USCIS automating pre-processing of immigration cases

New tools will dissect supporting evidence, making it easier for adjudicators to make decisions to award people benefits like green cards.
(Getty Images)

U.S. Citizenship and Immigration Services is focused on automating functions that will help pre-process immigration cases for adjudication, according to CTO Rob Brown.

Natural language processing helps harvest names for adjudicators and flag potential fraud when applicants’ stories don’t align, machine learning (ML) combs biographic and biometric data to identify people with USCIS benefits, and network analytics make connections regarding their relationships and employers, Brown said.

New tools will dissect supporting evidence related to immigration cases, making it easier for adjudicators to make decisions to award people benefits like green cards.

“Now we start to think about a lot of that pre-processing of adjudication really up front, as opposed to it being manually done or swivel chaired at an adjudicator’s workstation or workstations,” Brown said during an AI in Government event. “So providing a lot of that information upfront.”


Computer vision and optical character recognition will be used to validate documents and classify evidence, so adjudicators can click on what they want rather than sort through.

Identity proofing like mobile verification and sentiment analysis are proving more challenging, Brown said.

“We, I feel, need industry experts and assistance in looking at what does this mean from a privacy perspective and abating some of the challenges therein,” he said. “What does this mean from a security perspective?”

Identity validation presents a number of cyberattack vectors when doing something as seemingly benign as verifying photos or videos of people.

Presentation-layer, man-in-the-middle, and backend and data poisoning attacks are all possible.


“Simple things like Avatarify and even TikTok technologies have creeped in,” Brown said. “So I feel this is an area we need a lot of help with.”

Brown also hopes to deal with ML and artificial intelligence “sprawl” by consolidating toolsets and platforms to provide a more robust continuous integration/continuous delivery (CI/CD) pipeline.

Proper experimentation on algorithms that accounts for security and their sharing is also important, Brown said.

USCIS is still trying to solve the problem of data bias by automating algorithms to filter out biased data, audit pipelines and flag where data quality issues persist, Brown said.

Brown hopes to see more adaptive automated services embedding customer and adjudicator personas before 2025.

Latest Podcasts