Artificial Intelligence Use Case Inventory

Sentiment Analysis and Topic Modeling (SenTop)

The initial purpose of the Sentiment Analysis and Topic Modeling (SenTop) project was to analyze survey responses for DHS’s Office of the Chief Procurement Officer related to contracting. However, it has evolved to be a general-purpose text analytics solution that can be applied to any domain/area. It also has been tested/used for human resources topics. SenTop is a DHS-developed Python package for performing descriptive text analytics. Specifically, sentiment analysis and topic modeling on free-form, unstructured text. SenTop uses several methods for analyzing text including combining sentiment analyses and topic modeling into a single capability, permitting identification of sentiments per topic and topics per sentiment. Other innovations include the use of polarity and emotion detection, fully automated topic modeling, and multi-model/multi-configuration analyses for automatic model/configuration selection. The code has been established, performs an analysis, and provides a report but it is only accessed and run by one person per customer request.

What stage of production is the AI in? Development/testing: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

AIS Scoring & Feedback (AS&F)

Automated Indicator Sharing (AIS), a CISA capability, enables the real-time exchange of machine-readable cyber threat indicators and defensive measures to help protect against and ultimately reduce the prevalence of cyber incidents. AIS is offered as part of CISA’s broad authority to share information relating to cybersecurity risks, including authority to receive, analyze, and disseminate information, and fulfills CISA’s obligation under the Cybersecurity Information Sharing Act of 2015 to establish and operate the federal government’s capability and process for receiving cyber threat indicators and defensive measures, and to further share this information with certain other agencies, in some cases in a real-time manner. For more information, please visit: https://www.cisa.gov/ais.

AIS Automated Scoring & Feedback (AS&F), built on the AIS Scoring Framework, defines an algorithm by which organizations can enrich Structured Threat Information Expression Indicator objects, shared via AIS, with (1) an opinion value that provides an assessment of whether or not the information can be corroborated with other sources available to the entity submitting the opinion and (2) a confidence score that states the submitter’s confidence in the correctness of information they submit into AIS. When leveraged by CISA, AS&F uses artificial intelligence / machine learning to perform descriptive analytics from organizational-centric intelligence to support confidence and opinion classification of indicators of compromise. Together, these enrichments can help those receiving information from AIS prioritize actioning and investigating Indicator objects.

What stage of production is the AI in? In production: less than 6 months
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Automated PII Detection

CISA's Automated Personally Identifiable Information (PII) Detection and Human Review Process incorporates descriptive, predictive, and prescriptive analytics. Automated PII Detection leverages natural language processing tasks including named entity recognition coupled with Privacy guidance thresholds to automatically detect potential PII from within Automated Indicator Sharing submissions. If submissions are flagged for possible PII, the submission will be queued for human review where the analysts will be provided with the submission and artificial intelligence-assisted guidance to the specific PII concerns. Within human review, analysts can confirm/deny proper identification of PII and redact the information (if needed). Privacy experts are also able to review the actions of the system and analysts to ensure proper performance of the entire process along with providing feedback to the system and analysts for process improvements (if needed). The system learns from feedback from the analysts and Privacy experts.

Through the incorporation of the automated PII detection, CISA complies with Privacy, Civil Rights and Civil Liberties requirements of CISA 2015 and scaled analyst review of submissions by removing false positives and providing guidance to submission to be reviewed. Through continual audits CISA will maintain integrity and trust in system and human processes. For more information, please visit: https://www.cisa.gov/ais.

What stage of production is the AI in? In production: less than 6 months
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

CDC Airport Hotspot Throughput (PageRank)

TSA launched the “Stay Healthy. Stay Secure.” campaign, which details proactive and protective measures have been implemented at security checkpoints to make the screening process safer for passengers and our workforce by reducing the potential of exposure to the coronavirus. The campaign includes guidance and resources to help passengers prepare for the security screening process in the COVID environment. A big part of that campaign was the development of the Centers for Disease Control and Prevention's Airport Hotspot Throughput. This capability determines the domestic airports that have the highest rank of connecting flights during the holiday travel season to help mitigate the spread of COVID-19. This capability is a DHS-developed artificial intelligence model written in Spark/Scala that takes historical non-PII travel data and computes the highest-ranking airports based on the PageRank algorithm. TSA does not make decisions about flight cancellations or airport closures. These decisions are made locally, on a case-by-case basis, by individual airlines, airports, and public health officials. TSA will continuously evaluate and adapt procedures and policies to keep the public and our workforce safe as we learn more about this devastating disease and how it spreads.

What stage of production is the AI in? Development/testing: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Asylum Text Analytics (ATA)

USCIS oversees lawful immigration to the United States. As set forth in Section 451(b) of the Homeland Security Act of 2002, Public Law 107-296, Congress charged USCIS with administering the asylum program. USCIS, through its Asylum Division within the Refugee, Asylum & International Operations Directorate (RAIO), administers the affirmative asylum program to provide protection to qualified individuals in the United States who have suffered past persecution or have a well-founded fear of future persecution in their country of origin, as outlined under Section 208 of the Immigration and Nationality Act (INA), 8 U.S.C. § 1158 and Title 8 of the Code of Federal Regulations (C.F.R.), Part 208. Generally, an individual not in removal proceedings may apply for asylum through the affirmative asylum process regardless of how the individual arrived in the United States or his or her current immigration status by filing Form I-589, Application for Asylum and for Withholding of Removal. The ATA capability employs machine learning and data graphing techniques to identify plagiarism-based fraud in applications for asylum status and for the withholding of removal by scanning the digitized narrative sections of the associated forms and looking for common language patterns.

What stage of production is the AI in? In production: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

BET/FBI Fingerprint Success Maximization

USCIS's Customer Profile Management Service (CPMS) serves as a person-centric repository of biometric and biographic information provided by applicants and petitioners (hereafter collectively referred to as “benefit requestors”) that have been issued a USCIS card evidencing the granting of an immigration related benefit (i.e., permanent residency, work authorization, or travel documents). The Biometrics Encounter Tool (BET) / Federal Bureau of Investigation (FBI) Fingerprint Success Maximization Service center technicians can receive immediate feedback when a set of prints is likely to be rejected by the FBI by incorporating machine learning models into the BET application. The FBI will not disclose their quality grading criteria for fingerprints, leaving CPMS with the responsibility of determining quality to prevent unnecessary secondary encounters with applicants. Using even the simplest of models would catch 98% of rejected submissions, which could have potentially saved USCIS from scheduling 42,763 additional appointments in 2020. This would come at the cost of forcing recapture during 11% of encounters. This effort aims to maximize the number of successful FBI submissions while minimizing the number of fingerprint recaptures necessary. For more information, please visit: https://www.dhs.gov/publication/dhsuscispia-060-customer-profile-management-service-cpms

What stage of production is the AI in? Coordinating rollout, planned for early March 2022
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Biometrics Enrollment Tool (BET) Fingerprint Quality Score

USCIS's Customer Profile Management Service (CPMS) serves as a person-centric repository of biometric and biographic information provided by applicants and petitioners (hereafter collectively referred to as “benefit requestors”) that have been issued a USCIS card evidencing the granting of an immigration related benefit (i.e., permanent residency, work authorization, or travel documents). The Biometrics Enrollment Tool (BET) team has been working on enhancing their quality checks, with one of the new improvements being incorporation of the National Institute of Standards and Technology (NIST) Fingerprint Image Quality 2 (NIFQ2) algorithm (a trained machine learning algorithm) for scoring of fingerprints (https://www.nist.gov/services-resources/software/nfiq-2) into the BET application. This algorithm takes a fingerprint image and assigns a score between 0 - 100, with 100 indicating that this is the best quality fingerprint image that could be obtained. The higher the score, the more likely that the fingerprint will match when captured again. This algorithm has been in place for several Program Increments. BET had been providing Biometric Capture Technicians with a poor-quality indicator and encountered objections from technicians for the larger than expected number of recaptures required, based on contractual complications. The BET team continues to capture this data in the background, but this does not require recapture currently. For more information, please visit: https://www.dhs.gov/publication/dhsuscispia-060-customer-profile-management-service-cpms

What stage of production is the AI in? In use currently
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Evidence Classifier

USCIS is the component within DHS that oversees lawful immigration to the United States. USCIS receives immigration requests from individuals seeking immigration and non-immigration benefits. Once a benefit request form is submitted to USCIS, a series of processing and adjudication actions occur. One of the case management systems used to track and adjudicate certain immigration request forms is the Electronic Information System (ELIS). USCIS ELIS is an internal case management system composed of microservices to assist with performing complex adjudicative and processing tasks; one of those microservices is the Evidence Classifier. Until the introduction of the Evidence Classifier machine learning (ML) solution, those who are working cases and who are responsible for reviewing evidence documents would often have to sift through dozens, if not hundreds, of unlabeled pages to find one specific artifact — be that a green card, a birth certificate, or so on. To reduce the amount of adjudicative time spent on these repetitive tasks, a ML solution was built to systematically tag individual pages with some of the highest-volume, highest-impact evidence types. Calculated from September 28, 2021 , to May 20, 2022, the ML enhancements have saved around 24 million page scrolls, which amounts to approximately 13,348 hours saved, assuming it takes 2 seconds to review 1 page of evidence. This has nearly doubled cases with a 30-day adjudication rate from about 30% to 58%. For more information, please visit: https://www.dhs.gov/publication/dhsuscispia-056-uscis-electronic-immigration-system-uscis-elis

What stage of production is the AI in? In production: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

FDNS-DS NexGen

USCIS created the Fraud Detection and National Security (FDNS) Directorate to strengthen the integrity of the nation’s immigration system and to ensure that immigration benefits are not granted to individuals that may pose a threat to national security and/or public safety. In addition, the FDNS Directorate is responsible for detecting, deterring, and combating immigration benefit fraud, hence the creation of the FDNS Data System (FDNS-DS). FDNS-DS NexGen is a case management system, that is planning to use resolved artificial intelligence (AI) / machine learning (ML) entities from other applications to aid in investigative work, enhance investigative case prioritization, and detect duplicate case work. In the future, there are plans to integrate AI/ML into the predictive modeling for future system enhancements, working side-by-side with the business stakeholders to develop best practices. Fraud occurs in numerous ways; being able to discover and detect persons with multiple identities allows for more comprehensive investigations, reduces investigative cycle time, and improves performance . Those basic implementations will speed up processing by several magnitudes. For more information, please visit: https://www.dhs.gov/publication/dhsuscispia-013-01-fraud-detection-and-national-security-directorate

What stage of production is the AI in? Development/testing: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Sentiment Analysis

The USCIS Service Center Operations Directorate (SCOPS) provides services for persons seeking immigration benefits while ensuring the integrity and security of our immigration system. As part of that mission, we issued a two-part survey asking users both quantitative and qualitative questions. USCIS performed a statistical analysis of the quantitative results and then used Natural Language Processing modeling software to assign "sentiments" to categories ranging from strongly positive to strongly negative. This model was eventually enhanced using a machine learning model to have better reusability and performance. This capability has been deployed to production for more than one year.

What stage of production is the AI in? In production: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Testing Performance of ML Model using H2O

USCIS is the component within DHS that oversees lawful immigration to the United States. That means USCIS receives, processes, and maintains all applications for admission for Lawful permanent residents (LPRs), or adjustments to LPR status. Also known as “green card” holders, LPRs are non-citizens who are lawfully authorized to live permanently within the United States and are required to fill out Form I-90, Application to Replace Permanent Resident Card (Green Card). Since there has been a considerable influx of green card applications, USCIS used a combination of exploratory data analysis to determine the most used categories for applicants submitting I-90's, and machine learning to create predictions of workloads. USCIS used the H20 machine learning model to allow USCIS analysts to build and run several machine learning models on big data in an enterprise environment and identify the model that performs the best. It has already been successful in identifying the most accurate model for the I-90 Form Timeseries Analysis and Forecasting use case. This capability has been in production for more than one year.

What stage of production is the AI in? In production: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Timeseries Analysis and Forecasting

USCIS is the component within DHS that oversees lawful immigration to the United States. That means USCIS receives, processes, and maintains all applications for admission for Lawful permanent residents (LPRs), or adjustments to LPR status. Also known as “green card” holders, LPRs are non-citizens who are lawfully authorized to live permanently within the United States and are required to fill out Form I-90, Application to Replace Permanent Resident Card (Green Card). Since there has been a considerable influx of green card applications, USCIS used a combination of exploratory data analysis to determine the most used categories for applicants submitting I-90's and machine learning to create predictions of workloads. As a follow-on, USCIS used Autoregressive Integrated Moving Average (ARIMA) models on the I-90 form, which allowed the prediction of the total number of forms for a 2-year period. ARIMA is one of the easiest and effective machine learning algorithms to perform time series forecasting. This capability has been deployed in production for more than a year. This model was eventually enhanced using ML model to have better reusability and performance.

What stage of production is the AI in? In production: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Silicon Valley Innovation Program (SVIP) Language Translator

USCG operators must be able to communicate with vessel occupants - many who may be non-English speakers - while performing a variety of rescue and investigative missions. The accurate and swift translation of information is critical to the safety and security of Coast Guard boarding teams and vessel occupants. The DHS’s Science and Technology Silicon Valley Innovation Program (SVIP) Language Translator solicitation sought new capabilities to support the Coast Guard in facilitating real-time communications with non-English speakers and those who are unable to communicate verbally. The solicitation also included requirements for language translation technology to be capable of operating both online and offline because many Coast Guard interactions take place in extreme environmental conditions, and in locations without cell service or an internet connection. There are two performers, myLanguage and Kynamics, who have designed an online/offline voice-to-text speech recognition and text-to-text translation system that employs deep learning and artificial intelligence. myLanguage completed Phase I and Kynamics completed Phase 2, adapting their voice translation technologies for use into a rugged, hand-held mobile device that can withstand extreme temperatures and, customized model designs and training language models to fit Coast Guard use cases.

What stage of production is the AI in? Development/testing: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Agent Portable Surveillance

The agent portable surveillance system is a backpack mobile unit meant for single agent deployments. The system identifies border activities of interest by using artificial intelligence / machine learning to analyze data from Electro-Optical/Infra-Red cameras and radar. When an activity is detected, the system sends the information to agents through the Team Awareness Kit (TAK). Detections are shared with CBP TAK users to enhance efficiency and agent/officer safety.

What stage of production is the AI in? Development/testing: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Autonomous Surveillance Towers

The system permits autonomous detection, identification, and tracking of items of interest. The tower scans constantly and autonomously; radar detects and recognizes movement; and the camera slews autonomously to the items of interest and the system software identifies the object. The system utilizes artificial intelligence / machine learning to analyze the camera and radar data which alerts the user and autonomously tracks the item of interest. End users can monitor the system and see near real-time photos by logging into the User Interface on any USCBP device.

What stage of production is the AI in? In production: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

I4 Viewer Matroid Image Analysis

Matroid is a software that enables CBP end users to create and share vision detectors. Matroid detectors are trained computer vision models that recognize objects, people, and events in any image and in video streams. Once a detector is trained, it can monitor streaming video in real time, or efficiently search through pre-recorded video data or images to identify objects, people, and events of interest. Users can view detection information via a variety of reports and alert notifications to process and identify important events and trends. Detection data is also available through Matroid’s powerful developer Application Programming Interface and language-specific clients, so CBP applications can be integrated with the power of computer vision.

What stage of production is the AI in? In production: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Open-source News Aggregation

The platform enables users to make better decisions faster by identifying and forecasting emerging events on a global scale to mitigate risk, recognize threats, greatly enhance indications and warnings, and provide predictive intelligence capabilities. The artificial intelligence / machine learning models enable rapid access to automated intelligence assessments by fusing, processing, exploiting and analyzing open sources of data (including news, social media, economic indicators, governance indicators, travel warnings, weather and other sources). This system is an immediate and substantial force multiplier that shifts the traditional approach of monitoring and assessing the operational environment to focus on the forecast of the future geopolitical, socio, and economic environment.

What stage of production is the AI in? In production: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Data Tagging and Classification

The Homeland Security Investigations (HSI) Innovation Lab is developing an analytical platform called the Repository for Analytics in a Virtualized Environment (RAVEn). RAVEn facilitates large, complex analytical projects to support ICE’s mission to enforce and investigate violations of U.S. criminal, civil, and administrative laws. RAVEn also enables users to develop new tools to analyze trends and isolate criminal patterns as HSI mission needs arise. For more information, please read the DHS/ICE/PIA-055 - Privacy Impact Assessment 055 for the Repository for Analytics in a Virtualized Environment (RAVEn).

RAVEn leverages data tagging and classification to do the following:

The Email Analytics Tool streamlines how special agents and criminal analysts search, filter, translate, and report on electronic communications evidence and will help investigators more effectively determine the structure and organization of criminal enterprises.

The RAVEn - Lead Tracker is a centralized system where agents can send and receive leads and enter outcomes such as arrests and seizures. The goal is for all leads in the agency to be found in one place, rather than in various email inboxes.

The overarching goal of Mobile Device Analytics is to improve the efficiency of agents and analysts in identifying pertinent evidence, relationships, and criminal networks from data extracted from mobile devices.

What stage of production is the AI in? In production: less than 6 months
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

Language Translator

The Homeland Security Investigations (HSI) Innovation Lab is developing an analytical platform called the Repository for Analytics in a Virtualized Environment (RAVEn). RAVEn facilitates large, complex analytical projects to support ICE’s mission to enforce and investigate violations of U.S. criminal, civil, and administrative laws. RAVEn also enables users to develop new tools to analyze trends and isolate criminal patterns as HSI mission needs arise. For more information, please read the DHS/ICE/PIA-055 - Privacy Impact Assessment 055 for the Repository for Analytics in a Virtualized Environment (RAVEn).

RAVEn has incorporated a machine translation system for commercial vendors, Systran, which provides machine learning translation for over 100 different language combinations. Currently, the Innovation Lab has licenses for translating Chinese, Spanish, Arabic, Farsi, Russian, German, Ukrainian, and Filipino to English. Systran can translate plain text, word documents, and PDFs. Both a web-based user interface (UI) and Application Programming Interface (API) endpoints are available. Machine learning is used to increase the efficiency, accuracy, and quality of searching, analyzing, and translating speech.

What stage of production is the AI in? Development/testing: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

RAVEn Compliance Automation Tool (CAT)

The Homeland Security Investigations (HSI) Innovation Lab is developing an analytical platform called the Repository for Analytics in a Virtualized Environment (RAVEn). RAVEn facilitates large, complex analytical projects to support ICE’s mission to enforce and investigate violations of U.S. criminal, civil, and administrative laws. RAVEn also enables users to develop new tools to analyze trends and isolate criminal patterns as HSI mission needs arise. For more information, please read the DHS/ICE/PIA-055 - Privacy Impact Assessment 055 for the Repository for Analytics in a Virtualized Environment (RAVEn).

RAVEn CAT is being developed as part of an effort to modernize HSI’s Form I-9 Inspection Process. The goal is to use machine learning and automation to increase the speed and efficiency of ingesting and processing Forms I-9 data. Easy to use front-end interface workflow that increases work productivity and reduces manual entry. RAVEn CAT currently employs an Optical Recognition Service (OCR) model and software (Tesseract OCR) to identify pixel coordinates of handwritten and read/extract computer typed characters from ingested forms for processing. Additional research into opensource Machine Learning Object Detection models is being made to help further augment accuracy of text identification and extraction of ingested forms into the pipeline.

What stage of production is the AI in? In production: less than 6 months
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team

RAVEn Normalization Services

The Homeland Security Investigations (HSI) Innovation Lab is developing an analytical platform called the Repository for Analytics in a Virtualized Environment (RAVEn). RAVEn facilitates large, complex analytical projects to support ICE’s mission to enforce and investigate violations of U.S. criminal, civil, and administrative laws. RAVEn also enables users to develop new tools to analyze trends and isolate criminal patterns as HSI mission needs arise. For more information, please read the DHS/ICE/PIA-055 - Privacy Impact Assessment 055 for the Repository for Analytics in a Virtualized Environment (RAVEn).

Within RAVEn, HSI has utilized artificial intelligence / machine learning to verify, validate, correct, and normalize the accuracy and quality of addresses, phone numbers, and names. The normalization services let agents analyze both well-defined addresses (such as those in CONUS and Europe) and less well-defined addresses (such as addresses using mile markers); standardize phone numbers to their identified country and to the E164 ITU standard; and streamline the process of correcting data entry errors and/or pointing out purposeful misidentification, connecting information about a person across HSI datasets, and cutting down the number of resource hours needed for investigations.

What stage of production is the AI in? Development/testing: more than 1 year
If requested, would a FOIA exemption prevent the release of the use case?  No
For more information, please contact: DHS Artificial Intelligence Team