Autonomous vehicles could drive us crazy

A decade ago, the Society of Automotive Engineers (now SAE International) defined five levels of autonomy in autonomous vehicles — or six, if you count Level 0, designating traditional non-autonomous vehicles entirely under human control. Levels 1 and 2 involve driver-assistance features like blind-spot warnings and autonomous emergency braking systems now familiar on many newer vehicles. Level 3 is conditional automation, where the vehicle assumes full driving function but where the driver must be prepared to resume control upon notification; here the driver is essentially a passenger and co-pilot. A Level 4 vehicle is highly automated and is capable of performing all driving functions; at Level 5, the vehicle is fully automated and requires no driver at all. 

A key to developing autonomous vehicles is Light Detection and Ranging, or LiDAR, a technology originally developed by NASA to track satellites. LiDAR systems use pulsed lasers to map a three-dimensional model of an environment quickly and accurately. When these millions of pulses-per-second strike objects, the LiDAR sensors calculate distances and “see” what is around the vehicle — buildings, trees, other vehicles, pedestrians, even lane markers. With the proper algorithm, the AI “brain” of the automated vehicle can make effective operating decisions. 

Billions have been invested in LiDAR in recent years, but not everyone is sold on it. “We do not envisage using LIDAR,” said Tesla CEO Elon Musk, who announced earlier this year that Tesla vehicles would instead use radar to supplement their camera-based Autopilot system, which provides Level 3 automation. “LiDAR does not penetrate occlusions, it does not penetrate rain, dust, or snow, whereas radar does. … You can’t look at things in front of the car in front of you.”

True Level 4 and 5 vehicles — the vision for LiDAR development — are not yet available to the consumer market. “Realizing L4/L5 capability has proven to be more challenging than originally envisioned due to technical, safety, regulatory and cost considerations,” writes Sabbir Rangwala in Forbes. LiDAR more heavily targets the commercial market (trucking, logistics, ridesharing) for Level 4 and 5 vehicles while primarily developing value-added autonomy features at Level 2 and 3 for consumer vehicles. But this could change with time.

There is an old simplistic thought experiment in moral decision-making called the “trolley problem.” One of its modern variants envisions a fully automated vehicle heading full speed toward a crossing where a group of pedestrians suddenly emerges from the side. The only evasive action would send the vehicle into a concrete barrier, killing or injuring its occupants. What AI algorithm would guide the vehicle’s decision? Would it quickly calculate the likely casualty count for each outcome, or would it be programmed to save the vehicle’s occupants at all costs, even if it means plowing through the pedestrians?

There is no simple moral solution to such a situation — but it might not be a question best driven by an algorithm.