Developing self-driving cars has posed as AI’s supreme challenge. Presently, we acknowledge its substantial shortcomings, notwithstanding the colossal financial resources invested in striving to craft a commercially feasible vehicle. Furthermore, the recent discontinuation of a prominent provider of robotaxis in the United States, along with the implementation of stringent regulatory frameworks in the UK, hints at an even more distant possibility for developers to monetize this concept. The very viability of this notion hangs in the balance.
The inception of self-driving cars dates back to the mid-2000s with a challenge initiated by a US defense research agency, proposing a $1m reward for whoever could design a vehicle capable of completing a highly restricted journey in the desert. This morphed swiftly into a competitive race among various tech and automotive corporations (now identified as OEMs – original equipment manufacturers) to create what was assumed to be a revolutionary innovation: a vehicle capable of functioning in any environment without human intervention.
From the outset, the anticipation far exceeded the actual technological progress. In 2010, at the Shanghai Expo, General Motors showcased a video depicting a self-driving car rushing a pregnant woman to the hospital at breakneck speed, asserting its safety. The paramount pledge was enhanced safety, aiming to reduce the grim global annual road fatalities of 1.25 million, which was enticingly presented to the public by the proponents of autonomous vehicles.
However, this vision is now crumbling. Uber was the first casualty following an incident where one of its autonomous vehicles fatally struck Elaine Herzberg in Phoenix, Arizona. The vehicle was operating in autonomous mode, and its “supervisor” was allegedly engrossed in watching a TV show, neglecting to react when the vehicle collided with Herzberg, who had perplexed its sensors by crossing onto the highway with a bicycle carrying bags on its handlebars. Tragically, the AI system failed to comprehend this complex scenario.
Up until that point, Uber’s strategic model relied on the elimination of human drivers in the near future, envisioning a fleet of autonomous taxis. However, this plan was thwarted by Herzberg’s demise, leading Uber to withdraw from all autonomous taxi trials.
Now, Cruise, the entity acquired by General Motors to lead its foray into autonomous vehicles, is backing away swiftly as well. Another mishap, though non-fatal, caused severe injuries and prompted this retreat. In October, a woman was hit by a conventionally driven vehicle while crossing a street in San Francisco, subsequently colliding with a Cruise robotaxi without the latter stopping. According to its programming, the robotaxi veered to the right when faced with an unfamiliar scenario, running over the pedestrian. Although she survived, substantial compensation awaits her.
Following this incident, Cruise swiftly initiated damage control measures. Initially withholding pertinent details, the company later suspended its robotaxis in all US cities, and its CEO resigned. It was disclosed that the vehicles were not entirely autonomous, as human operators intervened every four to five miles. Mass layoffs ensued, casting uncertainty on the future of this initiative.
Tesla is also in defensive mode. The company has for a long time promoted its driver assistance software as “fully autonomous,” but in reality, it falls short of this claim. Drivers are required to remain vigilant and prepared to assume control, despite the car’s ability to function independently for most of the time, especially on highways. In the United States, where there have been multiple incidents involving Teslas in “full self-driving” mode, the automaker is currently facing several legal actions.
In the United Kingdom, Tesla will violate the recently enacted legislation, which prohibits businesses from deceiving the public about their vehicle’s capabilities. Tesla’s challenges have been worsened by disclosures from former employee Lukasz Krupski, who suggests that the autonomous driving features of Teslas pose a threat to public safety. Manufacturers will be mandated to define precisely which aspects of the vehicle—steering, braking, acceleration—have been automated. Tesla will need to revise its marketing strategy to abide by these regulations. Consequently, although the legislation was highlighted as a facilitator for the quicker adoption of self-driving vehicles, adhering to its stringent stipulations might present an insurmountable hurdle for their developers.
These incidents underscore the technological obstacles encountered in the transition to autonomous driving, as well as the fragility of the reasoning supporting the advancement of driverless cars. Every projection indicating the immediate realization of this technology within the next three or four years has proven overly optimistic. Ministers, including Chris Grayling in 2017, have been deceived, asserting the presence of self-driving cars on roads by 2021.
Technology firms have consistently underestimated the immense challenge of replicating, let alone surpassing, human driving abilities. This is where the technology has fallen short. Artificial intelligence is a more sophisticated term for the less glamorous concept of “machine learning”, and involves instructing the computer to interpret the complex road environment. The issue lies in the vast array of potential scenarios, from a frequently cited example of a stray camel on a city street to a mere obstacle on the road that might appear to be a harmless paper bag. Humans excel at promptly assessing these risks, but if a computer lacks knowledge about camels, it will struggle to react. The car’s computer was baffled by the plastic bags on Herzberg’s bike for a critical six seconds, as concluded in the subsequent analysis.
This illustrates the questionable priority of the government, led by tech entrepreneur Rishi Sunak, in introducing a bill concerning autonomous vehicles while neglecting reforms for the railway system or regulations for electric scooters, which operate in a legal gray area. The future may not necessarily revolve around driverless cars, and in the meantime, there is an ailing transportation infrastructure that urgently requires attention. If this represents the pinnacle of AI capabilities, perhaps concerns about its potential and its impact on human employment are exaggerated. For now, Sunak’s driver can feel reassured.
