Criminal investigators

Three parts of your work where AI is already doing real lifting, and what stays yours.

• Notebook + surveillance + informant net…1850–1920

  Notebook + surveillance + informant networks (Pinkerton investigative method)

  Allan Pinkerton formalized investigative method as systematically as anyone in the nineteenth century. His agents used tools that were entirely human: the ability to observe without being observed, to embed themselves in criminal networks, to interview witnesses systematically and document what they learned. The notebook, the investigator's primary instrument, recorded everything: names, dates, descriptions, movements. Physical surveillance, called 'shadowing,' was developed by Pinkerton into a professional discipline with its own rules: always maintain a fixed distance, never make eye contact, rotate operators to prevent recognition, use disguise when necessary. The rogues' gallery, a collection of photographs of known criminals, was Pinkerton's primary identification tool from the 1850s onward; Pinkerton accumulated the largest private collection of criminal photographs in the United States. Allan Pinkerton's 1884 book 'The Spy and the Thief' formalized many of these methods for the first time in print. The tools were entirely analog and entirely reliant on human memory, human judgment, and human presence.

  Effect on the work

  The Pinkerton method required highly skilled individual investigators capable of independent judgment over extended assignments. A single operative might spend months infiltrating a criminal gang before any arrest was attempted. This craft model of investigation shaped how detectives were trained, and compensated, for a century.

• FBI fingerprint identification division…1924–1980

  FBI fingerprint identification division + forensic laboratory (scientific investigation era)

  J. Edgar Hoover's most durable contributions to American law enforcement were technical rather than tactical. In 1924, the year he became FBI director at age 29, Hoover consolidated the fingerprint identification files of the Bureau of Investigation and the National Bureau of Criminal Identification into a single collection that would become the world's largest. By 1946 the FBI's identification division held about a hundred million fingerprint records; by 1971 it had become the Identification Division's computerized file. The FBI Laboratory, established in 1932, brought systematic forensic science (questioned-document examination, handwriting analysis, ballistics, trace-evidence chemistry) into criminal investigation for the first time as a government service available to any law enforcement agency in the country. The lab's work made it possible to build physical-evidence cases that did not depend entirely on confessions or eyewitness testimony. The era from 1924 to 1980 was defined by the principle that science could produce evidence where human memory and testimony were unreliable. It was also the era in which Hoover used the same scientific apparatus (covert surveillance, informant networks, COINTELPRO) to monitor and disrupt civil rights leaders including Martin Luther King Jr., a fact that remains the most serious institutional corruption in American law enforcement history.

  Effect on the work

  The scientific investigation model created the detective as a specialist in forensic evidence collection and analysis, not just in witness management. Detectives needed to understand fingerprint processing, crime-scene documentation, chain-of-custody requirements, and laboratory submission procedures: technical knowledge that distinguished investigation from simple policing.

• Automated fingerprint identification1980–1999

  Automated fingerprint identification (AFIS) + CODIS DNA database (first digital forensic era)

  The shift from manual to computerized fingerprint matching is one of the most consequential technological changes in the history of criminal investigation. Manual fingerprint comparison, comparing a latent print found at a crime scene against a suspect's ink-rolled card, was skilled but slow work: a trained examiner could compare perhaps many prints per hour, and a criminal with no prior arrest record had no prints on file to compare against. Automated Fingerprint Identification Systems (AFIS) were first deployed by individual state agencies in the early-to-mid 1980s; the San Francisco Police Department had an AFIS by 1984, the California DOJ by 1985. The FBI's national IAFIS (Integrated Automated Fingerprint Identification System) went live in 1999, holding 70 million fingerprint records. Response time for a criminal fingerprint submission: many minutes, compared to weeks under the manual system. Fingerprints that had sat in evidence boxes unsolved for decades could now be searched against millions of records. CODIS, the Combined DNA Index System, began as an FBI pilot program in 1990 with several state and local laboratories and received Congressional authorization through the DNA Identification Act of 1994. The National DNA Index System went live in October 1998. By September 2020 it held more than 14 million offender profiles and had aided in hundreds of thousands of investigations, producing 530,000 hits. DNA evidence transformed categories of crime (rape, murder, sexual assault) where conviction had historically depended on eyewitness testimony alone. The exoneration movement that followed, as post-conviction DNA testing cleared hundreds of wrongfully convicted defendants, was an unintended consequence of the same technology.

  Effect on the work

  AFIS reduced the labor required to clear a fingerprint match from weeks to minutes, but it did not reduce the number of investigators: it expanded the investigable universe. Crimes that previously went cold because no prints matched went to detectives who could now run rapid database checks. CODIS similarly expanded the case-closing capacity of investigative units, especially for cold cases.

• Digital forensics + mobile device extra…1999–2017

  Digital forensics + mobile device extraction + surveillance camera proliferation

  The spread of personal computing and mobile phones in the 2000s created a new category of evidence that did not exist before: digital traces. Email, browser history, GPS location data, cell-site records, financial transactions, social media messages: every serious crime in the digital era left a digital footprint that investigators had to learn to read. The first specialized digital forensic units in US police departments date to the mid-1990s; by 2010 they were standard in any agency handling serious felonies. Cellebrite, an Israeli company founded in 1999, emerged as the dominant provider of mobile forensic technology with its UFED (Universal Forensic Extraction Device), launched in 2007. The UFED could extract call records, text messages, photos, deleted files, and app data from smartphones: material that increasingly constituted the most probative evidence in homicide, trafficking, and drug cases. By the 2010s, Cellebrite had become effectively the standard tool for mobile extraction in US law enforcement. The ability to recover deleted WhatsApp messages and encrypted communications transformed what investigators could prove in court. Simultaneously, the proliferation of commercial surveillance cameras (in stores, gas stations, ATMs, traffic intersections) meant that investigators could often place suspects at or near crime scenes through camera footage that previously did not exist. The investigator's task shifted from reconstructing events entirely through witness accounts to assembling camera footage, cell-site records, and financial data into a timeline that either corroborated or contradicted witness testimony.

  Effect on the work

  Digital forensics created a new investigative specialty requiring technical training in computer science, network analysis, and mobile operating systems: skills not found in traditional detective training. Agencies struggled to hire and retain forensic examiners who could command private-sector salaries. The result was a persistent backlog of unexamined digital evidence, particularly in rural and smaller jurisdictions.

• Clearview AI facial recognition + genet…2017–2023

  Clearview AI facial recognition + genetic genealogy (identity at scale)

  The years 2017-2023 brought two technologies that, together, made identity virtually non-anonymous in a criminal investigation context. Clearview AI, founded in 2017 and exposed in a January 2020 New York Times investigation, provided law enforcement agencies with a facial recognition tool that could match a face against a database of billions of scraped social-media images, far larger than any government biometric database. By 2023, US police had run approximately 2 million annual searches across hundreds of agencies. A detective who had a single surveillance camera frame of an unidentified perpetrator could now run that image and receive a candidate match within seconds. The technology's error rates, particularly against darker-skinned individuals, produced documented wrongful arrests, and several cities banned police use entirely. Genetic genealogy, the technique of searching consumer DNA databases like GEDmatch to identify suspects by building family trees from partial DNA matches, was demonstrated at scale on April 24, 2018, when Sacramento County Sheriff's deputies arrested Joseph James DeAngelo, the Golden State Killer, after investigators used DNA uploaded to GEDmatch to identify relatives, narrowed the pool to DeAngelo through genealogical research, and confirmed the match from a covertly collected DNA sample. The case had been cold for over decades; it was solved in months once genetic genealogy was applied. By December 2023, the technique had solved hundreds of criminal cases and identified hundreds of perpetrators. Parabon NanoLabs generated investigative leads in hundreds of cases by January 2021. Forensic genetic genealogy transformed cold-case investigation more profoundly than any technology since DNA itself.

  Effect on the work

  Facial recognition and genetic genealogy compressed the identity-resolution phase of an investigation, the hardest part of many cases, from months or years of traditional legwork to days or weeks of database-assisted analysis. They did not eliminate the investigator; they made the investigator dramatically more effective against cases that previously had no viable suspect.

• AI case analysis + LLM report-writing +…2023–present

  AI case analysis + LLM report-writing + predictive lead scoring (generative AI era)

  The emergence of large language models in 2023-2024 brought AI into detective work in ways that mirror its impact on patrol: primarily as an administrative and analytic burden-reducer rather than a replacement for investigative judgment. Investigators spend a substantial share of their time on documentation (case reports, affidavits, search warrant applications, grand jury submissions) that is linguistically demanding but structurally repetitive. LLM-based drafting tools reduce this burden materially, as has been demonstrated by Axon's Draft One product (April 2024) for the related patrol-officer population. Vendors including Palantir Technologies have built case-management platforms (Gotham, Apollo) that use AI to surface connections across large evidence sets, linking phone records, financial transactions, surveillance footage, and database queries into a visual map of a criminal network that would previously require months of analyst work. Flock Safety's ALPR network (20 billion vehicle scans per month as of 2025) represents a passive surveillance infrastructure that feeds directly into investigative queries: a detective investigating a robbery can query which vehicles were in a three-block radius during a one-hour window and receive a list in seconds. License plate reader networks have effectively created a nationwide vehicle location history database without any legislative authorization for that purpose. The investigative utility is real; the civil liberties implications remain largely unresolved. The net effect of the generative AI era on this occupation is augmentation: faster evidence assembly, better pattern-recognition across large datasets, reduced administrative burden, but the terminal acts of criminal investigation (strategic judgment about which lead to pursue, building witness trust through an interview, presenting a case to a jury) remain irreducibly human.

  Effect on the work

  BLS projects essentially flat employment for detectives through 2034, consistent with a technology-augmentation scenario: the same number of investigators doing substantially more investigative work per head as AI tools compress the evidence-processing and documentation phases.