David Ley Professor of Geography at the University of British Columbia, in an ingenious study of the perceptual geography of crime, asked residents of the Monroe area of Philadelphia to draw lines on a map tracing the routes they would take when walking between their home and different destinations on the map. When the routes drawn by respondents were overlaid, a clear pattern emerged, delineating the spatial segmentation of the area into safe and dangerous zones. Ley did not directly question residents about their reasons for avoiding certain areas, but an inspection of the areas revealed a number of potential cues to danger, such as gang graffiti and abandoned buildings. Ley's work demonstrates the existence of microzones of danger (blocks, street corners).
If certain locations are commonly perceived to be dangerous, what areas are perceived to be safe? One answer is repeatedly present in research of ‘fear of crime’, in data from the National Crime Survey. It is evident that respondents were much more likely to view their own neighborhoods as safer, while other neighborhoods in the metropolitan area as more dangerous..
While not much thought is spent on understanding subtleties of dangers - many people are still afraid in their local environment; the crimes that people fear outside their home (as stipulated in the Gallup/GSS question) are personal offenses. A sample survey of Seattle residents report that their everyday fear of becoming victims of different crimes are:
- Having something taken from you by force
- Having strangers loiter near you
- Being threatened with a knife, club, or gun
- Being beaten up by a stranger
- Being approached by people begging for money
Although people fear crime, a perception of the crime areas in their local environment is clearly mapped- it would be logical to infer that they would avoid the dangerous areas. Nevertheless, police data from NYPD shows that in the vicinity of the intersection of Lexington Avenue and 125th Street is one of the most dangerous areas in Manhattan, in spite of the high volume of pedestrians and nearly continuous police presence. Similar behaviour is shown in many other high crime yet densely visited areas around the city.
Recently, computer scientists have tried to tackle this question with software: Can machines understand intent of human crime and assist in preventing crime?
The critical component of understanding the engineering in this type of behavior is arguably what makes us human. Our tendency to take comfort in repeated patterns of behaviour. Just as the Monroe county citizens collectively created a similar pattern of routes. On the flip side, criminals also follow a pattern of behaviour in finding opportunity to execute their crimes.
In the quest to create a machine model capable of understanding criminal intent and actual criminal intelligence superimposed on each geographical area, researchers from bLine Analytics’s Intelligence Laboratory created an algorithm capable of flushing out illicit behaviour, predicting and preventing crime.
This type of research can be used in a range of technologies aimed at reducing crime, safeguarding citizens, protecting our assets in transit and not the least in protection against viral infections.
Our ability to accurately model physical movement and behaviour, identify patterns and highlight anomalies is crucial for building machines that robustly infer illicit intent. However, when inferring human behavior AI systems must first accurately model movement in physical space so that variation in GPS signaling and cadence are still understood by the AI as a pertinent route. GPS signals are seen to be all but consistent. Reporting location of entities in different points along the same route at different times as routine intelligence. However, bLine’s ability to consistently identify routes in real time, albeit signaling inconsistencies, is key for a learning algorithm which is consistently able to analyze behaviour and movement, make midcourse suggestion or even autonomously correct and replace routes in avoidance of likely loss says Iddo Gill, CEO of bLine Analytics.
Our team developed bHive AI, a new AI programming platform recently developed at bLine, to combine pattern analysis with unsupervised AI clustering. We combined pattern analysis with unsupervised clustering to provide an optimal wayto augment patterns with huge amounts of data, delivering accurate results for an extremely large dataset with infinite permutations describing real life events.
Container shipping is closely correlated with developments and changes in the world economy, manufacturing, and consumption. As the world is becoming more interconnected, shipping availability and costs may vary extremely with crisis situations becoming more prevalent. While it is impossible to avoid paying high market prices for cargo shipments, emphasis can be focused on ensuring assets get from point A to point B safely and intactly, becoming one of the most important objectives for companies dependent on supply chains for business viability.
AI is in the process of abandoning the standard model where a fixed, known objective is given to the machine, Instead, the machine knows that it doesn't know what we want, which means that research on how to infer goals and preferences from human behavior becomes a central topic in AI. says Stuart Russell, Professor of Engineering at the University of California at Berkeley.
Following these guidelines, bLine’s model accurately represents normal behaviour without a fixed objective or intent, moreover it can group entities into clusters of similar behaviour. An additional unique output is the ability to extrapolate the anomalies from a deep understanding of the normal which usually provides the most interesting insights.
How it works
While there’s been considerable interest in analyzing physical movement involving advanced algorithms, mainly in security related government organization, not a lot of commercial solutions exist today with sophisticated capabilities. Moreover, at bLine’s labs we found that for certain commercial applications bHive AI can shine with high accuracy results - especially in applications for supply chain optimization and asset movement tracking. In these applications, both typical ‘regular’ asset movement is provided combined with movement involving illicit intent or consequences. Many times crime can be initiated by the driver of a truck or the ship captain, either with the intent of performing the crime or simply by taking actions that ultimately result in a crime. Therefore, the challenge is to be able to distinguish between ordinary typical routes and the abnormal ones that expose the asset to high risk of a crime. In both cases, a tracking device reporting location based data is available and it is up to the bHive AI to actually contrast and identify normal safe movement to those with high risk.
To illustrate, let us imagine a route of transferring tobacco products from a port in Mexico to a distribution center just outside Mexico city. The normal route performed by most drivers includes a 5 hour drive with a single stop around 2 hours into the drive. What bHive AI identified was that some of the cargo theft was assisted by the drivers of the trucks that performed suspicious stops at certain locations to aid the criminals in their act.
The team, however, was particularly inspired by the ability of the model to not only identify crime after the fact but rather to be able to predict behaviour leading to a crime. What we noticed is that as the crime unfolds, the pattern is matched and the bHive AI risk score assigned to truck position in real-time is increased which can warrant intervention before the crime occurs.
The team’s pattern detection and clustering algorithm, called “Spatial Movement Anomaly Detection (SMAD)”, models the driver's route, and modeling all routes and sub-routes from all origins to all destinations. By detecting and clustering all potential routes and sub-routes in advance, normal and abnormal routes are mapped and identified.
“One of our early insights was that if you want to infer someone’s illicit intent, there are 2 main approaches to this challenge. First, we realized that illicit activity can be repeated probably due to communication and coordination performed by several individuals that lead to the crimes. Second, an additional approach showed that in certain cases, illicit activity was a unique behaviour that simply stands out as an anomaly. These insights assist us in exploring ways to resolve fundamental limitations of current AI systems, and create a practical and consistent method for analyzing real world behaviour” says Gill.
Our work builds conceptually on earlier User Entity Behaviour Analytics (UEBA), which is based on analyzing user behaviour for detecting cyber threats and stopping user and entity malicious online activity as applied to real world events.