(New York) – Human Rights Watch has joined a coalition of over 50 human rights and other civil society organizations in opposing Immigration & Customs Enforcement’s (ICE) proposed new Extreme Vetting Initiative. The Extreme Vetting Initiative is ICE’s proposal to monitor much of the internet, including social media, to automatically flag people for deportation or visa denial based on broad, ambiguous criteria. The program will not only be ineffective due to technical limitations, but it risks hiding politicized, discriminatory decisions behind a veneer of objectivity. Additionally, it will potentially chill free expression as visa holders and applicants, and those they communicate with, may censor themselves online due to the surveillance.
* * *
The Honorable Elaine C. Duke
Acting Secretary of Homeland Security
Department of Homeland Security
Washington, DC 20528
November 16, 2017
Dear Secretary Duke:
We are a coalition of 55 civil rights, civil liberties, government accountability, human rights, immigrant rights, and privacy organizations. We write to express our opposition to Immigration & Customs Enforcement’s proposed new Extreme Vetting Initiative, which aims to use automated decision-making, machine learning, and social media monitoring to assist in vetting of visa applicants and to generate leads for deportation. As it is described in ICE documents,1 this program would be ineffective and discriminatory. It would also pose a signal threat to freedom of speech and assembly, civil liberties, and civil and human rights. We urge the Department of Homeland Security to abandon this effort.
In July 2017, the office of Immigration and Customs Enforcement (ICE) held an Industry Day to seek input from the private sector about an “overarching vetting contract that automates, centralizes and streamlines the current manual vetting process effort.” The goal of the Extreme Vetting Initiative is to “develop processes that determine and evaluate an applicant’s probability of becoming a positively contributing member of society as well as their ability to contribute to national interests,”2 using analytic capabilities including machine learning.3 ICE also seeks to “develop a mechanism/methodology that allows [the agency] to assess whether an applicant intends to commit criminal or terrorist acts after entering the United States.”4 ICE erroneously believes that because predictive analytics have, at times, been shown to be of value in the private sector, they will offer similar value for national security.5
In reality, as a group of prominent technologists advised in a recent letter, “no computational methods can provide reliable or objective assessments of the traits that ICE seeks to measure.”6 There is no definition anywhere in American law of what it means to be a “positively contributing member of society” or to “contribute to national interests,” posing a risk that ICE will exercise maximal latitude to discriminate beneath the cover of an unproven algorithm. Experts have also concluded that national security threats – in particular, acts of terrorism – are so rare that they are extraordinarily difficult, if not impossible, to predict, because the data are so scant that they do not provide a reliable basis for prediction.7
Instead, it is likely that any developer tasked with building this system will rely on proxies – for example, using an individual’s income to measure his or her contribution to society, or analyzing social media posts for “tone” considered threatening to the United States.8 Indeed, ICE documents place a heavy emphasis on social media monitoring,9 ignoring the fact – recently verified by a DHS Inspector General report – that DHS has failed to put into place adequate mechanisms to measure the value of its monitoring programs.10
What’s more, the criteria that ICE intends to use social media analysis to predict are outside the realm of existing technology. The meaning of content posted on social media is highly context-dependent. Errors in human judgment about the real meaning of social media posts are common.11 Algorithms designed to judge the meaning of text struggle to make even simple determinations, such as whether a social media post is positive, negative, or neutral.12 Moreover, most tools cannot reliably analyze text in languages other than English.13 Confirming that ICE’s focus is on quantity rather than quality, the agency has announced that the winning vendor for the Extreme Vetting Initiative contract must “generate a minimum of 10,000 investigative leads annually” – without regard to how many leads are actually appropriate.14
This initiative is tailor-made for discrimination. The proposed program’s aims – to evaluate whether an individual will become “a positively contributing member of society” or whether he or she “intends to commit criminal or terrorist attacks” – are lifted directly from the president’s “travel ban” executive orders,15 multiple versions of which federal courts enjoined as discriminatory.16 Indeed, President Trump has expressly said that his proposed ban on Muslim immigration had “morphed into extreme vetting for certain areas of the world” – the very name of this initiative.17
The Extreme Vetting Initiative will also undoubtedly chill free expression, contravening the First Amendment and international human rights, such as those contained in the Universal Declaration of Human Rights, for which the United States has registered official support, and the International Covenant on Civil and Political Rights, to which the U.S. is a party. ICE will require the chosen contractor to “analyze and apply techniques to exploit publically available information” from a breathtakingly wide range of online sources, including “media, blogs, public hearings, conferences, academic websites, social media websites such as Twitter, Facebook, and LinkedIn, radio, television, press, geospatial sources, internet sites, and specialized publications.”18 The knowledge that the government will be regularly scouring online statements to make admission or deportation determinations will unquestionably pressure both visa applicants and recipients – and the people with whom they communicate – to censor themselves online. With applicants’ online private lives open to government scrutiny, and without the ability to provide context to, or explanation for, online content or connections, visa applicants and holders are likely to self-censor or delete accounts with consequences for personal, business and travel-related activity. These risks are particularly acute in light of existing initiatives to ask travelers to identify all of their social media handles in order to obtain permission to travel to the United States.
It seems highly likely that this surveillance will extend to permanent residents, as well as temporary visa holders. The program’s Statement of Objectives says that the failure to continuously vet permanent residents creates “significant risk in ICE’s ability to identify emerging risks, such as radicalization, that may occur after an individual arrives in the United States.”19 When one contractor expressly asked ICE whether it wanted to continuously vet permanent residents, the agency said that it would answer the question at a later date.20
Through the Extreme Vetting Initiative, ICE seeks to automate the process by which the U.S. government targets, finds, and forcibly removes people from our country, likely including permanent residents. But this system will not work the way ICE says it will work. Instead, it risks hiding politicized, discriminatory decisions behind a veneer of objectivity – at great cost to freedom of speech, civil liberties, civil rights, and human rights. It will hurt real, decent people and tear apart families. We urge you to immediately and publicly halt work on this program.
18 Million Rising
Advocacy for Principled Action in Government
American-Arab Anti-Discrimination Committee
American Civil Liberties Union
Asian Americans Advancing Justice
Brennan Center for Justice
Center for Constitutional Rights
Center for Democracy & Technology
Center for Media Justice
The Center for Security, Race, and Rights, Rutgers Law School
Center on Privacy & Technology at Georgetown Law
Color of Change
Committee to Protect Journalists
The Constitution Project
Council on American-Islamic Relations
The Concerned Archivists Alliance Defending Rights & Dissent
Electronic Frontier Foundation
Electronic Privacy Information Center (EPIC)
Friends Committee on National Legislation
Government Accountability Project
Government Information Watch
Human Rights Watch
Human Rights Data Analysis
Group Immigrant Law Center of Minnesota
The Identity Project
Lawyers' Committee for Civil Rights Under Law
The Leadership Conference on Civil and Human Rights
Legal Aid Justice Center
Muslim Advocates Muslim Justice League
Muslim Public Affairs Council (MPAC) NAACP
National Hispanic Media Coalition
National Immigration Law Center
National Immigration Project of the National Lawyers Guild
National Iranian American Council (NIAC)
New America's Open Technology Institute
Online Policy Group
Open MIC (Open Media and Information Companies Initiative)
People for the American Way
Restore the Fourth
South Asian Americans Leading Together (SAALT)
Southern Poverty Law Center
Services, Immigrant Rights, and Education Network (SIREN)
Union for Reform Judaism
Woodhull Freedom Foundation
1 See Immigration & Customs Enforcement Homeland Security Investigations (hereinafter “ICE-HSI”), “ICE-HSI- Data Analysis Service: Solicitation Number HSCEMD-17-R-0010,” FedBizOpps.Gov, June 12, 2017.
2 ICE-HSI, “Extreme Vetting Initiative: STATEMENT OF OBJECTIVES (SOO),” June 12, 2017, available at FedBizOpps.Gov (hereinafter “Statement of Objectives”).
3 ICE-HSI, “Q&A from Vendor Emails,” July 27, 2017, available at FedBizOpps.Gov.
4 ICE-HSI, “Statement of Objectives.”
5 ICE-HSI, “Q&A July 19, 2017,” July 27, 2017, available at FedBizOpps.Gov (“We know this type of analysis is being used in private industry (i.e. looking for outliers and potential causes of risk) as part of their business. We are looking to do this for national security.”).
6 Letter from 30+ Technology Experts to the Honorable Elaine Duke, Secretary of Homeland Security, Nov. 16, 2017.
7 See, e.g., The MITRE Corporation, JASON Program Office, Rare Events, Oct. 2009 at 21 (concluding that “Social science-based models do not yet exist for anticipating and interceding in rare [weapons of mass destruction/terrorism] events” and that “It is unreasonable to aim for predictive models of specific rare events.”). 8 See, e.g., Aaron Cantú & George Joseph, “Trump’s Border Security May Search Your Social Media by ‘Tone,’” The Nation, Aug. 23, 2017, available at https://www.thenation.com/article/trumps-border-security-may-search-your- social-media-by-tone/.
9 ICE-HSI, “Background,” June 12, 2017, available at FedBizOpps.Gov (“Task 3: Social Media Exploitation.”).
10 See Department of Homeland Security, Office of Inspector General, “DHS’ Pilots for Social Media Screening Need Increased Rigor to Ensure Scalability and Long-term Success (Redacted),” Feb. 27, 2017. Curiously, contract documents also suggest that ICE-HSI has found a means to avoid technical measures erected by social media companies to prevent precisely this kind of surveillance. See ICE-HSI, “CTCEU Responses to Vendor Questions,” July 13, 2017, available at FedBizOpps.Gov (“[Q:] Certain social media sites have made their APIs private, such as Facebook. Can the vendor assume that the Facebook reference was just for illustration purposes? [A:] No, the Contractor shall analyze and apply techniques to exploit publically available information, including social media websites such as Facebook.”).
11 See, e.g., Fox News, “Twitter joke to ‘destroy America’ reportedly gets U.K. tourists barred from US,” Jan. 30, 2012.
12 See Ahmed Abbasi, Ammar Hassan & Milan Dhar, Benchmarking Twitter Sentiment Analysis Tools, Proceedings of the 9th Language Resources and Evaluation Conference (2014).
13 See Julia Hirschberg & Christopher D. Manning, Advances in Natural Language Processing, 349 SCIENCE 6245 (July 17, 2015). See also Su Lin Blodgett & Brendan O’Connor, Racial Disparity in Natural Language Processing: A Case Study of Social Media African-American English, Proceedings of the 2017 Fairness, Accountability, and Transparency in Machine Learning Conference (2017) (showing failure to perform on English text as used by a specific demographic community).
14 ICE-HSI, “Background,” June 12, 2017, available at FedBizOpps.Gov.
15 See Exec. Order 13,769 (Jan. 27, 2017) at sec. 4 (“[Various federal agencies] shall implement…a process to evaluate the applicant's likelihood of becoming a positively contributing member of society and the applicant's ability to make contributions to the national interest; and a mechanism to assess whether or not the applicant has the intent to commit criminal or terrorist acts after entering the United States.”). The order’s successor, Executive Order 13780, omitted the “positively contributing member of society” language, but still called for “a mechanism to assess whether applicants may commit, aid, or support any kind of violent, criminal, or terrorist acts after entering the United States.” See Exec. Order 13,780 (March 16, 2017) at sec. 5.
16 See Washington v. Trump, No. 17-35105, slip op (W.D. Wash.), rev’d, 847 F.3d 1151 (9th Cir. Feb. 9, 2017); Aziz v. Trump, 234 F.Supp.3d 724, slip op. (E.D. Va. Feb. 13, 2017); Hawaii v. Trump, 859 F.3d 741, 771 (9th Cir. 2017), cert. granted sub nom Trump v. Int’l Refugee Assistance Project, 137 S. Ct. 2080 (2017); Int’l Refugee Assistance Project v. Trump, 241 F.Supp.3d 539 (D. Md.), rev’d, 857 F.3d 554 (4th Cir. May 25, 2017).
17 “Transcript of the Second Debate,” New York Times, Oct. 10, 2016, available at https://www.nytimes.com/2016/10/10/us/politics/transcript-second-debate.html?mcubz=1&_r=0.
18 ICE-HSI, “Background,” June 12, 2017, available at FedBizOpps.Gov.
19 ICE-HSI, “Statement of Objectives.”
20 ICE-HSI, ““CTCEU Responses to Vendor Questions,” July 13, 2017, available at FedBizOpps.Gov.