Artificial Unintelligence

Artificial Unintelligence is not an argument against computers. Its target is what Meredith Broussard calls “technochauvinism”: the belief that once a problem can be digitized, a technical solution must be more objective, efficient, and advanced than human judgment. A former software developer who became a data journalist, Broussard explains the attraction of that belief from inside technical practice. She also shows how it permits institutions to skip elementary questions: does the software work, can the data represent the world, and should this activity be automated at all?

The title deliberately reverses “artificial intelligence.” Broussard distinguishes science-fictional general intelligence from the narrow systems that actually exist. These systems can perform impressively where rules and measurable goals are stable, but they do not understand context, meaning, or the people their outputs affect. Machine learning is not a mysterious mind; it is an arrangement of data, models, optimization targets, hardware, and organizational choices. Calling its outputs intelligent can conceal who defined success, which errors were accepted, and who bears the cost of failure.

The book is most persuasive when experiments in programming and reporting replace abstract alarm. Broussard rides in a self-driving car, rebuilds a Titanic survival prediction, investigates why students fail standardized tests, and attempts to map U.S. campaign finance with software. Again and again, the hard part is not an insufficiently powerful algorithm. The problem itself contains conflicting goals, incomplete evidence, institutional incentives, and values that require public argument. More elaborate code cannot by itself repair a bad exam, a distorted political system, or a dangerous road environment.

This critique is feminist because it refuses to treat “objectivity” as authority beyond interrogation. Records used in hiring, education, credit, healthcare, and public services come from unequal societies; missing data and classification choices influence who is visible, who is suspected, and who must prove themselves. Automation can also dismiss care, explanation, appeal, and the handling of exceptions as inefficient, even though women and marginalized workers disproportionately perform such labor. A statistically faster system may turn institutional responsibility into an interface decision that the person affected cannot correct.

Broussard’s contribution is not a demand to reject every technology but a recovery of judgment about appropriateness. Computers are excellent at repeated calculation, search, and large-scale organization. People remain responsible for choosing goals, understanding exceptions, explaining results, and answering challenges. When a public institution buys an AI system, accuracy is not enough. It should ask how the benchmark was built, which groups receive the errors, whether a nonautomated route exists, how a person can appeal, and what evidence would justify stopping the system. Technical literacy includes knowing when not to code.

Published before the current wave of generative AI, the book is both prescient and bounded. It clearly dismantles hype and the myth of general intelligence, but it does not cover at today’s scale the consent problems of training data, content labor, deepfakes, computational resources, or platform concentration. Its exploratory cases center the author’s own encounters, while race, gender, disability, and global supply chains receive less systematic treatment than in later data-justice work. It is best treated as a grammar for recognizing problems, not the final atlas of contemporary AI power.

For FemRes readers, “technochauvinism” can become a practical checklist. What social problem has been rewritten as a computable target? Whose data are absent? Who handles failure and exceptions? Is technological promise being used to avoid redistributing resources or accepting political responsibility? Read with Algorithms of Oppression, Automating Inequality, and Broussard’s later More Than a Glitch. The useful question is no longer whether machines will become intelligent like people, but who gets to decide what machines do and how those they judge can refuse.