Every year, federal agencies churn out an average of 3,700 proposed rules. They ask for the public to comment on them, and that results in a torrent of comments, often orchestrated campaigns by opponents or supporters—and some that are fabricated. A 2021 Government Accountability Office study of proposed rules found that depending on the agency, from 5 to 30 percent of the people whose emails were attached to comments didn’t make them.
If technology helps produce the flood of comments, including suspect ones, technology can help agencies wade through them. Artificial intelligence (AI), in particular natural language processing (NLP), can help federal workers comb through thousands of comments quickly to find themes and see that many are similar (suggesting a campaign) when manual reviews could take weeks or months. At the Environmental Protection Agency, for example, since 2013 Abt has reviewed more than 1.4 million comments for the annual Renewable Fuel Standard program. NLP can summarize themes in comments to enable agency staff to develop robust responses and help agencies rapidly identify any potential issues in their proposed rules.
Reviewing comments is a major government responsibility, but it’s far from the only way Abt’s 40 AI projects in the last five years have helped a wide variety of agencies. We’ve shown that AI can be transformative, making public services more effective and efficient. From creating predictions for PFAS blood levels and delays in nursing care to improving workflows by automating contract quality control and processing, AI sets our U.S. and global clients up for success in all the areas where we work, from health to housing.
Like any tool, AI must be used carefully and correctly. The underlying data that AI relies on may contain unintended biases or inaccuracies, leading to unfavorable results. To avoid such outcomes, Abt is building guardrails around our use of AI. For example, our commitment to equity means we include representation of key stakeholders in AI project design, implementation, and evaluation so that we can consider the implications for equity and any impact on communities and individuals. And to avoid poor quality, our subject-matter experts and computer scientists collaborate to produce AI tools that combine cutting-edge tech solutions within a nuanced understanding of the implications for our clients’ needs, the issues, populations, and solutions at hand.
Our clients use AI to reduce their work burdens and find novel solutions. We help them do that. AI in various forms—NLP, predictive analytics, and generative AI—is a critical development. No one knows for sure how it will evolve, but we are certain that it will. To see how we have provided and will continue to provide pragmatic and safe solutions to help public sector organizations achieve their goals, see examples of our work below.
AI has made Abt’s work better, faster, and more efficient. AI can rapidly process vast amounts of data and identify patterns and relationships that may otherwise remain hidden. This enables Abt to generate insights and recommendations with greater speed and accuracy than traditional research methods.
What the future holds: Generative AI will improve technical assistance by explaining complex topics in language anyone can understand. It will enhance quantitative research by spawning novel ideas; rather than merely synthesizing existing content, a generative AI algorithm can create something entirely new—images, music, text, ideas —based on the parameters and data it is given.
Abt is pursuing a wide variety of proofs of concept to contribute to future AI uses for social good. They include:
- Quality control. Abt worked on a demonstration for the Army Corps of Engineers to show how it could use AI and machine learning for cost estimates to incorporate more data than traditional methods and thus produce more accurate forecasts. And the U.S. Agency for International Development’s DHIS2, a health information database, experiences quality control issues that automation could mitigate.
- Disinformation. We are building social listening tools to detect and counter disinformation campaigns for COVID, tuberculosis, and vaccinations.
- Climate prediction. Health agencies could use AI to predict extreme weather events and provide warning to vulnerable populations with special health needs. With the predictions in hand, social services agencies could connect those affected to the care they need and send money to residents’ checking accounts before a natural disaster hits.
Support to the Per- and Polyfluoroalkyl Substances (PFAS) Work
Client: Centers for Disease Control and Prevention
Per- and polyfluoroalkyl substances (PFAS) are man-made contaminants found in drinking water that lead to a variety of health impacts, from low birthweight to impaired immune systems to increased cancer risk. Understanding the amount of PFAS in a given person’s blood is important for that person and for governments and agencies trying to set clear guidance on risk levels in communities. But PFAS blood monitoring is not readily accessible. Our team used AI predictive analytics to develop a model both for researchers and the general public to predict blood levels based on drinking water exposure. They focused on two parameters and incorporated hundreds of datapoints from multiple studies to refine the estimates of those two parameters. The model improves our ability to estimate PFAS blood levels.
Intermodal Transportation Recommender System
Client: U.S. Army Corps of Engineers
Abt built a recommender system using vessel types, commodity properties, dock properties, and intermodal transportation availability to recommend alternative docks when the intended dock is inaccessible.
Zika Africa Indoor Residual Spraying Project
Client: U.S. President’s Malaria Initiative/U.S. Agency for International Development
Abt developed a smartphone web app to use a smartphone camera to automate the counting of individual mosquito eggs for field-based mosquito surveillance efforts — an important process in addressing mosquito-borne illnesses such as malaria and Zika.
Accountable Health Communities Evaluation
Client: Centers for Medicare and Medicaid Services
Abt used NLP to analyze 4,711 open-ended survey responses for the Accountable Health Communities Evaluation Beneficiary Survey. The goal was to identify the 10 most discussed topics. The analysis, which took about 10 minutes, confirmed that respondents discussed food-related topics most commonly and that topics didn’t differ between intervention and control respondents.
Predicting College Graduation Rates
Client: Miami Dade College
Abt used a gradient boosting algorithm to predict students’ associate degree completion up to four years from first enrollment with up to 93% accuracy. A secondary analysis also determined a ranked list of variables that most influenced degree completion.
Healthcare and Human Services Delivery
Abt’s ready-to-deploy, data-driven operating model places a health equity-focused, Social Determinants of Health framework as central to identifying and mitigating the impact of long-standing structural barriers to health and economic well-being.
We’re using machine learning tools to help social service agencies allocate resources more equitably and efficiently. That enables them to fine-tune service delivery, leading to better outcomes for beneficiaries. Learn more.