Are self-driving cars racist?
A self-driving car is a motor vehicle capable of driving on an open road without the intervention of a driver. The concept aims to develop and produce a vehicle that can really run on public roads in traffic without human intervention in all situations, eventually. It is a typical application of the field of mobile robotics and artificial intelligence in which many actors are engaged. Nevertheless, many technical, legal, psychological and legal issues remain unresolved.
Autonomous cars are becoming more numerous and will soon become mainstream cars. Before investing in your first autonomous vehicle, discover how they work: lidars, sonars, cameras, satellites, artificial intelligence and so on.
How do they work ?
Not all manufacturers use exactly the same technologies for the operation of their autonomous vehicles, but their technologies are similar. We will describe the operation of Nissan’s standalone Leaf project which uses the main sensors: radars, sonars, lidars.
The Leaf project uses laser scanners to detect objects and its environment. Front radars can see the road up to 200 meters in front of the vehicle and rear radars up to 70 meters.
This laser remote sensing technology is called LIDAR or light detection and ranging. This is the detection and estimation of distance by light / laser. Technically, it is about measuring distances by analyzing the properties of a light beam that is reflected towards its transmitter.
Sonars make it possible to validate the data provided by the laser scanners. The range of sonar is more limited than that of laser scanners and they are effective at very low speed, so they are especially useful for detecting nearby objects including parking.
Cameras are also used for a 360 vision. The cameras detect in particular the lines and strips of the road so that the car stays on it. The cameras serve to respect the rules of the road. They can detect red lights, signs and turn signals of other vehicles to adapt the behavior of the car.
All this is wonderful, but besides detecting red lights, other cars, and other obstacles; self-driving cars should also detect humans. To be able to implement an artificially intelligent system, it has to understand the human physics (anatomy, movement, and skin color).
According to this article, researchers from Georgia Institute of Technology found out that after a series of investigations on different AI systems and different targets and obstacles (including pedestrians of all sorts of races), the ones facing pedestrians of darker skins are the ones that perform the poorest.
I think before explaining why this happens; the ones among you who are not familiar with Artificial Intelligence should know that, in order to create a robot or an AI system, we need to train it.
Training a bot means we give it enough information to work with. Think of it like human beings. When we were kids we did not know about what roads are or what cars do, so we got trained and educated to cross the roads in a certain way but only where we grew up. This means that we trained our brains enough that we can cross any road in the world.
The same thing applies to bots, in order for a self-driving car to learn how NOT to hit a human being, it should be trained for all sorts of races, skin colors, heights and weights.
So to be able to explain this, one can only assume that this problem happens because AI systems are not trained enough with dark skin humans, that’s why they perform poorly.
Racism is not only a human behavior, but it can also lead to technology disasters if not treated well.