Notes on Public Interest Technology for Social Workers
I usually introduce myself as a computational social scientist and a public interest technologist. Neither term is widely used in social work, and both are recent: the article most often cited as the founding statement of computational social science appeared in 2009, and public interest technology spread mostly after 2016. This post explains what each term means, how I use both in my own work, and how public interest technology differs from technosolutionism. Please read it as one person’s account and not as a definition of either field.
Computational social science
Computational social science, as David Lazer and colleagues described it in Science in 2009, studies social life through the digital traces people leave and the computational methods needed to analyze them at scale. Most of my methods come from this field: natural language processing for what people write in online communities and public documents, and geospatial analysis for where services are and are not. In International Migration Review my colleagues and I introduced natural language processing to migration researchers through online migration forums, and in the Journal of the Society for Social Work and Research we measured the racial, ethnic, and linguistic diversity of the clinical social work workforce and how far people live from it. I learned much of it through the Summer Institutes in Computational Social Science, which Chris Bail and Matt Salganik started in 2017, first as a participant and later as an organizer of SICSS-NYU Silver. I now co-organize the data science and AI special interest group of the Society for Social Work and Research.
These methods suit my research for practical reasons. Immigrants, refugees, and small language groups are often too few in probability surveys to analyze on their own, or are excluded altogether when a survey is offered only in English, so much of what I want to know about them has to come from other records. Whether mental health care is offered in a given language, and how far it is from where people live, is recorded in provider directories and on organization websites, and geospatial methods link those records to Census data on the surrounding population. How people describe their own needs, often in languages other than English, appears in online forums and public documents, and natural language processing makes it possible to read that text in large volumes.
Public interest technology
The term spread around 2016, mostly through foundations. The Ford Foundation describes public interest technology as a field of people who “work to ensure technology is created and used responsibly,” who “call out where technology can improve for the public good, and sometimes question whether certain technologies should be created at all.” Ford counts designers, lawyers, artists, activists, and members of affected communities as technologists, and it says the field was “modeled after the framework of public interest law.”
The Public Interest Technology University Network (PIT-UN), which Ford, the Hewlett Foundation, and New America launched in 2019, uses a shorter academic version: “the study and application of technology expertise to advance the public interest, generate public benefits, and promote the public good.” The University of Michigan was one of the 21 charter members, with the Ford School’s Science, Technology, and Public Policy program as its hub. Tara Dawson McGuinness and Hana Schank, in Power to the Public (2021), give a more practical definition, “the application of design, data, and delivery to advance the public interest and promote the public good,” as quoted in a 2025 PIT-UN report. Bruce Schneier, who has written about the field for years, adds that “you do not need a computer-science degree” to do this work; the people he has in mind come from technology, policy, and law, and work in government, nonprofits, research institutions, companies, and the press.
PIT-UN’s version is about applying expertise. Ford’s version also includes deciding whether a technology should exist at all, and that second part is the one I lean on most.
The field has also changed quickly. In March 2025 the General Services Administration eliminated 18F, its in-house digital services team, after an internal email described the office as “non-critical,” and the January 2025 executive order that created DOGE “publicly renamed” the U.S. Digital Service as the United States DOGE Service. PIT-UN is now “a fiscally sponsored project of the New Venture Fund.” For anyone hoping to do this work inside the federal government, the route seems less certain than it did a few years ago.
Background
I grew up in two villages in Korea, and the computer and the internet were how I first saw a wider world. I built HTML websites in an after-school program in elementary school, and in 2018 I started learning Python, mostly for natural language processing. I came to language access through work with immigrants and refugees in my twenties. Ingrid Piller, a sociolinguist, describes language as a “key gatekeeping mechanism” for education, employment, health care, and welfare, and that matches what I saw: the barrier was usually an institution that had not adapted to the languages its clients spoke.
The tools I have built come out of that. The Immigrant Support Atlas maps immigrant-serving organizations and access to mental health care across the United States. langaccess is a Python package that checks how a website handles languages other than English. With Migration to Asia Peace, a refugee organization in Korea where I serve on the board, I work on MAP Refugee News, a multilingual information service for refugees, and the full list is on my Data & Tools page.
Publics and indirect stakeholders
Computational social science describes how the work is done, and public interest technology describes whom it is for, how I judge it, and who has a say in it. I use the first when I am talking about how a study was done and the second when I am talking about why a tool exists.
John Dewey’s definition of a public helps me with the second. In The Public and Its Problems (1927), the public “consists of all those who are affected by the indirect consequences of transactions to such an extent that it is deemed necessary to have those consequences systematically cared for.” A person with limited English proficiency who receives an untranslated benefits notice was not party to the contract between the agency and its software vendor, and still lives with the result. Value sensitive design makes the same point with its category of indirect stakeholders, people who “never or rarely interact with the system” but “are nevertheless affected by the system.”
Social work and technology
Social work seems to me a natural home for this kind of work. The NASW Code of Ethics names the profession’s “dual focus on individual well-being in a social context and the well-being of society,” and person-in-environment thinking applies to a website or an algorithm as much as to a neighborhood. The 2017 Standards for Technology in Social Work Practice ask social workers to advocate for access to technology for people with “limited proficiency in English,” among others. Their interpretation of Standard 2.20 is close to my own work: when people have to use text-based application forms to obtain services or benefits, social workers “should consider options to help people who prefer to use a language other than English.”
Social work scholars have been making this argument for a decade. A 2015 working paper for the Grand Challenges for Social Work quotes Lauri Goldkind and Lea Wolf: social work is “uniquely positioned and ethically obligated” to ensure that technological change does not reproduce or widen existing inequalities. Goldkind later led a PIT-UN project on competencies for public interest technologists that draws directly on the competency model of social work education. Desmond Upton Patton, who has worked with computer scientists on violence prevention, described what that collaboration required: “we MUST have community support AND buy-in.” In September 2026, CSWE created a national advisory board on AI and technology.
Technosolutionism
Public interest technology is easy to confuse with technosolutionism, the habit Evgeny Morozov criticized of recasting messy social situations as problems that software can solve. Mark Latonero made a similar point about corporate “AI for good” programs, writing that the fanfare “smacks of tech solutionism.”
The difference, as I understand it, is where the work starts. Technosolutionism starts from the tool and looks for a problem. Public interest technology starts from the people affected and treats building nothing as one of the options. Langdon Winner’s “Do Artifacts Have Politics?” (1980) is still the clearest statement of why: if we judge technology only by “tools and uses,” he writes, we “will be blinded to much” that matters. His example of the mechanical tomato harvester involves no plot, only “deeply entrenched patterns that bear the unmistakable stamp of political and economic power.”
In practice this means asking, before building, whether a human interpreter or a better staffed office would serve people better than a tool, and keeping the people affected in the decision. The Design Justice Network puts it as its second principle: “We center the voices of those who are directly impacted by the outcomes of the design process.” For the translation work with MAP, I am developing a workflow in which a reviewer from each language community approves high-stakes translations before they are published.
Some of the best known public interest technology in Michigan involves no AI at all. Working with the design studio Civilla, the Michigan Department of Health and Human Services rewrote its combined application for food, cash, health care, and child care assistance as 18 pages, “down from 42 pages” in what the department called the longest application in the country, and clients in the pilot counties spent about 20 minutes on it instead of 45. Virginia Eubanks, who studied automated systems in public assistance, describes the cost of the alternative: “Building these systems well is incredibly hard and incredibly resource intensive, and building them poorly is only cheaper and faster at first.”
Big Tech and technocapitalism
Luis Suárez-Villa’s term technocapitalism describes “a new version of capitalism, grounded in technology and science,” in which creativity and research become commodities owned by corporations. AI follows the same pattern, and Meredith Whittaker argues in “The Steep Cost of Capture” that recent advances in AI were “primarily the product of significantly concentrated data and compute resources that reside in the hands of a few large tech corporations.” The share of new AI PhDs going to industry rose from 21% in 2004 to almost 70% in 2020 (Ahmed, Wahed, and Thompson, 2023), and nearly 90% of notable AI models released in 2024 came from industry (Stanford AI Index 2025). The 2026 AI Index adds one counterpoint: new AI PhDs in the United States and Canada grew 22% from 2022 to 2024, and the graduates behind that increase “took jobs in academia, not in industry.”
I do not think a public interest technologist can stand outside this. I use commercial language models and cloud credits from technology companies, and the public alternatives depend on them too: the federal NAIRR pilot provides researchers with compute from NVIDIA, Microsoft, Amazon, Google, OpenAI, and Anthropic. I try instead to make the dependence visible and to put the parts that matter in public hands: open code, data deposited in public archives, and tools that a community organization can run without me. Mozilla’s case for public AI states the reason plainly: “We can’t just rely on a few companies to build everything our society needs from AI.”
Refusal and erosion
Audre Lorde’s warning that “the master’s tools will never dismantle the master’s house” states the case for refusal. Erik Olin Wright, in his late work on anticapitalist strategy, distinguished smashing, taming, resisting, and escaping capitalism, and argued for a combination he called eroding capitalism: building more democratic and participatory relations “in the spaces and cracks” of the existing system until they displace it. The closest name I have found for my own position is Phil Agre’s “critical technical practice,” which he said would “require a split identity,” with “one foot planted in the craft work of design and the other foot planted in the reflexive work of critique.” I criticize the concentration of AI in a few companies and also use their models to build public interest tools for immigrants, and I do not think that contradiction resolves. Iris Marion Young’s social connection model gives me a way to think about it: “all agents who contribute by their actions to the structural processes that produce injustice have responsibilities to work to remedy these injustices.”
Machine translation
Machine translation shows why the judgment matters more than the tool. Under the 2024 rule implementing Section 1557 of the Affordable Care Act, a health program that uses machine translation for text critical to a person’s rights or access “must be reviewed by a qualified human translator.” In July 2025, after Executive Order 14224 made English the official language, the Department of Justice told federal agencies to “minimize non-essential multilingual services” and encouraged “responsible use of artificial intelligence and machine translation” to save costs. The department has also “temporarily suspended the operations of lep.gov,” the federal site that held language access guidance. The same technology appears in both documents, once as something to check and once as a reason to spend less on language services.
A public interest technologist, as I use the term, is someone who can tell those two uses apart and build the version that adds access, including deciding that some tools should not be built.
