Social Challenges of Automated Driving: From the development of AI technology to the development of relevant rules

MANAGI Shunsuke
Faculty Fellow, RIETI

Development of AI rules

Artificial intelligence (AI) is on the move, so much so that it is no exaggeration to say that not a single day goes by without seeing an article mentioning AI in newspapers. A major challenge at the moment is a lack of progress in the discussion of how our society should determine rules to be programmed well in advance when AI comes into greater use.

No matter how the volume of big data or the speed of computing may be increased, and regardless of how the entire big data process may be streamlined, we cannot expect any drastic change in society—such as one in which nearly half of workers would be replaced by AI—without first building consensus over what should be the basic objective of using AI, how agreement can be reached on the objective, and how to come to terms with various issues that could pose a social dilemma (Note 1).

A driver's failure to stay alert while in self-driving mode due to misunderstanding about the level of automation could result in a serious accident, for instance, when the car turns out to be not as automated as had been assumed by the driver (Note 2).

Going forward, more automated driving systems are expected to come into use and replace relatively simple driving assistance systems. And if any ambiguity exists as to whether it is humans or AI to hold ultimate decision-making authority, it must be clarified in advance.

Expectations on automated driving and challenges ahead

Autonomous cars (also known as robotic cars or driverless cars), which are capable of navigating and reaching a destination without a human driver behind the wheel, are drawing much attention including media coverage. In what follows, I would like to introduce how automated driving is perceived as a foundation for considering the ongoing debate on AI and future debate on artificial life.

In a bid to promote our understanding of the current level of public acceptance of automated driving in Japan, we conducted a large-scale questionnaire survey covering more than 240,000 individuals. The primary purpose of this survey was to assess the user acceptance of fully autonomous driving. Based on the assessed level of acceptance, we then examined the market potential challenges of fully autonomous vehicles under the business-as-usual scenario. We also explored what type of people would purchase a fully autonomous vehicle under varying conditions (price and functions), and what concerns they have in introducing or purchasing one.

Asked how soon they think they might want to purchase a fully autonomous car, most respondents (more than 80%) answered between one to 15 years' time, bringing the average timing of purchase to approximately 9.5 years' time or FY2025. With more than 70% of the respondents found to have in mind some idea about the timing for purchasing a fully autonomous car, public awareness and expectations of autonomous driving are fairly high.

Next, we asked whether they would purchase an optional feature that enables autonomous navigation and driving, if they are to purchase a car. Those who responded positively accounted for 47%. A closer look at those respondents reveals a significant difference by gender, with a disproportionately large number of male respondents showing willingness to purchase it. Meanwhile, 41% of those who currently do not own a car and 44% of those who do not have a driver's license responded positively. From the viewpoint of automobile makers, they can be seen as a new layer of customers, because autonomous driving capabilities would be a must-have feature for them.

Asked when or in what situation they would activate the autonomous driving mode, approximately half of the respondents said they would do so while driving on highways. Also, more respondents said that they would use the mode on lightly trafficked roads than those who intend to use it on heavily trafficked roads. Thus, their general assumption is that autonomous driving is for use when roads are easy to navigate and do not require sophisticated driving technique, as would also be the case when driving on highways. The most conspicuous difference between genders was observed in their responses to whether or not they would rely on autonomous driving while on highways, which is attributable to the combination of two factors, i.e., that female drivers use highways less often than their male counterparts and that women are less inclined to purchase autonomous driving capabilities.

We also examined respondents' willingness to pay (how much they would be willing to pay) by type of functions: 1) autonomous driving at high speed, 2) autonomous driving in a traffic jam, 3) autonomous parking, and 4) fully autonomous driving. The average amount the respondents would be willing to pay was approximately 110,000 yen for autonomous driving at high speed, 100,000 yen for autonomous driving in traffic jams, 90,000 yen for autonomous parking, and 190,000 yen for fully autonomous driving, when including those who chose "zero yen" as their answer. When limited to those with an intention to pay a certain amount, the figures rise to approximately 170,000 yen (n=153,625 people) for autonomous driving at high speed, 160,000 yen (n=157,409) for autonomous driving in a traffic jam, 160,000 yen (n=137,985) for autonomous parking, and 290,000 yen (n=163,200) for fully autonomous driving. As such, the amounts people would be willing to pay for the partial autonomous driving capabilities, particularly for autonomous parking, are high relative to the amount they would pay for fully autonomous driving. However, those amounts are far below the prices at which automobile makers wish to pay for those capabilities, indicating that there remains a significant gap to close before they can sell those products to enough consumers.

Next, we examined the amount the respondents would be willing to pay for each autonomous driving capability by type of respondents. Respondents without a driver's license showed less willingness to purchase a fully autonomous driving capability, but would pay a higher amount for the feature than the average respondent. Meanwhile, a comparison of respondents owning a car and those not showed that the latter would pay less for fully autonomous driving capability. We also found that elderly people are willing to pay a relatively high amount for fully autonomous driving capability.

Advantages and disadvantages of autonomous driving

We also asked respondents what advantages and disadvantages they see in fully autonomous driving. First, they found the following advantages.

The elimination of concerns about elderly drivers was found to be the greatest advantage of fully autonomous driving, reflecting respondents' anxiety about the dangers of elderly driving. Second, fully autonomous driving is seen as a necessary effective measure to reduce traffic accidents. Meanwhile, more than 10% of respondents said that the possible elimination of the need to obtain a driver's license as an advantage of fully autonomous driving. This is incompatible with the idea of requiring some sort of license, such as the one currently being discussed in California. Serving as a status symbol was the least selected as an advantage of owning a fully autonomous car, showing a distinctive difference from the time when hybrid cars first hit the market and celebrities rushed to buy them.

The greatest disadvantage was uncertainties about the safety of the technology, indicating that there still remain deep-rooted concerns. Since this is based on the survey conducted before the recent fatal Tesla accident, the level of concerns may be even higher today. Those who cited the possibility of information leakage and the impossibility of driving at a speed above the statutory limit were small in number. Instead, the possibility of children traveling on their own without their parents or guardians knowing was cited by more than 40% of respondents, pointing to the need to establish an appropriate licensing system and/or regulations for users in order to increase the public acceptance of fully autonomous cars.

What can be expected in the future?

Many consumers see advantages in having fully autonomous cars on the roads. One big reason is that autonomous cars are expected to reduce car accidents based on the way their functions are designed. However, from the viewpoint of consumers, things look different. Unable to understand the mechanism of autonomous driving, they are worried about the possibility of accidents and consider it as a disadvantage, and this poses many potential challenges to the deployment of autonomous cars.

Regarding the Tesla accident, automotive specialists point out that the vehicle in question cannot be defined as an autonomous car in a strict sense because the autonomous driving technology used in it is at a substandard level. However, the company sold the vehicle as an autonomous car and thus cannot excuse itself by saying that its autonomous driving feature was still in public beta mode.

It is undesirable to make plans for commercialization based on the assumption that people understand what is explained. It is reasonable to expect more accidents involving autonomous driving in the coming years. Autonomous driving might be able to prevent up to 90% of the traffic accidents we see today. However, as the distance traveled by autonomous cars is increasing across the world, responding to accidents may pose a huge challenge. Ethical issues could arise as well. Suppose that autonomous driving is not functioning and an accident is unavoidable. In this situation, should the human driver hit a wall to stop the car or hit a pedestrian instead? We will be facing, and be required to find an answer to, various situations where an individual's ethics is questioned socially as is the case in the above example. This is not a technological issue but a social issue that we face. If we are to deploy autonomous driving from an early stage of this technology, we must promote vigorous open discussions and deepen our understanding as to how we should solve various issues—including ethical ones—surrounding autonomous driving.

Lastly, we need to understand the limits of a human's ability to stay alert. Suppose that in the future we have a highly advanced technology that allows for almost fully autonomous driving. Someone, who used to drive in the past, is traveling aboard an autonomous vehicle, relaxing and doing almost nothing. Now, if suddenly encountered with a situation that cannot be dealt with automatically or if the vehicle's autonomous driving capabilities malfunction, would the human ex-driver be capable of coping with the situation? Even if a human aboard an autonomous car is authorized to take control of the vehicle when the needs arise, he or she may be too panicked to respond quickly. In promoting the automation of driving, rules must be developed by taking into account human behavior in an unfamiliar situation. Although some jobs will be lost to automation, the development of new rules will create new types of jobs because rulemaking is an act of humans. (Note 3)

July 26, 2016
  1. ^ With an aim to address those issues, we have launched a research project entitled "Economics of Artificial Intelligence" at RIETI.
  2. ^ In the United States, the driver of a Tesla Model S electric car was killed on May 7, 2016 after his car collided into a trailer. As the Tesla was operating in autopilot mode at the time of the accident, the automated driving systems have been cited as a possible cause of the crash. From the very beginning, Tesla Motors, Inc., has been insisting that it is the driver’s responsibility to keep his or her own safety. While the National Highway Traffic Safety Administration (NHTSA), a unit of the U.S. Department of Transportation, is still investigating the operation of the systems at the time of the accident, a consumer group has been denouncing Tesla Motors, saying that the automaker should stop calling its technology "autopilot" if it holds drivers responsible for any accident involving the technology. What we can see from this particular accident is that it is for humans—not AI—to decide the degree of responsibility to be assumed by humans as drivers.
  3. ^ Similar arguments can be made about artificial life, which is another direction in which the ongoing debate on artificial intelligence is going. As the term literally indicates, artificial life is about artificially providing life functions to machines. Artificial life would enable us to exploit numerous judgments that have been made by humans in similar situations, instead of trying to achieve social consensus. The ongoing efforts for the development of artificial life are focused on fundamental research. However, as with the case of artificial intelligence, things are moving in the direction of taking the aspect of real-world applicability into greater consideration. With all such technological possibilities in mind, we need to consider how we should assimilate those possibilities into our society.

August 24, 2016