J. Edgar Hoover was 24 years old when he was appointed head of the Justice Department's General Intelligence Division in 1919, tasked with identifying and cataloguing radicals, anarchists, and political dissidents in the United States. He had trained as a librarian before studying law, and he brought that training to bear on a problem that was fundamentally one of information architecture: how do you track thousands of people, map their relationships, anticipate their behavior, and make that information retrievable at scale? His answer was an index-card system — ultimately containing more than 450,000 names — organized to allow cross-referencing of individuals, organizations, publications, and geographic locations. It was not metaphorically an algorithm. It was, structurally, exactly what we now call one: a set of rules for sorting and retrieving information that produces outputs (targets, risk assessments, associations) from inputs (names, affiliations, observed behaviors).
The mechanics of the system are worth understanding because they illuminate how information infrastructure shapes outcomes independent of the intent of any individual operator. Each card contained not just a name but a network — who a person associated with, what meetings they attended, what publications they read. Cross-reference cards connected these nodes. A query on one name would surface not just that individual's file but a map of their relationships, each relationship a potential target in turn. This is association mapping, and it is the foundational logic of every modern social graph, threat-scoring system, and predictive policing algorithm. The technology changed; the logic did not. The political presumption embedded in the system — that association with disfavored ideas or people is itself evidence of risk — also did not change.
The Index Card as Infrastructure
Modern algorithmic surveillance differs from Hoover's card index in scale, speed, and the granularity of data available. It does not differ in its fundamental architecture. Data aggregation platforms compile records of purchases, movements, communications, and associations — cross-referenced in exactly the way Hoover's clerks cross-referenced names. Risk-scoring systems used by law enforcement and immigration authorities assign numerical values to behavioral patterns in exactly the way Hoover's system assigned political classifications. The language of objectivity has changed — "data-driven" now substitutes for "scientifically organized" — but the underlying claim is the same: that a system of records can produce reliable predictions about human behavior and intent. The historical record on that claim is not encouraging.
What distinguishes an information system from an accountability system is the question of who answers for error. Hoover's index generated thousands of wrongful deportations during the Palmer Raids, destroyed careers, and enabled decades of illegal surveillance of civil rights leaders, labor organizers, journalists, and politicians. None of the architects of those systems faced meaningful accountability. The systems themselves were treated as neutral infrastructure — the problem, when problems were acknowledged at all, was attributed to individual bad actors rather than to the design of the apparatus. We are repeating this pattern. When a predictive policing system flags a neighborhood for over-patrol based on historical arrest data that reflects historical over-policing, the error is not random. It is structural. And structural errors require structural accountability, not individual correction.
The Central Valley has its own history on the receiving end of federal surveillance and classification systems. Agricultural labor organizing in the 1930s brought FBI surveillance to the Valley's fields and labor camps. Japanese American families in Stockton, Fresno, and across the region were incarcerated based in part on records compiled through exactly the kind of cross-referenced association mapping Hoover had pioneered. Undocumented residents today are subject to data-sharing arrangements between local law enforcement and federal immigration authorities that function as a contemporary version of the same architecture — a distributed system of records that converts presence into risk and association into evidence. The Valley is not peripheral to this history. It is central to it.
"Every algorithm is a filing system. Every filing system is a theory of who matters and who doesn't."
Why does this history matter now? Because the policy debates that animate current arguments about algorithmic accountability — who gets to see what data, what counts as probable cause, what associations can be used as evidence of intent — are not new debates. They are the same debates that followed Hoover's excesses, that produced the Church Committee reforms, that created the Privacy Act and the Freedom of Information Act. Those reforms were real but incomplete, and they have been systematically eroded. Understanding that the current moment is an iteration of a long argument, not a new problem created by new technology, is the prerequisite for understanding what actual accountability would require. The filing system is not neutral. It never was.