Residential Burglary Expert System
REBES, short for Residential Burglary Expert System (also known as Baltimore County Burglary System), was the first offender profiling software used for local crime investigation in the United States. It was built in the late 1980s for the Baltimore County Police Department by the Jefferson Institute for Justice Studies to help detectives with residential burglary cases. The system was tested and then discontinued in the early 1990s after its experimental period.
In crime work, experts distinguish between major crimes (like murder) and volume crimes (like burglary). In the 1980s, U.S. efforts to apply artificial intelligence to crime fighting mainly focused on major crimes, while interest in burglary-focused systems began in the United Kingdom. After a 1985 Exeter pilot and a Devon and Cornwall burglary system, the United States funded grants in 1986 to test such expert systems. With these grants, the Jefferson Institute adapted the Devon and Cornwall approach for Baltimore County.
Residential burglary is a volume crime with many offenses, often by serial offenders, and a relatively low rate of solving cases. An experienced burglary detective who has worked many cases can solve new ones by drawing on knowledge from past investigations. It was believed that retirements of seasoned officers could create a “brain drain” that computers might help prevent.
REBES was designed to capture human expertise. It worked by taking crime-scene data and comparing it to patterns from known offenders stored in the system, then returning a likely offender with a probability. If there were several possibilities, the system produced a ranked list. Development took about 14 months, and the core knowledge came from Baltimore County police detectives.
The software was built as a rule-based system, using if/then rules and heuristics to simulate human reasoning. Development steps included defining rules, designing data-collection forms, and building a database. Officers identified burglary categories, turned them into rules, and examined them for accuracy. After discarding weak rules, the remaining ones formed REBES’s rule base.
To populate the system, data from roughly 3,000 solved and 1,700 unsolved cases were entered, covering about 675 suspected or known burglars. In April 1988 REBES was ready for testing. During 1988–1990, it received about 100 queries per month. Investigators often already had a suspect and used REBES as a backup to generate a short list of potential offenders.
REBES was shared with police departments in Rochester (New York), Tucson (Arizona), Charlotte (North Carolina), and Tampa (Florida). The developers claimed that Baltimore County’s burglary detection rate rose by about 2.5% thanks to REBES, but the system was ultimately discontinued.
Two major criticisms emerged. First, concerns about the system itself, and second, broader criticisms of early expert systems in criminal investigation. While supporters praised REBES as guidance for younger officers during its trial, critics cited factors like high turnover among users, new users disagreeing with the stored knowledge, changing knowledge, and a lack of integration with existing computer systems.
In particular, German scholar Jo Reichertz argued against treating burglary as easily solvable by a fixed set of rules. He questioned the forensic approach as overly simplistic and noted that early expert systems were amateur prototypes that could only repackage entered data. They did not generate new ideas or hypotheses. Reichertz and others argued that real investigative reasoning relies on abductive thinking, not just inductive rules, and that early systems’ limitations helped push law enforcement AI toward primarily database-driven applications rather than standalone expert systems.
This page was last edited on 3 February 2026, at 02:30 (CET).