Potential disadvantages
A direct impact of widespread adoption of automated vehicles is the loss of driving-related jobs in the road transport industry. There could be resistance from professional drivers and unions who are threatened by job losses. In addition, there could be job losses in public transit services and crash repair shops. The automobile insurance industry might suffer as the technology makes certain aspects of these occupations obsolete. A frequently cited paper by Michael Osborne and Carl Benedikt Frey found that automated cars would make many jobs redundant.
Privacy could be an issue when having the vehicle’s location and position integrated into an interface in which other people have access to. In addition, there is the risk of automotive hacking through the sharing of information through V2V (Vehicle to Vehicle) and V2I (Vehicle to Infrastructure) protocols. There is also the risk of terrorist attacks. Self-driving cars could potentially be loaded with explosives and used as bombs.
The lack of stressful driving, more productive time during the trip, and the potential savings in travel time and cost could become an incentive to live far away from cities, where land is cheaper, and work in the city’s core, thus increasing travel distances and inducing more urban sprawl, more fuel consumption and an increase in the carbon footprint of urban travel. There is also the risk that traffic congestion might increase, rather than decrease. Appropriate public policies and regulations, such as zoning, pricing, and urban design are required to avoid the negative impacts of increased suburbanization and longer distance travel.
Some believe that once automation in vehicles reaches higher levels and becomes reliable, drivers will pay less attention to the road. Research shows that drivers in automated cars react later when they have to intervene in a critical situation, compared to if they were driving manually. Depending on the capabilities of automated vehicles and the frequency with which human intervention is needed, this may counteract any increase in safety, as compared to all-human driving, that may be delivered by other factors.
Ethical and moral reasoning come into consideration when programming the software that decides what action the car takes in an unavoidable crash; whether the automated car will crash into a bus, potentially killing people inside; or swerve elsewhere, potentially killing its own passengers or nearby pedestrians. A question that programmers of AI systems find difficult to answer (as do ordinary people, and ethicists) is “what decision should the car make that causes the Ďsmallestí damage to people’s lives?”
The ethics of automated vehicles are still being articulated, and may lead to controversy. They may also require closer consideration of the variability, context-dependency, complexity and non-deterministic nature of human ethics. Different human drivers make various ethical decisions when driving, such as avoiding harm to themselves, or putting themselves at risk to protect others. These decisions range from rare extremes such as self-sacrifice or criminal negligence, to routine decisions good enough to keep the traffic flowing but bad enough to cause accidents, road rage and stress.
Human thought and reaction time may sometimes be too slow to detect the risk of an upcoming fatal crash, think through the ethical implications of the available options, or take an action to implement an ethical choice. Whether a particular automated vehicle’s capacity to correctly detect an upcoming risk, analyze the options or choose a ‘good’ option from among bad choices would be as good or better than a particular human’s may be difficult to predict or assess. This difficulty may be in part because the level of automated vehicle system understanding of the ethical issues at play in a given road scenario, sensed for an instant from out of a continuous stream of synthetic physical predictions of the near future, and dependent on layers of pattern recognition and situational intelligence, may be opaque to human inspection because of its origins in probabilistic machine learning rather than a simple, plain English ‘human values’ logic of parsable rules. The depth of understanding, predictive power and ethical sophistication needed will be hard to implement, and even harder to test or assess.
The scale of this challenge may have other effects. There may be few entities able to marshal the resources and AI capacity necessary to meet it, as well as the capital necessary to take an automated vehicle system to market and sustain it operationally for the life of a vehicle, and the legal and ‘government affairs’ capacity to deal with the potential for liability for a significant proportion of traffic accidents. This may have the effect of narrowing the number of different system operators, and eroding the presently quite diverse global vehicle market down to a small number of system suppliers.