The US Patent Office recently published a patent by Ford for an autonomous police vehicle that could be programmed with “machine learning tools (e.g., deep neural networks) to find good hiding spots to catch violators of traffic laws”. First spotted by Motor 1, the patent – which represents more of a moonshot project than a pending invention – would nonetheless really round out the most dystopian visions of our future.
Exceeding the capabilities of current policing robots by security companies such as Knightscope, the autonomous fleet would have immense surveillance capabilities. The patent filed in 2016 references cameras, road sensors, licence plate readers, touch sensitive panels, speakers, LIDAR, ultrasound sensors and microphones, satellite connectivity, and radar detectors to record the speed of other vehicles (ominously referred to as a “laser gun”).
Further, the patent references machine learning and neural networks throughout. AI is both rapidly altering law enforcement and prompting alarms from privacy advocates concerned about a dawning surveillance state. Ford imagines the robocar would connect to “a locally stored record of drivers” or even larger government databases to verify drivers’ licenses. With these powerful recording and storage opportunities, even fleeting interactions with the car could potentially land the public in a database somewhere.
The filing includes these diagrams of how the car might operate:
This first diagram is an overview of the many things the robocar could do in varying scenarios. You’ll note the tree at 180. According to the patent, the robocar “may, based on machine learning through deep neural network(s), find a spot behind an object 180 (shown as a tree in FIG. 1) and park at that spot behind object 180 so as to be inconspicuous”. Speed traps such as this are an old police trick, but training AI to hide from humans is all kinds of disturbing.
Ford imagines the vehicles could be attached to a local network of surveillance cameras that will send signals to robocars when they record traffic violations. As shown in the patent image, the cams could be attached to stoplights and stop signs themselves, triggering as soon as a driver, say, runs a red light or changes lanes without signalling. A network of cameras designed to catch speeders, yet the AI still needs to hide? Weird.
Police already use artificial intelligence to send squad cars to places they expect more crime, a practice called called “hot spot policing”. It wouldn’t be a stretch to imagine departments sending robocars to areas where there are more traffic violations, but there’s a serious risk for bias here. After the Ferguson, Missouri riots in 2015, a DOJ report found that local police over-patrolled black neighbourhoods and stopped black drivers significantly more often that white drivers. If autonomous police vehicles relied on traffic violations recorded by area, how do we know they wouldn’t repeat these biases?
The second diagram in the patent shows a hypothetical situation where the robocar has spotted someone speeding. The car would use “wireless communication” to contact the driver of the speeding vehicle and pull them over. From there, the robocar would establish the identity of the driver. The patent isn’t clear on how exactly this would work, only saying that valid responses from the pedestrian car may include “an image of a driver’s licence of a human driver of the first vehicle… verifying the authenticity of the driver’s licence”.
Perhaps the most troubling aspect of Ford’s patent is that it states its robocar would be used for “routine police tasks”. Of course, there’s nothing routine about being hunted by an robot police car that learns to hide in foliage like special forces. Right now, officers are using drones to keep tabs on drivers, but this represents the ultimate end of smart policing: A driverless surveillance bot that can’t be harmed, reasoned with, or stopped.g