembled gallery. we didn't consent anything. that's the wrong person from which is didn't find a match of people traveling. it could be b environmental or operational reasons -- >> how many were false positives? >> there may be a small handful, i'm just not aware of them, not of any. as we build and test this, we are not seeing that. >> i think my time has expired. thank you. >> 9/11 tradition recommended the use of biometrics for those entering and leaving the united states and i believe
4:05 pm
technologies are good and stopping terrorists and we've seen it time and time again and my first question is, my understanding is that the entry exit program, american citizens can't opt out the program, correct? >> yes, correct. >> there's no requirement they have to be subjected? >> no but people have to establish their identity. once we determine the review of the passport or by using technology, they are a u.s. citizen, they are excluded from tracking requirements. >> it's like we use the global entry, most of my constituents love global entry. i have a clear program, associated with tsa, put your fingerprints down and gets ahead
4:06 pm
of the tsa. they made it easier for the traveling public but also the great thing is, it doesn't lie. the last congress we passed on the committee, the biometric identification transnational migration of our program, i know this is not cbp, the task overwhelmingly bipartisan way, it reauthorized the program and obama administration that secured jake johnson and i talked a great deal about how can we identify when this people are coming in, they may change their names multiple times.
4:07 pm
month to get to the united states, together facial recognition biometrics don't. this has been a very successful program and keeping terrorists, human traffickers and bad actors out of this country. in fact, this program has enrolled over 155,000 my encounters are persons of interest in 460 known and suspected terrorists including arresting violent criminals and racist involved in transnational criminal organizations, so mr. wagoner, can you comment on why that program is so valuable? >> it's critically important commented, people do change their biographical details. most are watchlist searches, bio graphically based but if we can
4:08 pm
identify people, especially people traveling through the air, we have national security concerns about it and entering our hemisphere. we can work with our partners down there. establish on a biometric basis who that person is so mattery what identity they show up and later, if they show up on the us-mexico border and from the biometric conversations, when they first flew into the hemisphere. >> and travel documents can change and passports are stolen and manufacturers. that's not accurate biometrics don't lie. >> people change documents, borrow documents and purchase documents. it's harder to alter them now but the ability to get a legitimate document that looks like you and if you can pass
4:09 pm
inspection collecting it in a photograph on it, then yes. >> it's unfortunate the senate did not pass this bill they stole. bipartisan way to test congress into that's unfortunate. i would hope we get past this bill again with congress. we have to look at civil liberties and privacy as well. i do think this is entry exit opt out. it applies primarily to americans with one option and foreign nationals, it applies really to foreign nationals themselves. i want to thank the witnesses for testimony. thank you for having this hearing. >> chair recognizes the young lady from new york. >> thank you.
4:10 pm
i think our ranking member and expert witness who testified before today. it's time we face the facts. unregulated facial recognition is just not an option. we can debate and disagree about the exact situation where we permit the use of facial recognition but we should all agree there's no situation where facial recognition should be used without face cards advanced bias for privacy. right now in termsy. of recognitionfacial is still in the wild west. meanwhile, facial recognition technology are routinely this identifying women and people of color. although there are some promising applications for facial recognition, these benefits don't outweigh the risks of automating discrimination. we see what happens when technology is widely and, they can impose meaningful safeguards. let's look before weekly. mr. wagoner, some of our staff
4:11 pm
observed issues with facial recognition technologies, for example, we've seen passengers and particularly, not able to b matched due to poor lighting or other factors. the cbp tracks how often it's the system failed to capture photos of sufficient quality for matching. >> we track the number, we don't own all the cameras. it's difficult for us to track one airline, how many pictures i might be taking before they submit one for matching. in the departure environment, they on them. we are tracking how many pictures we receiveck and match rates against them. >> i was wondering about the quality because of the photo quality isn't goodbe enough, the algorithm is irrelevant. >> absolutely. the picture has to be of this
4:12 pm
quality. >> track the numbers of photos that don't meet your standard. you said you have all these other partners using their materials that now you are dealing with something that has become irrelevant. if you don't know what subset of those don't meet your quality control. >> we know the pictures they transmit to us. >> rightht but. >> we don't know how many items they've made. >> you normal the quality or how much of that, what percentage of that doesn't meet our standard. >> we look at the number of passengers. >> how do they plan to address these issues to ensure they capture high-quality images of travelers were successful facial recognition? >> we have a high performing algorithm, how we look at the operational variables to make
4:13 pm
that even for high-performing. >> what i would say to you is until you met that standard,t you're not doing of quality control. >> we are developing that. what we are seeing, 97 -- 90% rate -- >> wetland gone to another question. when you are at 3%, it doesn't matter. >> we are not seeing demographic based in the 3%. they help us understand that better. >> i understand this is our last operational testing in the entry exit system of the airport. the results indicated the accurately matched images when captured but the rate in capturing an image were significantly lower than
4:14 pm
expected. 80% compared to 97%. >> most of these issues were attributed to airlines referring to manually processing passengers in the boarding process. are you aware of the timing? >> yes. that's as we were developing the operational variables to look at them. one doesn't even work. can we make it work? now we look at and work with the airlines do not shut down the boarding. what is the ease of the application of the traveler?hu >> quickly, what steps are they >>king to work with airlines with these images? >> one is publishing the regulation which would then put the requirement onto the foreign national who has to comply withs the biometric congressional mandate. then we can work with both leaders to increase -- >> can you provide those steps? that would be helpful.
4:15 pm
i've run out of time. you spoken your testimonyn about impermissible bias. i was just wondering since you use that u technology, is the inverse, is that part of your, is there something called permissible bites? >> i think if i understand correctly, the reason why we use that term is because there are a lot of reasons why there may be an act, providing an environment but we are focused on an error that's created based on a particular characteristic like sex or h. that's what i'm referring to. >> there's no bias that it is permissible? if there is a quirk of some sort and you find it to be so
4:16 pm
inconsequential that it becomes part of your standard, that becomes permissible? i'm just trying to understand. >> one of focusing on is what is prohibited by law that our offices look at which is really based on projected characteristics. trying to cpp are eliminate any bias, any reason. however, in terms of what we do at the policy office, we are focused on potential for bias in those protected areas. >> thank you. >> chair recognizes the gentleman from new york. >> thank you. i appreciate you having this hearing. i commend all my colleagues for all these questions because it's important. when i was first prosecutor, dna evidence was this weird science
4:17 pm
think that nobody knew we knew about. as i got better, it became a very potent tool not just for law enforcement to exonerate people who were accused of crimes. the biometric technology is taking a similar role in it will help law enforcement and also do a dramatically good thing to prevent his identification. one of the things i heard today, the highest performing algorithms have no statistical anomalies. that means that algorithms will get front lines that encourage you to get them to the front lines quickly. never let your guard down and always follow the problems of the system and make it better. in the end, we are all going to benefit. i trust my colleagues to ask the
4:18 pm
appropriate questions. i have to ask something that occurred yesterday was very important. my constituents in general but the new york state internal. the 25th 2020, they can no longer have drivers licenses as part of the formula. that's because new york state under the green light law, which they passed, forbids access by cbp and ice through the database. would you please summarize for us and asked it to be into the public record, being incorporated into the record. >> okay. my understanding is new york state because of the law they passed, without the access to
4:19 pm
motor vehicle data which included drivers license information, license plate registration, vehicle registration and information, i our operations, any of the work we do where we would use that information to help validate an identity and address in a vehicle the ownership of the vehicle is impacted by not being able to do that directly. the breath of our mission goes way beyond what the loss is about immigration reinforcement. national security mission and all the other areas in which we operate. >> is there any other state in the country that is having this problem? >> we've worked some other agreements with other states to continue to access the data for the work we do. >> and my understanding that new york state customs and border protection's for ice and access to driver databases? >> yes.e
4:20 pm
>> even california? >> we have a separate agreement with them for continue access to the information. >> i want to know some of the things, tell me if this is correct. on a daily basis, ice uses the data in an effort to combat transnational narcotics smuggling, human trafficking and otheran contraband, child exploitation, child exploitation? technology, fraud and identity theft.ef it's fair to say i don't have access to the database that hampers the investigation at times. >> any law enforcement practice where we you would use that information would be impacted. >> i yield back. >> thank you. cbp was made aware of the policy
4:21 pm
before the announcements? >> yes. >> so you were aware of it, about blocking access. >> i don't know. >> personally, we received an influx about questions about the policy overnight. the thousand new york state residents were affected. they have pending double entry enrollment applications for renewal. this is going to have an enormous impact on people. many of whom entered into this program because their jobs ededire them to travel internationally.
4:22 pm
what you plan to do about all those people who be impacted? >> without the ability to help validate their identities through -- >> you have theirli fingerprint. >> if they haven't been rprested, then i have that. what will the finger prints tiger? >> what are you trying to find out? >> trying to validate their address, where they live, these are things important as we establish that low risk traveler status that we avoid people in that program. how would we do that? so new york state shut off without consultation, our access to the information in december. how would we continue to operate and validate who people are? >> going forward, what about the people who already havee t it? i have global entry. >> -- i will be able to do that when i go to the north. here i am a sitting congresswoman. so to me, i understand the
4:23 pm
distinction you're making, your think you have individual agreements with all ofua them wl have access to this database. >> are not aware of any other state blocking our access to information. >> i would like you, we will follow-up. there are 15 other states that allow undocumented people to get drivers licenses. >> i was not allowed aware of them blocking our information. >> you not being aware is not a sufficient answer because there could be other states that do. it seems to me this is once again an attempt by this administration, specifically donald trump who formerly lived in new york to punish new york. you either follow up on that, i appreciate you trying to answer these questions but we need more
4:24 pm
information. i appreciate your attempt to answer and i yield back. >> thank you. can a person have global entry without a drivers license? >> yes, i believe so. >> i'm trying to figure out how you will get all these people. some of them don't even drive. >> it's a new york state identification. >> but they have a >> it validates the address of where they live. >> my drivers license has a po box. i'm just trying to figure out,. >> why is the information box
4:25 pm
for this purpose then? >> i don't know, i'm saying why would you cancel it? >> where they block this information? >> identification or security? improvement with other documents.seth >> care recognizes the gentleman from louisiana. >> i yield one minute to my colleague. >> thank you. a quick follow-up question with my colleague, it's a simple one. first of all, it's clear the investigation clears the ability to get hundred and of dictation available in the dmv database in new york state. my colleague from new york and many other states that have
4:26 pm
possible drivers licenses. that's not the issue. the issue is, is there any other state in the united states of america that completely blocks customs and border protection and isis access to dmv records? >> i don't believe so. >> in my opinion, i do not believe this is a political exercise. all new york would have to do is enter a similar agreement with customs and border protection and ice they simply verify it won't use it for immigration enforcement purposes. i use it for law enforcement purposes and global entry into those things. correct? >> i think that's a discussion we would have the state. >> thank you. i yield back. >> a follow-up on the new york, it's a fascinating document.
4:27 pm
are you aware of negotiations or communications prior to the new york legislative body with customs and border protection? >> our access was just turned off one day in december and are officers and agents in the field : and said what happened to our access? >> as far as youcc know, you can certainly advise this, but as far as you know, is there an ongoing communication during the course of the development of thisn legislation in the state f new york with law enforcement agencies like border protection and ice? >> anna. i don't know. >> you mentioned blind box
4:28 pm
protecting. facial recognition provided by vendors that tested without your knowledge of who the vendor is, strictly looking at the results of a algorithms themselves. >> when i used the phrase black block testing, we don't have any insight into the algorithm itself. we publish api. >> you know the identity of the offender? >> just to clarify that, can any vendor submit an algorithm for testing? >> yes. >> in the process by submitting that product is the standardized? >> it is. >> the top-performing algorithm like customs and border
4:29 pm
protection use, is there a wide variance between what you refer to performing algorithms and say academic projects perhaps submitted for testing? >> yes, there's a wide variance of algorithms at the top. >> in your scientific assessment of this testing and evaluation of facial recognition technologies, would you say what we arese referring to as the algorithms used by customs and border protection are far and beyond some of the common products presented here? >> they are significantly bett better. >> can you confirm for this committee that it is a topir performing algorithms at this time is being used by federal law enforcement agencies?
4:30 pm
>> i have no way to independently verify that. >> they are using the top -- >> on to confirm that. >> i didn't say that.ha we are using the previous version of it. using a high n performing -- >> all right, i think that vaguely answers my question. your time is expired. >> chair recognizes the gentleman from new jersey. >> thank you. who and where does all this facial recognition data store, please describe under specific circumstances they allow to be
4:31 pm
shared or used or transferred if that is the case? >> we are using as a database travel document.databases. these are photographs collected by the u.s. government for the purposes of putting on a travel document like a u.s. passport or u.s. visa or a photograph that a foreign national when they arrive in the u.s. like under the visa program, we take their photograph or read it from a passport and store back. that's what forms the baseline gallery that we match against. new photographs we take of a person, we match up to a u.s. passport, the photos are discarded. they go over to the dhs runs
4:32 pm
with a store under the particles of the system of the data retention which i believe is 75 years. >> a follow-up with that, we are living in an age where everything is being hacked. what type of security measures productions have been in b place regarding the security? >> they are housed within the u.s. government. cbp doesn't necessarily keep her own any of those permanent databases. they are owned by department of state and other participants. we access a lot of the information and use it match against the input information back into them. >> okay. i continue to have, this came
4:33 pm
across my desk, facial recognition technology and racial bias, it's my understanding that it continues to misrepresent and irregularly identify people of color and women. in my hearing from majority of the panel been that that is not the case? it keeps coming to us. there has too be some validity. >> in our testing for the 11 identification algorithms, we do see evidence of demographic effects, differences with regard to race and sex and age in the one to many identification testing we i did with the algorithms we tested, there's a
4:34 pm
small set of high performing algorithms that had undetectable differentials. but the majority of the algorithms still exhibit those characteristics. >> a description of the difference between the two? >> yes. in the case of verification, verifying and identity, a biometrics is matched solely against -- >> is that the one to one? >> yes. the revocation is to try to determine if you say who you say you are. prince matched against a gallery of one. >> was the one to many? >> it's matched the case of cbp's application, one, two perhaps thousands for the airline traveling public or one
4:35 pm
to millions in the case law enforcement to try to identify the suspect. >> your think the percentage of identifications in the 120, you have more incidents of this bias? >> in the algorithms that we tested, that is correct. however, many of the vendors who chose to participate in the one to many testing did not choose to participate in the one to one. those are some of the highest performing. >> thank you, i yield back. >> chair recognizes the gentleman from north carolina. >> thank you. i like to yield to interment of louisiana. >> the effectiveness of the technology that you tested
4:36 pm
regarding, is there potential from fleet assembled a gallery of photographs of children, some of them would be exploited falsifications presented how hot is like technology work with children compared to mistakes and errors in other demographics? being exploited across borders coming into our country and if so, what can we do to protect the privacy of the children's due to the fact that they are minors? >> the applications specifically is something we don't test. but we have tested is the effectiveness of the algorithms in terms of error rates and we doit find for children one to 11
4:37 pm
he just described, there are demographic effects there, there differentials. there rates are higher in the 11 cases with respect to any it's more difficult based on our testing, it appears more difficult. >> but there's no gallery, there's no one to many. >> we have no such gallery. >> perhaps this could be a tool to protect children. >> we can undertake many different kinds of testing to determine the effectiveness. >> thank you. >> thank you. mr. wagner, is it true a biometric entry system uses personal identifiable information current system we have in place? >> yes.
4:38 pm
currently you open your passport booklet and show it to an individual to either check your bags or go through screening, the cbp officer, you are exposing your name, date of birth, all the information on your passport. they could be looking over your shoulders, your disclosing into a person who doesn't actually need to know all that additional information. person standing in front of a camera with no identifiable identifiable information of in your face. that person knows you've been validated by the government record to proceed.o sharing less information in this instance. >> on a scale of one to ten, how would you rate this as continuing toou develop, the highest security possible for
4:39 pm
travelers compared to anything else? >> on top of everything else we are doing, it brings us closer to ten which is where we want to w >> when he talked about what you are involved in, are there any numbers based on people who have caught either involved in human trafficking or other various activity quickly because of the facial recognition? >> on the land border, we cut 247 imposters so far. the judgment documents belonging to someone else. eighteen of those 7% were under the age of 18, considers children. seventy-three of them had u.s. passports or cards, 46 of them are almost 20% had criminal records they were trying to hide. >> you believe these were identified strickler because of the use of facial recognition? was any aspect?
4:40 pm
>> they are good at identifying behaviors in the present when they resent the travel documents. i can be acute that the person is hiding something. the technology on the skills and abilities should bring us toil that security realm. >> are there any policy systems different for citizens versus noncitizens? >> everyone has to establish identity by law. after produce identification. p >> the process scrubbing this after is what? >> we discard after 12 hours, we are looking at shrinking back to last time. we've got to restore everything. >> thank you.
4:41 pm
i find it interesting the more you talk. a very busy airport, one of the busiest in the country, a lot of international tourists come through. i know we talked a lot about the use of facial recognition for security reasons. i'd like to talk about it in artifacts the passengers experience. we want people in las vegas to have a good experience from the time they land until they leave. how do you work to coordinate using this for security and also reducing wait times or serving the passenger as opposed to making it more difficult? >> it absolutely supports those goals as well. it makes much better passenger experience more convenient and consistent.
4:42 pm
going to theuc airport, the numr of stops you have to make to produce a piece of paper or opening a passport again or providing forms of validations to go forward, you can use the facial recognition and camera to have that same process and it's quick enough that you get your picture protection, then you move forward. we are seeing reduced wait times, the airlines as they incorporate it into the boarding process are reducing their boarding times of the aircraft, sometimes as much as 45%. it's a different atmosphere for the traveler because you're not fumbling for documents or forgetting where you put your boarding pass or get stuck in line behind the person whose phone went dead or forgot where they put their passport. it's creating a better atmosphere for the traveler, it's moving the lines quicker because it's creating, you can't leave your face on the plane.
4:43 pm
you can't leave your face in the bathroom, you can't forget that like they do with travel documents. it's making it easier because everybody knows how to take a picture.o people are enjoying this process a lot better for them. we are seeing clients reduced. >> are you working with tsa or law enforcement to make this run smoothly? >> we are working closely with tsa. we front a few pilots went because we built that gallery as a person print out there boarding pass. any place now where they have to show their passport at the airport, departing the u.s., you could take a picture and validate it against the gallery. your outside of the airport where he walked into their air airport, your picture because into that gallery so steps like checking your bags where you show your id to the airline
4:44 pm
person, he have a character that does that. tsa can take a photograph and transmit the gallery because we built it to the biometric exit requirement, we want to make that environment available to all the other places where he would show your passport to do that. yes for tsa, you can take a picture then he go for screening. t go to board the plane, elling takes her picture and comes back to our gallery and we confirm and you board the plane without even trying your passport to the airline or showing your boarding pass to the airline. >> sometimes you find somebody doesn't match. guess goes through and law enforcement is busy or a person responsible for checking out their meds are doing something else, you have a staffing model? it comes through an app and there's no action you are
4:45 pm
supposed to take, sometimes they just ignore it. >> depending on where you are in the airport. generally the airline. they look at the physical passport which is what you are presenting now. if we have doubts about the picture on your passport, which happens, they may call us over, we may ask a person for anotherr form of id or additional questions. we make duke further inspection on them. if you don't look like your passport photo, from a visual review, is the same think i would occur. >> we've had a lot of confusion about going to the real id from regular drivers license. people don't know they have to do that. some states didn't provide the funding to go to real id. is that transition part of your
4:46 pm
system? >> it is separate. >> it's not going to make a difference? >> not really. >> people who use passports instead of real id, that won't matter? >> these are international commerce we are talking about. >> you don't see this moving national as well as international? >> i would refer to tsa on how this might apply to a domestic flight. i think there's some good discussion to have their that of people have passports and you can electronically confirm them should the traveler opt into this, it would be good government to build a system like this if that's what people would want. >> thank you. >> the check recognizes the young lady from illinois. i'm sorry, gentlemen from texas.
4:47 pm
>> thank you.>> this has been an interesting hearing to watch. i want to go over misinformati misinformation, the technology we are discussing here today, it seems to be abnormally controversial. we are not talking about id government surveillance, not like china has. we're not talking about facial recognition or times square downtown. we are talking about facial recognition air, land and sea ports of entry, not just authority but the responsibility to know who comes into our country in fair checking ckidentification. it seems from the answers we've gotten thatit cbp is using the best algorithms with almost no bias whatsoever. locations where facial recognition technology is employed, locations are marked.
4:48 pm
correct? >> yes, it's where you would normally present your passport. >> places of interest required to present -- you already answered that one. entrants are allowed to opt out of facial recognition technology and present photographic identification who were then compared to the physical appearance of the entrance and identification presented. >> correct. >> it stored no more than 12 hours in a private cloud. >> correct. >> data for entrance or not u.s. persons are stored. >> correct. >> given the above, visual recognition technology used requirements present photo id, ability to opt out and secure storage, was the major privacy concerns i might be missing? how can we improve it? >> i think what we've heard from a visit community is people get
4:49 pm
used to the convenience of this technology and it bleeds over into the commercial role in a private sense and faith may be more likely to allow that to happen outside of the government requirements. my discussions, also an expectation for the public that they have us in means in their private life, why should the government interactions be integrated? why should they be antiquated and manual and frustrated? do they expect the inconvenience should apply when traveling internationally? >> one way this could be used in a positive sense, is there a way that this can be integrated with other tools like spotlight, battle sex trafficking.
4:50 pm
>> it helps our core betting processes are biographical based. through tsa, we vet into the background checks, protecting? who purchased a ticket? then you can use a biometric to validate and fed at the right person, you have the assurances that that is the person who's actually traveling is not just their passport is traveling under a different person that's being trafficked. it helps us close the vulnerabilities of imposters are being victimized to be able to do that using imposter documents. >> can this be used to combat these as well? >> we've actually, we track them
4:51 pm
primarily through the biographic information but by implementing this system, we have diametrically confirmed almost 44000 over states. the biometric validation that these peoples overstayed will be later than they were authorized to do so. just about 44000. i healed back. >> check recognizes the gentleman from new york. >> thank you. the nypd has, in the past, used facial recognition to compare photos from crime scenes, some state lawmakers want to take the ability away from the nypd as the agencies do support the agencies using facial recognition in the course of the criminal investigations.
4:52 pm
>> that's not necessarily an issue we looked at, we are primarily looking at the facial recognition technology and in particular, we focus primarily less on the identification piece for you have, you're trying to map a photo to our gallery and we are looking at a narrower verification. again a role there is to make sure we are addressing the concerns regarding temperamental bias whether race, national origin -- >> one thing in this conversation in the way in which civil liberty can be infringed upon the absence of these technologies. speak to this for a bit, false positives, people who are being
4:53 pm
arrested or at least questioned further based off just a verbal description. >> i think it's obviously critically important to blend both the use of technology as well as the end user in this process. i don't think it's a proposition as we've advised the components, we've got a perspective, that's really where you see the greatest benefit, it's that technology there. for example, if there was a false negative, and you would have the officer looking at their actual passport or travel documentation. benefit matters, they go along and aboard the flight. >> is important to note that the
4:54 pm
use of technology has consistently been implemented to preserve our public safety but also to further protect civil liberties. this is being lost in this conversation and yet again the same people are unnecessarily politicizing to keep us safe. you all have some work to do to make it even better. you will have to here from another. >> i am not a supporter of this new york legislation passed, i think it is unfortunate andd wrong you were not notified. but two wrongs do not make a right. i'm asking simple questions. if you were setting up to be a professional force that you are and do this professionally, if you think in advance of announcing this, you should have told congress what was wrong and
4:55 pm
what would happen if it wasn't fixed or addressed? >> i would have to refer to dhs on that. >> this is ridiculous. it's a simple question.s we heard about this from fox news. politics at its worst. acting like professionals right now.w. if there is a problem and needs to be addressed and you are all doing this, you think it was appropriate that we were not told well in advance so we could check provide a solution? do you think that's okay? is that the way you would want to do that? >> are not going to comment on that. >> the fact that you have given very clear answers previously, i think we can all assume what you are thinking and unwilling to say right now. let's commit to trying to solve
4:56 pm
problems here. as of congress that won't be able to renew something, the cae are helping the balance. colleagues, all types of people. it's just politics. if you are really making an effort to address a problem, there would have been a system, a proposal, a negotiation, a conversation, letters written, there's no way this is conductive. let's put that aside. what you now commit, now that we've all engaged in ourd politics to having sensible meanings and conversations about a way forward to solve this issue? you would commit to that? okay, thank you.
4:57 pm
>> your time has expired. check recognizes the young lady from illinois. >> many of my constituents have ri drive over an hour to get to the airport. we would like to learn more about technology that can take down the time. it's important to ensure dhs uses facial recognition and other technology in fair and reliable and effective way. mr. wagoner, although children under the age of 14 are not required to be screened, if any do go through screening that collects biometric information. hardly store and secure the information? i'm talking about under 14. >> if you're outside the scope of the biometric tracking requirements, which is 1479, i believe we discard all of the information. >> would you be willing to provide that information about the procedure and policy?
4:58 pm
>> yes. >> are there any differences in how they collect, use or secure children's biometric information in comparison to adults? so if a child presents, do they take it and immediately delete or will they go through a filter they are in? december 2019 from this report found children are more likely to be misidentified during screening. other groups, people of color and misidentified them. they correct their identified report. >> again, we are using high-performing algorithm that we are not in the demographic based. if they do not match the gallery or to the document they are presenting, we'll examine the document physically. and really look at the picture and if we have confidence edits
4:59 pm
the present, we can do it through questioning or verification, we do it through inspection of the person, sometimes it's just looking at the passport and sink go ahead. it all depends on how discrepant you look from your travel document photograph. >> unaware about how to opt out of the screening month they expand the screening program and tend to reevaluate the best way ndof communicating the information. >> right now, we've got signage at the airports, a lot of people don't read signs of the airport. we got announcements that the airlines try to make before boarding. there's always competing announcements going on to understand what's being said. could we print things on the boarding pass? can we give notification when people get their ticket or when they are getting fact checking
5:00 pm
information on their electronic messages? we are looking at additional ways to do that. we started taking out privacy advertisements in providing people requirements of other options. >> it is my interest in making sure every passenger understands this is happening and that they have a choice to opt out. ...
5:01 pm
my question is capturing and reporting by port of entry so we want to know are we seeing more is certain places along the border or in certain areas? >> we announcing false patterns -- we are not seeing false positives with this technology. >> or mistaken identity? >> more likely you don't match against anything so we get no informationg in return. >> okay. can you elaborate on algorithm developers can improve accuracy? >> report assessing that we do does not result in the recommendation specifically to the vendors other than to take the data that we provide, the evaluation results and strive to use those resultsd to improve i.
5:02 pm
>> you're saying you don't have a lot of interaction with developers. >> we have informal interaction with them in the sense that the scientist to do this by metric r biometrics community. we see the vendor representatives and scientists have meetings and so on but with regard to that itself testing tg the feedback that we provide to is the test rule. >> so you all are doing leading with industry and helping them improve their products? >> we do host events but more as a convener to get the community to gather to discuss different techniques. we don't provide other than in the general scientific community since we don't provide specific recommendations for their improvementse. stick i recognize my time has expired. we would like to get more
5:03 pm
information. thank you sir. >> the chair recognizes the gentlelady from new jersey. >> thank you mr. chairman and thank you for your testimony. a couple of questions. i want to talk more about the role of the t.r. cl and it seems to me there has not been much coordination across the dhs and directions from dhs to each component regarding deployment of biometric technology and correct me if i'm wrong. is there any sort of departmentwide strategy in place for the use of biometric technology or our components like yours given wide latitude to stand up to biometric programs as you see them? >> i'm sorry?
5:04 pm
>> are you a lone ranger? >> are we what? i'm sorry i didn't hear that. >> okay. did this court nation doesn't seem like it's a departmentwide oversight? i want to know whether or not you're getting directions from others. what is your role and to what degree have you been involved in the oversight and in signing off on how these are being done in? >> as we build out new programs we are balanced by certain statutes that require us to publish your systems of records notice your privacy impact assessment where things are reviewed by our internal counsel or our privacy office and to make sure we make and meet all the requirements of the statute. you have the authority to collect what you are doing? is your timeline for storing and sharing is that all permissible
5:05 pm
in the law and is it consistent with your mission? are you authorized to do those things. >> are you offering your silo, this is what the law is with regard regard to what you can do. is this how you execute based upon what your interpretation is of that or is there a dhs component that plays into this as well and says okay but this is how we want to see this? >> depending on acquisition process. there's a multitude of people at dhs that look at the acquisition , the resources spent. there's a whole process to go through that for approval before various boards that authorized expenditures and the investment and that. there's a dhs privacy officer and there's a lot of oversight by dhs already in this process. certainly the rulemaking's would go through dhs counsel, the dhs
5:06 pm
policy so there is a lot of oversight in court nation. >> it's my understanding though that there is no body within the department that gives the program a stamp of approval and certify that o they are ready fr primetime? is that correct? have they approved your program? do you know? >> they would necessarily go to them for approval. >> approval in maintaining and protecting privacy rights. >> things are reviewed by them. >> what authority do you have sir? >> why don't i enter that in a couple of different ways congresswoman. let me step back and talkif abot the first part of your question regarding the enterprise level review. one of the ways in which that dialogue is by serving on enterprise level wide working groups across the department that include representatives of cbp and the office of biometrics identity management where we are
5:07 pm
talking about a lot of these issues. we don't have a privacy impact model however we do work closely with the privacy office regarding not facial recognition technology but certainly other forms of biometrics identification that the department uses. with regard to relationship with cbp we work with them in a couple of ways. first is offering them b advice and also we work with cbp and privacy civil liberties oversight as well. >> is your role in the thing more than an observation and advice clicks you have no authority to say no that's not working and that's a violation, that's it, right? >> that's not entirely accurate. what i would say is yes we do have advisories. we offer recommendations. >> and if they don't follow
5:08 pm
them? family can elevated if necessary. >> i have one last question and the question has to do with just the whole system that is used when we are taking pictures. who is in charge of determining whether or w not the lighting is good the background iss adequate the cameras are good they are placed right so that we can get the best that we need to get? is there anyone in charge of that? >> cbp would be and that's going to beou based on our results of the matrix. you can have an airport with a bank of boots in the windows are such that the sunlight comes in and affects the boots during the morning and these boots in the afternoon and these are things we have to look at as we deploy this. one of the environmental factors that will influence all the different locations that we are going to do this and then we try to adjust and we might have the airport had the windows.
5:09 pm
>> how do you do that? >> we do that internally by reviewing the data and the results. >> daily, monthly, whatever? do you know whatt time the sun comes in that window and the sun comes in that window. >> we do continuously to get the best protection we can. >> thank you very. much. >> thank you very much. the chair recognizes the ranking member. >> that the consent request to enter into the record. >> without objection so ordered. the chair recognizes the gentlelady from texas. >> i thank the chairman very much and let me acknowledge all of the witnesses and thank them for their service. let me renew the inquiry that will be pursued by chairwoman rice but i will add to it and that is mr. wagner a better understanding of providing information to the committee on
5:10 pm
the denial of clear traveler as it relates to states that may not have the laws that you think are appropriate or an instance of new york closing out access to the issue driver's license and i raise the question because we should look at the federal government as what other identification options might eval it and i know we have known each other for a long time but i would think you'd be willing to look at that so that we can find common ground. let me pursue this line of reasoninge and please witnesses understand that i'm not saying this is what you are doing. any punter stand your thinking. to the deputy assistant commissioner wagner would you accept the fact that bias could be introduced by technology and the application developer of the program into how an application
5:11 pm
reacts to different types of people because its technology. i would also make the point it's a little bit humorous i'm sure the people in iowa thought they had something going on there and we all could see where we are at this point. would youse accept that mr. min? and you would accept that mr. romine? he yes? algorithms and again this is not pointing to what you have written all-black males wearing dreadlocks. a mr. wagner? this is in terms of how technology could read. >> i guess he could. >> you can see in the record to your knowledge you are not using that. that would be very good. i'm sure the judge would be glad of that. mr. mina. >> that is possible in a review
5:12 pm
as well. >> all right all right mr. romine. you could certainly possible. >> uriarte and my colleagues over here that are members of the judiciary committee we are in a -- i wrote the violence against women act and put in dna enhancements that we understand the new add it toality but as the department of homeland security we made a commitment post-9/11 with george bush going to the trade and saying he heard the firefighters but at the same time he also heard muslims who were indicating it's not the blanket world for people who happen to be muslim. in particular mr. mina i want to try to find out what aggressive role do you play in helping to not have platitudes and i'm not suggesting you do but to aggressively assure that the biases against black women with
5:13 pm
dreadlocks and men with dreadlocks, muslims and sheiks wearing a -- and i've been on this committee is not technology but technology is not undermining the civil liberties and civil rights of this nation and those coming in innocently to the country. can use these new technologies while and to mr. romine let me find out how are you continuing to do your assessment of these algorithms to ensure that it looks like you are not able to get the exact one that mr. wagner's team is using. that concerns me. i need you to get every bit of information in that light you to say mr. mina what aggressive is are you doing to protect the travelers, the american people? >> i thank youou for the questi. we are doing a lot of cross examination of this program and policy. again i want to focus our
5:14 pm
attention is really on the application not so much the algorithm itself but applied by the dhs program. we do that on a policymaking site working directly with the component, advising on proposed regulation and implementing policies as well as offering suggestions as it relates to applications for example folks that have headwear or objective topography or people who are disabled or injured that aren't able to take pictures. we also do it for community and members of the community across the country. i have the information from me regarding certain areas. the issue's been recent portland atlanta a chicago in 20 west thd spent in california primarily l.a. and orange county in orange city area. we have heard concerns regarding facial recognition technology in her primary role is to be the
5:15 pm
eyes and ears of the department of former colleagues at cbp. here the concerns we are seeing and how do we work together to try to address some of these problems or potential problems that will have a greater effect and also on the back and we have a robust compliance process. while they don't have an active investigation right now it's something we are looking at. we most certainly will open an investigation and at lies in that way as well. >> i just want to say this for the record that we can get answers in terms of retaining information. >> you can answer the question and submitted in writing. >> thank you. mr. romine my question was --
5:16 pm
>> i beg your pardon maam? >> my question was what you doing to be accurate in your te? he said he didn't know whether you had the accurate apps that they were using. what are you doing to make sure we don't have the bias in the algorithm? >> the tests we undertake are intended to determine whether they are democratic differences. some call it bias. the fact that i know there is strong interest in testing testh data that is more representative and we signed a recent mou with this cbp to undertake continuede doing the very best that we can to provide the information that they need to make sound decisions. senate thank you very much. i yield back. >> i recognize the gentleman from texas mr. green.
5:17 pm
>> thank you mr. chairman. i thank the witnesses for appearing as well and i would like to address some intelligence that has been afforded me. the indication is that nist found african-american faces were 10 times more like lee, 10 to 100 times more likely to be misidentified than white faces and i'mha curious as to whetherr not there is something inherent in the technology that creates an inverse relationship with reference to the identification of whites juxtaposed to african-americans? is there something inherent in the technology meaning if you want to absolutely identified whites will there be something that you cannot adjust so you will get the same absolute
5:18 pm
identification with minorities, african-americans or if you want to absolutely identify african-americans will u.s. a result of technology not able to properly identify white's? >> it's a very interestingou question and let me clarify first that those differentials that we have observed were not in the case of misidentification but their occasion -- verification of one-to-one rather than the one to many tes. we saw those demographics for and pacificican islanders as well. but in the case that you are talking about our work has not to date focused on cause and effect. what is it that causes theno algorithms to exhibit certain types of behavior? we are really just testing the performance.
5:19 pm
>> my question was interesting as you put it. your answer is intriguing because this is not the first opportunity for the word to be heard that we have these difficulties and at some point it would seem that they would move from testing technology as it is to understanding why technology performs the way it does. help me too understand why we haven't made that move? >> the question you asked is a very challenging open research do have ant we issue. there are algorithms that have been submitted to our testing tg from asian countries that do not exhibit the demographic differentials on asian faces so we can't guarantee it that we think that's an indication that
5:20 pm
the training data that are being used for the algorithm development may have a significant impact on your ability to discern our exhibit demographic basis for different populations. >> do you believe it's important for us to move expeditiously to answer this question, to resolve this issue especially if we don't find ourselves having deployed something en masse that we know to be defective or have some degree of inefficiencies associated with it. of this isy important. >> yes sir. those are two different things to think about. performance testing that we currently execute has helped have helped operational agencies ensure that they are not applying different demographic
5:21 pm
differentials. the research question that you tee that up which is interesting is a much deeper question much more difficult i think. >> is it fair to say that the country, and i have about 45 seconds left but a country that employs technology that has indicated to you they are having fewer challenges is it fair to say that technology also captures white men sufficiently? in the testing we did before the specific one that i'm referring to the high-performing algorithms from asian countries that don't exhibit demographic of differences, there is no difference in the performance were no discernible difference on caucasian faces and asian faces under certain asian
5:22 pm
developed algorithms and the onlytr speculation is that maybe the training data that are used. >> thank you mr. chairman to baio back. >> thank you very much for the chair recognizes the gentleman from rhode island for five minutes. >> thank you mr. chairman and thank the witnesses for your testimony here today and thank you for your dedication to this very important issue. i believe that technology is important solutions to some of our p most active issues and challenges including an ever-growing number of international travelers. what i wanted to ask mr. wagner or mr. mina we know in technological solutions such as facial recognition software the algorithms are only as good as the data that informed them so i want to know how this cbp adjusted for the data that it
5:23 pm
uses to train facial recognition software and what are you doing to ensure software continues to be updated and algorithms are incorporated? >> we work closely with the nac and we work closely with them to incorporate there up dates and their latest and greatest products into how we are using them. as we review the data we look to make those operational adjustments which do impact metrics and again that's going to be the quality of the photograph the quality of the camera, the human factors or the size of the gallery is important. this gallery is up to 12 million people. on the margins of the capabilities of these algorithms we are doing these in a couple thousand and interesting correlations or how much better
5:24 pm
improved is your impact on potential demographic vices on this much smaller gallery space. what we are getting at earlier one of the variables that we can riser lower to help address some of what the error rates are showing us. >> to that point how this cbp you incorporate feedback from officers about errors that facial recognition software has made in the field? when the officer is looking and interacting the software doesn't get a credulous that feedback comes back to the system it doesn't learn. >> in the look at system logs themselves but we talked the officer they provide feedback and we are also on sight to witness and observe and discuss with those officers as we deployed these. >> understand the trusted traveler program shares information with othertr countrs
5:25 pm
and how this cbp share biometric information with other countries and what steps do you take to ensure those countries use the data responsibly? is that accurate and my understanding is our regarding that data? >> i'm trying to think. i'm not aware of how we would share or if we are sharing the trusted traveler program. we don't share -- we might share person status in the program but we are not actually sharing fingerprints. >> what they asked that for the record and i would ask that you give back to me. what types of information do we share under the trusted traveler programr and because i think that's important for us to know and if we do share whatever information we share i want to know what steps we take to ensure the countries use that
5:26 pm
data responsibly. i know this question has been touched on earlier so i'm going to ask it in a different way. prior to her hearing on this topic last july we were notified on the network of the cbp subcontractors. someone claiming to be a foreign agent send tens of thousands of photos of drivers is faces and license plates along the port of entry along the southern border. how does cbp ensure that the personal and collective facial recognition by the government directory or vendors or the -- are being protected from inadvertent or unauthorized access and also what can you tell the committee are the root causes of the may 2019 breach
5:27 pm
being addressed is to reduce the likelihood of another breach? we have assigned sets of business requirements with which they commit to not storing not sharing not saving any of the photographs that are transmitted to us and purging them from their system. one of the other conditions as they have to be available for cbp to audit their cameras in their technology to ensure that they are following those rules. we are about to commence an audit of one of the airlines in the next couple of months. but to make sure that that's not happening. >> i would ask that the director get back to me in writing isn't as possible on the trusted traveler program and what information you share with others. thank you mr. chairman. >> thank you.
5:28 pm
the gentlelady from texas wants to ask a question. >> mr. chairman thank you so very much. first of all i ask unanimous consent to place in the record not to the witnesses that the headline reads amazon facial recognition is confusing 20 congressmen with no criminal -- july 26, 20 tina mr. wagner are you using amazon technology? >> we are not using their matching algorithm. >> thank you mr. chairman my question as you gave congressman langevin a detailed response so let me try to change it around -- do you have a team that is directly responsible not just for the implementation but for the internal analysis of the utilization of the technology that you are using so it's on-site so you are able to get
5:29 pm
firsthand knowledge of the violations or abuses by way of thee technology. is that information coming back to your office and when i say that, your sector? part of our office has been working engine -- conjunction with their field location. >> give a team that's responding to that? >> we have reviews of the compliance of the officers using the technology. >> mr. chairman i know we have a lot of work and maybe that can be a classified reefing. i'd like to know how that is done and how that is kept and how long they keep the data on mr. jones or mr. amman or various persons. >> we will work through it. >> the data is stored in compliance with the system of
5:30 pm
record notices of where that data is stored. the photograph of a u.s. citizen that we take is only held for 12 hours and then purge. picture a foreign national is sent to the department database words stored for 75 years. the record of order crossing the biographical information is then stored another suit systems attributable to those. >> let me insert in the record a letter from the electronic privacy information center and the press release from the u.s. travel association. mr. wagner if you are aware of any notification requirements that a


info Stream Only

Uploaded by TV Archive on