LONDON. Tens of billions of dollars have been invested in autonomous vehicle startups promising to create the first fully autonomous car that doesn’t need a driver. However, experts in the field believe that these vehicles will always require a human inspector who can remotely take control of the vehicle in the event of a problem.

Despite the hope that the combination of computers and artificial intelligence will drastically reduce the number of accidents, the implementation of this theory remains extremely difficult. It is very difficult for a robotic car to drive more safely than a human because self-driving software cannot predict and respond to impending danger.

“I see no reason to remove the inspector. Our customers will have more peace of mind knowing that there is a person behind the automation software,” says Kyle Vocht, CEO of a subsidiary of General Motors. Like air traffic controllers, these inspectors can be thousands of miles away, recording the path of an unmanned vehicle using cameras. Inspectors will be able to take control if they think the driver is at risk.

Earlier this month, GM recalled and updated software for 80 self-driving cars after two people were injured in a crash involving one self-driving car. U.S. regulators assessed that the withdrawn software “incorrectly predicted” the path of a vehicle in the oncoming lane.

The idea that controllers will continue to operate ostensibly autonomous vehicles calls into question the reliability of this technology.

For some, the idea that controllers will still drive ostensibly autonomous cars casts doubt on the reliability of the technology. In 2018, GM filed for approval for a fully autonomous vehicle without a steering wheel, brakes or accelerator. This vehicle, known as the Cruise Origin, is expected to roll off the assembly line in the spring of 2023. In 2019, Tesla CEO Elon Musk promised that by 2022 his company would build “a million taxi robots.” In a June interview, Musk acknowledged that building self-driving cars is “much more difficult than I expected.”

Many self-managing startups today are using remote inspectors. These people, although an additional financial burden, help to “manage” marginal risks such as interruption of road markings due to construction work or unpredictable sudden movements of conventional vehicles or pedestrians while driving an autonomous vehicle.

Autonomous driving systems are not as efficient as humans because the prediction algorithms they use do not have the capacity of the human brain to make small decisions. When a person sees a ball crossing the road, for example, they assume that the child might be running after the ball and apply the brakes a critical split second before the AI ​​software.

Experts believe that autonomous vehicles are off-road ready until their top speed exceeds 30 km/h.