Government agencies are tapping a facial recognition company to prove you’re you – here’s why that raises concerns about privacy, accuracy and fairness

 

 Beginning this summer, you might need to upload a selfie and a photo ID to a private company, ID.me, if you want to file your taxes online.

Oscar Wong/Moment via Getty Images

James Hendler, Rensselaer Polytechnic Institute

The U.S. Internal Revenue Service is planning to require citizens to create accounts with a private facial recognition company in order to file taxes online. The IRS is joining a growing number of federal and state agencies that have contracted with ID.me to authenticate the identities of people accessing services.

The IRS’s move is aimed at cutting down on identity theft, a crime that affects millions of Americans. The IRS, in particular, has reported a number of tax filings from people claiming to be others, and fraud in many of the programs that were administered as part of the American Relief Plan has been a major concern to the government.

The IRS decision has prompted a backlash, in part over concerns about requiring citizens to use facial recognition technology and in part over difficulties some people have had in using the system, particularly with some state agencies that provide unemployment benefits. The reaction has prompted the IRS to revisit its decision.

a webpage with the IRS logo in the top left corner and buttons for creating or logging into an account

 

 

 

Here’s what greets you when you click the link to sign into your IRS account. If current plans remain in place, the blue button will go away in the summer of 2022.
Screenshot, IRS sign-in webpage

As a computer science researcher and the chair of the Global Technology Policy Council of the Association for Computing Machinery, I have been involved in exploring some of the issues with government use of facial recognition technology, both its use and its potential flaws. There have been a great number of concerns raised over the general use of this technology in policing and other government functions, often focused on whether the accuracy of these algorithms can have discriminatory affects. In the case of ID.me, there are other issues involved as well.

ID dot who?

ID.me is a private company that formed as TroopSwap, a site that offered retail discounts to members of the armed forces. As part of that effort, the company created an ID service so that military staff who qualified for discounts at various companies could prove they were, indeed, service members. In 2013, the company renamed itself ID.me and started to market its ID service more broadly. The U.S. Department of Veterans Affairs began using the technology in 2016, the company’s first government use.

To use ID.me, a user loads a mobile phone app and takes a selfie – a photo of their own face. ID.me then compares that image to various IDs that it obtains either through open records or through information that applicants provide through the app. If it finds a match, it creates an account and uses image recognition for ID. If it cannot perform a match, users can contact a “trusted referee” and have a video call to fix the problem.

A number of companies and states have been using ID.me for several years. News reports have documented problems people have had with ID.me failing to authenticate them, and with the company’s customer support in resolving those problems. Also, the system’s technology requirements could widen the digital divide, making it harder for many of the people who need government services the most to access them.

But much of the concern about the IRS and other federal agencies using ID.me revolves around its use of facial recognition technology and collection of biometric data.

Accuracy and bias

To start with, there are a number of general concerns about the accuracy of facial recognition technologies and whether there are discriminatory biases in their accuracy. These have led the Association for Computing Machinery, among other organizations, to call for a moratorium on government use of facial recognition technology.

A study of commercial and academic facial recognition algorithms by the National Institute of Standards and Technology found that U.S. facial-matching algorithms generally have higher false positive rates for Asian and Black faces than for white faces, although recent results have improved. ID.me claims that there is no racial bias in its face-matching verification process.

There are many other conditions that can also cause inaccuracy – physical changes caused by illness or an accident, hair loss due to chemotherapy, color change due to aging, gender conversions and others. How any company, including ID.me, handles such situations is unclear, and this is one issue that has raised concerns. Imagine having a disfiguring accident and not being able to log into your medical insurance company’s website because of damage to your face.

 

 

 

Facial recognition technology is spreading fast. Is the technology – and society – ready?

Data privacy

There are other issues that go beyond the question of just how well the algorithm works. As part of its process, ID.me collects a very large amount of personal information. It has a very long and difficult-to-read privacy policy, but essentially while ID.me doesn’t share most of the personal information, it does share various information about internet use and website visits with other partners. The nature of these exchanges is not immediately apparent.

So one question that arises is what level of information the company shares with the government, and whether the information can be used in tracking U.S. citizens between regulated boundaries that apply to government agencies. Privacy advocates on both the left and right have long opposed any form of a mandatory uniform government identification card. Does handing off the identification to a private company allow the government to essentially achieve this through subterfuge? It’s not difficult to imagine that some states – and maybe eventually the federal government – could insist on an identification from ID.me or one of its competitors to access government services, get medical coverage and even to vote.

As Joy Buolamwini, an MIT AI researcher and founder of the Algorithmic Justice League, argued, beyond accuracy and bias issues is the question of the right not to use biometric technology. “Government pressure on citizens to share their biometric data with the government affects all of us — no matter your race, gender, or political affiliations,” she wrote.

Too many unknowns for comfort

Another issue is who audits ID.me for the security of its applications? While no one is accusing ID.me of bad practices, security researchers are worried about how the company may protect the incredible level of personal information it will end up with. Imagine a security breach that released the IRS information for millions of taxpayers. In the fast-changing world of cybersecurity, with threats ranging from individual hacking to international criminal activities, experts would like assurance that a company provided with so much personal information is using state-of-the-art security and keeping it up to date.

[Over 140,000 readers rely on The Conversation’s newsletters to understand the world. Sign up today.]

Much of the questioning of the IRS decision comes because these are early days for government use of private companies to provide biometric security, and some of the details are still not fully explained. Even if you grant that the IRS use of the technology is appropriately limited, this is potentially the start of what could quickly snowball to many government agencies using commercial facial recognition companies to get around regulations that were put in place specifically to rein in government powers.

The U.S. stands at the edge of a slippery slope, and while that doesn’t mean facial recognition technology shouldn’t be used at all, I believe it does mean that the government should put a lot more care and due diligence into exploring the terrain ahead before taking those critical first steps.The Conversation

James Hendler, Professor of Computer, Web and Cognitive Sciences, Rensselaer Polytechnic Institute

This article is republished from The Conversation under a Creative Commons license. Read the original article.

James Hendler, Professor of Computer, Web and Cognitive Sciences, Rensselaer Polytechnic Institute
This article is republished from The Conversation under a Creative Commons license. 

WSTNet Lab Profile: RPI

RPI (the Renssellaer Polytechnic Institute) is based in Troy, New York state comprising 30 research centres and over 750 PhD students. These research teams are engaged in projects worth over $100 million.   

RPI’s Center for Computational Innovations (CCI) is home to one of the most powerful (eight petaflop) supercomputers on the November 2019 Top 500 ranking of supercomputers (named AiMOS – Artificial Intelligence Multiprocessing Optimized System – in honor of Rensselaer co-founder Amos Eaton).

RPI is making AiMOS available (in partnership with IBM, academic institutions, and national labs) as well as access to the expertise of world-class faculty in data, artificial intelligence, networking, therapeutic interventions, materials, public health, and other areas necessary to understand and address the threat of COVID-19.

RPI hosts the Tetherless World Constellation (TWC) which is an active WSTNet laboratory associated with the Web Science community.

 

Deborah McGuinness is a professor in the computer science and cognitive science departments and the director of the Web Science Research Center at Rensselaer. She is a leading expert on knowledge representation and reasoning languages, ontology creation and evolution environments, and provenance.  

 

She is a long-time friend and supporter of Web Science and best known for her research on the Semantic Web and in bridging artificial intelligence and eScience. An extension of the World Wide Web, the Semantic Web allows computers and other electronics and robotics to communicate and interact without requiring human intervention. The Semantic Web uses information encoded in Web ontology languages to allow computers to “talk” to and understand one another. She is a professor and Lab director at the Tetherless World Constellation 

McGuinness’ work on ontology languages and semantic environments opens the Semantic Web to a broader user base and enables semantic applications to proliferate. McGuinness is one of the founders of an emerging area of semantic eScience—introducing encoded meaning or semantics to virtual science environments. Within this intersection of artificial intelligence and eScience, McGuinness is engaged in using semantic technologies in a range of health and environmental applications.

She has published more than 200 papers on semantic eScience, data science, knowledge-based systems, ontology environments, configuration, search technology, and intelligent applications, and holds five patents. She recently won the Robert Engelmore Memorial Association for the Advancement of Artificial Intelligence Award for leadership in Semantic Web research and in bridging artificial intelligence and eScience, as well as significant contributions to deployed artificial intelligence applications.

McGuinness earned a bachelor’s degree in computer science and in mathematics from Duke University, a master’s degree in computer science and electrical engineering from the University of California at Berkeley, and a doctoral degree in knowledge representation from Rutgers University.

Click here to see TWC’s WSTnet Page
Click here to visit the RPI website

 

Real-time Twitter Visualisations for the US 2016 Presidential Elections

Twitter Visualisation at RPI For the 2016 US Presidential election, researchers at the University of Southampton with support from the EPSRC funded project SOCIAM,  built a real-time data visualization that combined traditional polling data with social media posts. The application was built and designed for the Rensselaer Polytechnic Institute EMPAC Campfire, a novel multi-user, collaborative, immersive, computing interface that consist of a desk height panoramic screen and floor projection that users gather around and look into. The application is also a part of the Web Macroscope (a visualization platform developed at the University of Southampton) and uses data from the Southampton Web Observatory.

Data collection for the polling data was taking from the Huffington Post Pollster API, which collects all the popular polls and their results. The social media data was collected on Twitter, using both their Streaming and Search API. The Streaming API was used to create a stream of data that included 1% of all tweets that had any of the popular and official hashtags and words used by each campaign to show support for their candidate. This hashtag list included tags like ‘TeamTrump’, ‘maga’, ‘TeamTrump’, and ’draintheswamp’ in support for Donald Trump, and ‘LoveTrumpsHate’, ‘ImWithHer’, ‘StrongerTogether’, and ‘WhyIWantHillary’ in support for Hillary Clinton. Any tweets that mixed hashtags and words from both candidates were removed as this was normally done in a way to not show support for a candidate, but to react to supporters on the other side.Campfire visualisation of US election Twitter activity
Results from the visualizations showed different levels of support on Twitter for each candidates over time. In the days leading to the election on November 8th, tweets in support for Trump were 1.5 times greater than those in support for Clinton. Interestingly, on the day of the election, this ratio switched and levelled off. Around the 2pm EST on November 8th, tweets in support for Clinton were almost equal to the number of tweets supporting Trump. Later in the night of the election, the ratio of support changed again, with tweets in support of Trump being 1.14 times larger than those in support for Clinton.
Another interesting result from the data, was the how many tweets that had geographic information tagged to them were overwhelmingly in support for Clinton throughout the day leading and on the election. Most tweets streamed through the visualization had no GPS lat/long data embedded in them (these tweets often come from mobile phones using the Twitter App, with the optional GSP location data option enabled). As a whole, these geographic tweets are a small minority of the data collected from the Twitter Stream (about 1%). Interestingly, these geographic tweets supported Clinton 15 times more than Trump. Why this is the case is hard to say. It looks like Clinton supporters use mobile apps with location data more than Trump supporters.
Two other studies – one from researchers at USC, and another from Oxford University, the University of Washington and Corvinus University of Budapest,both showed that AI controlled bots were spreading pro-Trump content in overwhelming numbers. This created the illusion of more support for Trump on Twitter than make naturally been. Our results of geotagged tweets in support for Clinton, despite overall support from Trump on Twitter might be due to this issue of bots.
Authored by Dominic DiFranzo, 18 November 2016.

ACM Web Science 2017 at Rensselaer Polytechnic Institute, Troy NY

WebSci17 is taking place at Rensselaer Polytechnic Institute (RPI) in Troy, New York, co-chaired by Professor Deborah L McGuinness (Tetherless World Senior Constellation Chair and Professor of Computer and Cognitive Science at RPI) and Professor Peter Fox (Tetherless World Constellation Chair and Professor of Earth and Environmental Science, Computer Science and Cognitive Science at RPI). Program Chairs are Dr Katharina Kinder-Kurlanda (GESIS) and Professor Paolo Boldi (Univ Milano).

Save the dates!

  • Notify intention to submit: 1 March 2017
  • Submit papers: 8 March 2017
  • Submit extended abstracts: 1 May 1 2017
  • Conference: 26 – 28 June 2017
  • Workshops: 25 June 2017

WSTNet Lab Directors Meet at WebSci16

WSTNet Lab Directors Meeting, Hannover, 22 May 2016.

WSTNet Lab Directors Meeting, Hannover, 22 May 2016

WSTNet Lab Directors got together at the start of the Web Science Conference this week in Hannover, Germany. Highlights of the meeting include the election of Steffen Staab as Chair and Pete Burnap as Vice-Chair, planning for this years’ Web Science Summer School at University of Koblenz (30 June to 6 July – ), and firming up of arrangements for World Wide Web Week – a global event celebrating 10 years of Web Science to be held later this year.

Who’s who in the photo (from left to right): Thanassis Tiropanis (WSI), Manfred Hauswirth (FOKUS), Steffan Staab (Institute WeST), Noshir Contractor (SONIC), Sung-Hyon Myaeng (KAIST), Les Carr (WSI), John Erickson (RPI), Susan Davies (WST), Hans Akkermans (VU Amsterdam), Dave De Roure (Oxford e-Research), Anni Rowland-Campbell (Intersticia), Pete Burnap (Cardiff University), and Wolfgang Nejdl, (L3S).

The Science of Magic

Troy, N.Y. – An interdisciplinary team of researchers at Rensselaer Polytechnic Institute is collaborating with Walt Disney Imagineering Research & Development, Inc., part of the theme park design and development arm of The Walt Disney Company. Together, they are exploring how the cognitive computing technology being developed at Rensselaer can help enhance the experience of visitors to Disney theme parks, cruise ships and other venues.

Walt Disney Imagineering Research & Development, Inc. and Rensselaer researchers are exploring a range of cognitive computing technologies. These include information extraction techniques to help computers better understand words written or spoken by a human, as well as agent-based techniques for investigating how computers and humans can engage in more natural conversations.

“Walt Disney Imagineering Research & Development, Inc. is part of the creative force behind the iconic Disney attractions and experiences and is on the forefront of natural interactive character-based experience technologies. Walt Disney Imagineering Research & Development, Inc. has a rich history of creating, developing and bringing to life ground breaking technologies in the field of Audio-Animatronics® Figures. Rensselaer is a world-class research university and a leading force in computational science and engineering, including in the emerging field of cognitive computing. The possibilities of what we can accomplish together are endless,” said Jonathan Dordick, vice president for research at Rensselaer.

“Walt Disney Imagineering Research & Development, Inc. is excited to partner with Rensselaer, a recognized leader in knowledge extraction and natural language understanding. We believe Rensselaer’s world class text and language processing tools, in conjunction with Walt Disney Imagineering Research & Development, Inc.’s cutting edge autonomous character platforms, will enable a new class of Guest/character experiences,” said Jonathan Snoddy, R&D Studio Executive at Walt Disney Imagineering Research & Development, Inc.

Leading the project for Rensselaer is James Hendler, Tetherless World Senior Constellation Professor and director of The Rensselaer Institute for Data Exploration and Applications (IDEA). An expert in web science, Big Data, and artificial intelligence, Hendler said the collaboration with Walt Disney Imagineering Research & Development, Inc. is an important step forward for all of the data-related research taking place as part of The Rensselaer IDEA. Rensselaer faculty members Mei Si, assistant professor in the Department of Cognitive Science, and Heng Ji, the Edward P. Hamilton Development Chair and associate professor in the Department of Computer Science, will collaborate with Hendler on the project.

“Unstructured data, that is the information inherent in written texts and spoken dialog, is an increasingly important part of the Big Data landscape,” Hendler said. “Our goal in this project is to work with Walt Disney Imagineering Research & Development, Inc. to transform the leading-edge tools and techniques into fully developed applications that will help make the Disney experience even more enjoyable for people and families around the world. We look forward to an incredible collaboration with Walt Disney Imagineering Research & Development, Inc.”

Contact: David Brond, Rensselaer Polytechnic Institute
Office: (518) 276-2800
Email: brondd@rpi.edu

– See more at: http://news.rpi.edu/content/2015/03/19/science-magic-rensselaer-and-walt-disney-collaborate#sthash.IK76gZ0d.dpuf