Federal government agencies such as the Department of Homeland Security (DHS) have dramatically expanded their social media monitoring programs in recent years, collecting a vast amount of user information in the process – including political and religious views, data about physical and mental health, and the identity of family and friends. DHS increasingly uses this information for vetting and analysis, including for individuals seeking to enter the United States and for both U.S. and international travelers.
But while the government has justified its expansion in the name of national security, there is little indication that social media monitoring programs – or the algorithms that sometimes power them – are effective in achieving their stated goals. Additionally, there is evidence that DHS is using personal information extracted from social media posts to target protestors and religious and ethnic minorities for increased vetting and surveillance. In a new report, Social Media Monitoring, the Brennan Center provides an overview of DHS social media monitoring programs and the new set of challenges that they are surfacing.
The effectiveness of social media monitoring programs is unproven
The recent growth of DHS surveillance systems, including its social media monitoring programs, has been rapid. In Social Media Monitoring, the Brennan Center outlines how social media monitoring is used across various arms of DHS, including Customs and Border Protection (CBP), the Transportation Security Administration (TSA), U.S. Immigration and Customs Enforcement (ICE), and U.S. Citizenship and Immigration Services (USCIS).
DHS social media monitoring programs have expanded in parallel with the proliferation of social media information and, consequentially, the growing number of companies creating products that claim to interpret that information. “It’s a coupling of the explosion of information on social media and the emergence of algorithmic tools that purport to be able to analyze it and come up with meaningful results,” said Faiza Patel, co-director of the Brennan Center’s Liberty and National Security Program.
But despite their expansion, the DHS programs have not proven successful, even based on the department’s own measures. For example, after USCIS piloted five social media monitoring programs in 2016, the agency’s own evaluations found the programs largely ineffective in identifying threats to public safety or national security. Indeed, for three out of the four programs used to vet refugees, “the information in the accounts did not yield clear, articulable links to national security concerns, even for those applicants who were found to pose a potential national security threat based on other security screening result,” according to a DHS brief.
These DHS pilot programs and their subsequent evaluations highlight several of the central challenges associated with social media monitoring. One major challenge is the difficulty of actually interpreting what’s in the social media messages and connecting them to actual threats. These interpretation problems become even more complex when a non-English language or unfamiliar cultural context is involved. The programs themselves also carry civil liberties risks. “They give the government a pool of information about people’s personal lives and political and religious beliefs that can easily be abused. And research shows that people censor themselves when they know the government is watching,” said Rachel Levinson-Waldman, senior counsel in the Brennan Center’s Liberty and National Security Program.
Social media monitoring algorithms are unreliable – and could be discriminatory
There is yet another key complicating factor in social media monitoring programs – the increasing use of algorithmic tools to review social media posts. These tools and methods, which include natural language processing and algorithmic tone and sentiment analysis, have high error rates. This makes it questionable that they are actually capable of achieving DHS objectives, particularly because of the open-ended nature of the evaluations they are used for, such as identifying national security threats.
Equally troubling, the algorithms that are deployed for social media monitoring are susceptible to bias. “Our experience with algorithmic tools shows that they tend to operate in a discriminatory fashion,” said Patel. “They make judgments based on proxies, and when these proxies reflect biases, the results produced by an algorithm simply reproduce those biases. For example, the biases evident in the early versions of the Trump administration’s Muslim ban could be coded into an algorithm, resulting in the flagging of many Muslims as a national security threat.” Since even before the ban, federal agencies such as the FBI and the Department of Defense have used religious beliefs as markers of dangerousness.
The hard questions that DHS needs to consider
One barrier to addressing DHS’s expansion of its social media monitoring programs is the lack of visibility into the full scope of the department’s surveillance capabilities, a gap this report seeks to address. In addition, there is currently minimal oversight of these programs, including from Congress.
“Congress should look closely at these DHS programs and ask the basic questions,” said Patel. “In what contexts is the Department monitoring social media? How is it verifying the accuracy of accounts being attributed to individuals? What kinds of decisions is it using this data for? How is the information being shared? And how is the effectiveness of these programs being measured?”
Read the full Brennan Center report, Social Media Monitoring.