Artificial intelligence did not slow down or signal due to a failure in the logic of object recognition.
The self-managed Uber car knocked a bicyclist in Arizona to death because of a software crash that determines which objects on the road should be ignored and what to look for. This information was reported by The Information sources close to the investigation of the incident.
According to the interlocutors of the publication, the system could not “not see” bicyclist Helen Herzberg before the collision, as she was directly in front of the lidar (the “eyes” of the car) and the front chambers. However, the car did not signal and did not even slow down.
The Information sources explained that the part of the software of the machine, responsible for the logic after recognition of objects, is to blame for everything. To prevent the self-controlled machine from “panicking” at every new obstacle, the system laid the possibility, in the manner of people, to ignore some insignificant things, for example, a motorcycle parked on the road.
The specialists manually set the sensitivity of this function. And it seems that the car that hit the bicyclist was determined to ignore even the objects that it was supposed to pay attention to. She recognized the girl, but did not find her presence on the road a reason for stopping.
March 19 homing machine Uber death was hit by a cyclist when she was crossing the road in the wrong place. There was a driver behind the wheel of the car, but he was diverted from control for a few seconds, just at the moment when the car could still stop.
After that, Uber brought the family’s dead condolences and agreed on a pre-trial settlement of the incident. All tests of self- controlled machines were suspended for an indefinite period, and American authorities joined the investigation.