How your data is used by police, and where it goes wrong

"Cape Up" is Jonathan 's weekly podcast, talking with news and key people behind our culture. Subscribe to places to listen to Apple Podcasts, Stitchers, and other podcasts

"Thanks to this new information and this new feature, we are beginning to see the police change what they are doing.

In most of their history, the police station has followed the healthcare model we call the "rest / repair" model. Something went wrong, you should contact the doctor or police to fix it. For example, you are sick or your house is broken. After the event, the doctor or police will always arrive to manage the damage. But what happens when the police can get to the scene before the incident? Better yet, let's prevent the crime from happening first! So how does it work? Instead of resorting to predictors based on specific characteristics of a particular police station, Predictive Crime Analytics analyzes multiple overlapping data sets and analyzes factors such as crime and weather patterns, public transport situations, social media activity, To determine the correlation between. Other factors

One of our favorite interventions of this year against predictive vigilance and prejudice is the white collar crime application. By focusing on wealthy and strong people, this normally reveals the predictive policing data and overrides the invisible people. Mapping Financial Crime Data of FINRA to all blocks in the United States. Suddenly the downtown financial district in Boston was exposed as a hotbed for crime! Also this year, the AI ​​field slowly began to face its own bias problem. New approaches like AI4ALL directly address diversity and comprehensive issues. Founded by AI researchers Fei-Fei Li and Olga Russakovsky, the AI4ALL cooperates with leading universities to provide young women and people of different colors with opportunities to understand and design their AI systems .

Unfortunately, the AI ​​project data may be incorrect. The biggest failure is the lack of availability of data and failure to provide appropriate data to AI. Companies often fall into a trap of thinking that they have all the data, but experience shows that they are unavailable, inaccessible, storable, incomplete, or biased with experience. To overcome this defect and obtain data suitable for AI, we need strong vision and leadership support. There are still places where there is a possibility that it will not go well even after the project is launched. The lack of quality or accuracy of decision-making can result primarily from the lack of data selection and preparation, and the lack of effort to train AI. Do not use data sources that are incorrect, understand data dependencies, organize data, or have enough data to train. Biased data is a small field that has a major impact on the quality of AI's decision.