Jeff Brantingham is close enough to face the controversial practice of "predictive surveillance". Over the past decade, the anthropology professor at the University of California-Los Angeles has adapted his research funded by the Pentagon to predict casualties on the battlefield. Iraq forecasting crime for US police departments, patenting their investigation and founding a for-profit company called PredPol, LLC.
PredPol quickly became one of the market leaders in the incipient field of crime prediction around 2012, but was also criticized by activists and civil libertarians who argued that the company provided a kind of "technological laundering" for a racially biased and ineffective policing. methods
Now, Brantingham is using military research funds for another technical and police collaboration with potentially damaging repercussions: use of machine learning, criminal data from the Los Angeles Police Department and an obsolete gang territory map to automate the classification of "related to gangs" crime
Being classified as a member of a gang or related to a gang crime may result in additional criminal charges, heavier prison sentences or inclusion in a civil gang mandate that restricts the movements of a person and the ability to associate with other people. In general, the application of the law determines the links between gangs through a highly subjective and individualized evaluation of criminal records, arrests, interviews and other intelligence information. In recent years, activists in California, Illinois and other states have rejected gang policing measures, such as databases and gang orders, and in the case of California, they got residents the right to review and appeal. your gang classification.
but in a document on "Partially generative neural networks for the classification of gang delinquency" presented in February in the first edition of Artificial Intelligence, Ethics, and the Society conference (AIES), Brantingham and his coauthors propose to automate this complex and subjective evaluation.
The document attempts to predict whether the crimes are gang-related using a neural network, a complex computer system modeled after a human brain that "learns" to classify or identify elements based on the ingestion of a training data set . The authors selected what they determined were the four most important characteristics (number of suspects, primary weapon used, type of premises where the crime was committed, and narrative description of the crime) to identify a gang-related crime since 2014- 16 data LAPD and cross references of criminal incidents with a 2009 LAPD map of the band territory to create a set of training data for your neural network.
The researchers tested the accuracy of the network's predictions by seeing how well it classified the crime data without a characteristic key: the crime description of the narrative text, the information that the police spend the most time collecting it. This is where the "partially generative" aspect enters in the title. In the absence of a written description, the neural network generates new text, in fact, an algorithmically written crime report based on the other three characteristics used in the training model. The generated text is not actually read by anyone, nor is it presumed to provide a meaningful narrative context that replaces a police report, but it becomes a mathematical vector and is incorporated into a final prediction of whether it is of a crime. related to gangs.
This document is the first to be published by a research team co-directed by Brantingham that studies "Theory of spatial and temporal games and real-time machine learning for adversary groups" in the University of the Center for Artificial Intelligence and Society of Southern California (CAIS). The mission of CAIS sets a goal of "[sharing] our ideas on how AI can be used to address the most difficult social problems".
Funding for the USC research team that includes the Brantingham project comes from the Minerva Initiative, a Pentagon research program aimed at improving the military's understanding of the social, political and behavioral drivers of the conflict. According to the Minerva Initiative website, funds are provided for projects that address "specific thematic areas determined by the Secretary of Defense." By e-mail, the CAIS co-founder and co-author of the Milind Tambe document said that Minerva's grant for this project is "approximately $ 1.2 million, which will be distributed over three years."
The website for the research team's efforts , including the gang classification document, opens with references to ISIS and Jabhat al-Nusra before switching to street gangs in Los Angeles, a merger echoing Brantingham's earlier work funded by the DOD he co-found PredPol PredPol has sold its services to the police everywhere from California to Georgia, as well as in the UK In 2015, PredPol lobbied unsuccessfully before the Arizona legislature to approve a $ 2 million appropriation act for use the company's forecasting technology to predict gang activity.
Reported by p Science the newspaper received criticism and some outrage at the AIES conference when it was presented by junior coauthor Hau Chan. At one point, Chan answered a question about the ethical implications of the investigation with the statement "I'm just an engineer," which prompted an assistant to run out of the place. However, reporting on the document and its consequences did not mention Brantingham's commercial connections with PredPol or the military funding of its past and present research.
When asked in a telephone interview if this investigation could inform future commercial efforts, Brantingham said: "This is a separate project, and that is how we are thinking." Noting that it took a decade for his previous military-funded research to become PredPol, Brantingham emphasized that the document reflected very preliminary work. "Our job is to conduct careful basic research and make sure we understand how and why things are the way they are, long before you think about using them in the field."
However, preliminary research could be good. the intentions of their authors are, the role and participation of Brantingham arouse astonishment among critics of the increasingly automated data-based policing technology.