⚡ Quick Summary
Published by TechTonic Justice, this November 2024 report is written by the organisation's founder and president, Kevin De Liban, and describes itself as "part report, part warning, part roadmap, and part vision." It sets out to explain and quantify, for a broad public audience, how AI-based decision-making reaches low-income people in the United States, and states that essentially all 92 million people below 200 percent of the federal poverty line have some basic aspect of their lives decided by AI.
Its central metric is AI exposure, an estimate of the number of low-income people subjected to AI that makes or recommends decisions about their lives. Exposure figures are given per domain: Medicaid 73 million, SNAP 42 million, housing 39.8 million, employment 32.4 million low-wage workers, private health insurance 30.6 million, Medicare Advantage 16.5 million, K-12 education 13.25 million children, Social Security disability benefits 10.6 million, language services 5 million, domestic violence 2 million, unemployment insurance 1.1 million and child welfare 72,000 children.
The report has three substantive parts and an appendix: an issue-by-issue account of AI uses and harms; an analysis of how AI-based decisions differ from human ones and of existing protections; and five recommendations covering advocacy capacity, policy infrastructure, new laws, enforcement and a positive vision for AI.
🧩 What’s Covered
The document contains three substantive parts, an appendix, a definitions section and a methods discussion.
- Executive Summary and Key Findings: states the scope of AI decision-making across public benefits, health insurance, housing, employment, K-12 education, language services, domestic violence and child welfare, with point-in-time exposure figures and the harms reported for each.
- Understanding How AI is Different: nine features distinguishing AI-based decisions from human-centred methods, including a "cloak of unwarranted rationality", expanded scale of risk, previously infeasible harmful functions, policymakers weaponising insufficient government capacity, less predictable lives, more data collection, persistence of past events, harder contestation of decisions, and defiance of accountability mechanisms.
- Defining Artificial Intelligence and AI Exposure: a broad definition spanning automated decision trees, process automation, data matching, surveillance such as facial recognition, machine learning, large language models and generative AI; the OECD definition of an AI system; what falls outside scope; and the methods behind the exposure metric.
- Introduction to TechTonic Justice: the author's background as a legal aid lawyer, the Arkansas litigation against an algorithm-based system for in-home care eligibility, and the organisation's mission to strengthen local justice movements.
- Report Overview: describes Part One, "How AI Decides the Lives of Low-Income People", Part Two, "Innovating Injustice", Part Three, "The Way Forward", and the Appendix, "Definitions and Methods".
- Actionable Recommendations: build capacity for integrated, multidimensional advocacy; create infrastructure connecting affected people to AI policy discourse; push for laws regulating AI use and protecting economic stability; invest in enforcement; and develop a positive vision for AI technology.
💡 Why it matters?
For advocates, legal aid lawyers, benefits advisers and oversight bodies, the report gathers in one place where automated decision-making reaches low-income people, how large the exposed population is estimated to be, and what harm is reported in each domain. It connects those harms to the limits of existing accountability routes — conversations with decision-makers, complaint processes, administrative hearings, affordable counsel and community organising — and argues that both enforcement of current law and new protections are needed. Its treatment of exposure as involuntary risk gives auditors and policymakers a population-level way to frame review, rather than a case-by-case one.
❓ What’s Missing
No new data was collected; the report relies on existing sources, some dating to 2015 or 2017, which the author says likely underestimates current AI use. Exposure figures are point-in-time counts that cannot be summed, and private-sector uses by landlords and employers are poorly documented. Several decisional domains are excluded — the criminal legal system, immigration, tax and child support enforcement, credit, health care treatment and voting — as are non-decisional harms such as environmental and labour impacts. The author also notes that available information may skew toward harmful uses. The scope is the United States; no other jurisdiction's law or standard is analysed.
👥 Best For
Best for legal aid and public benefits attorneys, community organisers and policy advocates building campaigns against automated decision-making; for journalists and researchers who need exposure figures and domain-by-domain examples of AI use; and for public-sector oversight staff, auditors and officials who want to see where AI touches benefit, housing, employment, education and child welfare decisions.
📄 Source Details
Inescapable AI: The Ways AI Decides How Low-Income People Work, Live, Learn, and Survive is written by Kevin De Liban, founder and president of TechTonic Justice, and dated November 2024. The supplied file has 27 pages and is in English; it prints no series or reference number. The closing page lists the organisation's website, techtonicjustice.org, and social media handles as contact details. The input was a text extraction of all 27 pages, covering the front matter, executive summary, key findings, definitions and methods, introduction and report overview; the detailed chapters of the three named parts do not appear in it.
