PRE2020 3 Group11
The acceptance of self-driving cars
Problem statement
What are the relevant factors that contribute to the acceptance of self-driving cars for the private end-user?
Self-driving cars are believed to be more safe than manually driven cars. However, they can not be a 100% safe. Because crashes and collisions are unavoidable, self-driving cars should be programmed for responding to situations where accidents are highly likely or unavoidable (Sven Nyholm, Jilles Smids, 2016). There are three moral problems involving self-driving cars. First, the problem of who decides how self-driving cars should be programmed to deal with accidents exists. Next, the moral question who has to take the moral and legal responsibility for harms caused by self-driving cars is asked. Finally, there is the decision-making of risks and uncertainty.
There is the trolley problem, which is a moral problem because of human perspective on moral decisions made by machine intelligence, such as self-driving cars. For example, should a self-driving car hit a pregnant woman or swerve into a wall and kill its four passengers? There is also a moral responsibility for harms caused by self-driving cars. Suppose, for example, when there is an accident between an autonomous car and a conventional car, this will not only be followed by legal proceedings, it will also lead to a debate about who is morally responsible for what happened (Sven Nyholm, Jilles Smids, 2016).
A lot of uncertainty is involved with self-driving cars. The self-driving car cannot acquire certain knowledge about the truck’s trajectory, its speed at the time of collision, and its actual weight. Second, focusing on the self-driving car itself, in order to calculate the optimal trajectory, the self-driving car needs to have perfect knowledge of the state of the road, since any slipperiness of the road limits its maximal deceleration. Finally, if we turn to the elderly pedestrian, again we can easily identify a number of sources of uncertainty. Using facial recognition software, the self-driving car can perhaps estimate his age with some degree of precision and confidence. But it may merely guess his actual state of health (Sven Nyholm, Jilles Smids, 2016).
The decision-making about self-driving cars is more realistically represented as being made by multiple stakeholders; ordinary citizens, lawyers, ethicists, engineers, risk assessment experts, car-manufacturers, government, etc. These stakeholders need to negotiate a mutually agreed-upon solution (Sven Nyholm, Jilles Smids, 2016). This report will focus on the relevant factors that contribute to the acceptance of self-driving cars with the main focus on the private end-user. Taking into account the ethical theories: utilitarianism, kantianism, virtue ethics, deontology, ethical plurism, ethical absolutism and ethical relativism, the moral and legal responsibility, safety, security, privacy and the perspective of the private end-user.
State-of-the-art/Hypothesis
What are the relevant factors that contribute to the acceptance of self-driving cars for the private end-user?
Survey
Demografie
Wat is uw leeftijd?
Open vraag
Wat is uw geslacht?
- Man
- Vrouw
- Anders
Wat is uw opleidingsniveau?
- Geen opleiding/ onvolledige basisonderwijs
- Middelbaar
- Hoger beroepsonderwijs
- Universiteit bachelors diploma
- Universiteit Masters diploma
Heeft u een auto rijbewijs?
- Ja
- Nee
Hoe vaak gebruikt u een auto?
- (Bijna) elke dag
- Vaak
- Af en toe
- (Bijna) nooit
General/User perspective
Hoe bekend bent u met zelfrijdende auto’s?
Zeer bekend, bekend, enigszins onbekend, onbekend
Hoe waarschijnlijk denkt u dat de volgende voordelen optreden bij het gebruik van volledig zelfrijdende auto’s?
zeer waarschijnlijk, enigszins waarschijnlijk, enigszins onwaarschijnlijk, zeer onwaarschijnlijk
- Minder ongelukken
- Verminderde ernst van ongelukken
- Minder verkeersopstoppingen
- Kortere reizen
- Lagere voertuigemissies
- Betere brandstofbesparing
- Lagere verzekeringstarieven
Security/safety
(zeer bezorgd, bezorgd, enigszins bezorgd, niet bezorgd)
Hoe bezorgd bent u over de volgende kwesties gerelateerd aan volledig zelfrijdende voertuigen?
- Autorijden in een voertuig met zelfrijdende technologie
- Veiligheids consequenties van apparaatstoring of systeem mislukking
- Wettelijke aansprakelijkheid voor chauffeurs/eigenaren
- Systeembeveiliging (van hackers)
- Gegevensprivacy (locatie en bestemming)
- Interactie met niet-zelfrijdende voertuigen
- Interactie met voetgangers en fietsers
- Zelfrijdende voertuigen leren gebruiken
- Systeemprestaties in slecht weer
- Verwarde zelfrijdende voertuigen door onvoorspelbare situaties
- Zelfrijdende voertuigen rijden niet even goed als de mens zelf in het algemeen
- Rijden in een voertuig zonder dat de bestuurder kan ingrijpen
Ethical setting
In sommige onvermijdbare ongelukken zal de auto een keuze moeten maken tussen verschillende mensenlevens, bijvoorbeeld tussen die van de bestuurder en voetgangers. Het zou zelfs mogelijk zijn een instelling aan een zelfrijdende auto toe te voegen die bepaalt welke keuze de auto zou moeten maken in het geval van een ongeluk. De onderstaande vragen gaan over deze ethische instelling.
Met welke ethische instelling ziet u het liefst zelfrijdende auto’s op de weg? Rangschik van favoriet (1) naar minst favoriet (5):
1. De auto zou altijd het leven en gezondheid van de inzittende(n) moeten prioriteren boven die van omstander(s).
2. De auto zou altijd het leven en gezondheid van de omstander(s) prioriteren boven die van de inzittende(n).
3. De auto zou altijd moeten kiezen om de minste hoeveelheid schade aan te richten bij de minste hoeveelheid mensen, of ze nu omstander of inzittend zijn.
4. De auto zou geen expliciete keuze mogen maken tussen mensenlevens, en zou dus bij een onvermijdbaar ongeluk niet moeten ingrijpen. Als gevolg zal er dus willekeurig een slachtoffer vallen.
5. De keuze die de auto maakt zou gebaseerd moeten zijn op wat het merendeel van de weggebruikers wilt.
Note to ourselves: from top to bottom they represent the following ethical theories: egoism, virtue ethics (kinda), utilitarianism, deontology, contractualism.
In hoeverre bent u het met de volgende stellingen eens:
zeer eens, eens, oneens, zeer oneens
- Zelfrijdende auto’s moeten niet verkocht worden met een door de gebruiker instelbare ethische instelling. In plaats daarvan moet deze instelling vastgesteld worden door bijvoorbeeld de regering of de fabrikant.
- Ik zou liever een zelfrijdende auto kopen met een instelbare ethische instelling dan een zelfrijdende auto met een onaantastbare ethische instelling.
- Als ik een een zelfrijdende auto gebruik, zou voor mij de specifieke ethische instelling van belang zijn.
- De auto zou altijd mijn leven en gezondheid moeten prioriteren boven die van omstander(s).
- De auto zou altijd het leven en gezondheid van de omstander(s) prioriteren boven die van mijzelf.
- De auto zou altijd moeten kiezen om de minste hoeveelheid schade aan te richten bij de minste hoeveelheid mensen, of ik het nu ben of omstanders.
- De auto zou geen expliciete keuze mogen maken tussen mensenlevens, en zou dus bij een onvermijdbaar ongeluk niet moeten ingrijpen. Als gevolg zal er dus willekeurig een slachtoffer vallen.
- De keuze die de auto maakt zou gebaseerd moeten zijn op wat het merendeel van de weggebruikers wilt.
Responsibility
In hoeverre zou u het volledig zelfrijdende voertuig gebruiken in de volgende stellingen?
Zeer waarschijnlijk, enigszins waarschijnlijk, enigszins onwaarschijnlijk, zeer onwaarschijnlijk
1. Fabrikanten zijn volledig aansprakelijk als een zelfrijdende auto een ongeluk maakt, ook als dit ze ontmoedigt om zelfrijdende auto’s te produceren.
2. Fabrikanten zijn gedeeltelijk aansprakelijk zodat ze wel produceren, maar altijd aangemoedigd zijn om fouten te verbeteren.
3. Ik ben zelf aansprakelijk als mijn zelfrijdende auto een ongeluk maakt, ook al kan ik niet zelf ingrijpen (volledig autonome auto).
4. Ik ben zelf aansprakelijk als mijn zelfrijdende auto een ongeluk maakt, omdat ik de mogelijkheid heb om in te grijpen (semi-autonome auto).
5. Iedereen met een zelfrijdende auto is aansprakelijk als een zelfrijdende auto een ongeluk maakt door middel van een verplichte verzekering of belasting.
https://deepblue.lib.umich.edu/bitstream/handle/2027.42/108384/103024.pdf?sequence=1&isAllowed=y
Ethical theories
A key feature of self-driving cars is that the decision making process is taken away from the person in the driver’s seat, and instead bestowed upon the car itself. From this several ethical dilemmas emerge, one of which is essentially a version of the trolley problem. When an unavoidable collision will occur, it is important to define the desired behaviour of the self-driving car. It might be the case that in such a scenario, the car has to choose whether to prioritize the life and health of its passengers or the people outside of the vehicle. In real life such cases are relatively rare [reference 1] , but the ethical theory underlying that decision will have possibly an impact on the acceptance of the technology. Self-driving vehicles that decide who might live and who might die are essentially in a scenario where some moral reasoning is required in order to produce the best outcome for all parties involved. Given that cars seem not to be capable of moral reasoning, programmers must choose for them the right ethical setting on which to base such decisions on. However, ethical decisions are not often clear cut. Imagine driving at high speed in a self-driving car, and suddenly the car in front comes to a sudden halt. The self-driving car can either suddenly break as well, possibly harming the passengers, or it can swerve into a motorcyclist, possibly harming them. This scenario can be regarded as an adapted version of the trolley problem. One could argue that since the motorcyclist is not at fault, the self-driving car should prioritize their safety. After all, the passenger made the decision to enter the car, putting at least some responsibility on them. On the other hand, people who buy might buy the self-driving car will have an expectation to not be put in avoidable danger. No matter the choice of the car, and the underlying ethical theory that it is (possibly) based on, it is likely that the behaviour and decision-making of the car has more chance of being socially accepted if it can morally be justified. Therefore in this section there is first highlighted some possible ethical theories, and then we will discuss some relevant aspects that surround the implementation of all ethical theories. - Different ethical theories explanation (utilitarianism, Kantianism, deontology, contractualists, egoism, ethical plurism/absolutism/relativism). Seems to me that people prefer utilitarianism if they are not in the car themselves, but always want it to save the driver if they are the driver. Therefore I predict that contractualism or egoism seems best. Utilitarianism is also not as straightforward as it seems: the utilitarian solution might be to always save the owner, resulting in more people buying the car, resulting in safer roads. - Explicitly choose ethical setting vs neural nets (which would effectively choose one, being a black box) - Ethical knob and letting the user set the ethical setting (pluralism) - Game theory - Relevance of ethical theories - Ethical theories that are bad for users might not be popular - Knowing the theory might change people’s behaviours (pedestrians who know the car will prioritize them might not pay as much attention; a perfect car might do everything in its power to avoid collision with them)
Ethical theories under consideration Although there are not a lot of actions a car could take in the above-described scenario, there are a lot of ethical theories that can help to inform the car to make such a decision. The most prominent ethical theories that might prima facie be useful are utilitarianism, deontology, virtue ethics, contractualism, and egoism. There are also three meta-ethical frameworks which should be considered: Relativism, absolutism, and pluralism. These frameworks can by themselves not influence the decision-making process as they make no normative claims of their own, but they are useful in deciding how the possible ethical knob of a self-driving car might work (see section …). Utilitarianism considers consequences of actions, as opposed to the action itself. This means that the correct moral decision or action in any scenario is the one that produces the most good. Although ‘’good’’ is a subjective term, in most versions of utilitarianism this usually refers to the net happiness or welfare increase for all associated parties. [reference]. Circumstances or the intrinsic nature of an action is not taken into account, as opposed to Deontology. Deontology does not judge the morality of an action based on its consequences, but on the action itself. Deontology posits that moral actions are those actions which have been taken on the grounds of a set of pre-determined rules, which hold universally and absolutely. This means that for a deontologist, some actions are wrong or right no matter their outcome.
The third major normative ethical theory is virtue ethics. Virtue ethics emphasizes the virtues, or moral character, as opposed to rules or consequences. Virtues are seen as positive or ‘’good’’ character traits. Examples of such traits are courage, or modesty. A moral person should do actions which realize these traits, and therefore moral actions are those which cause a persons’ virtues to be realized.
Other than the three major classical normative ethics theories, there are two more prima facie relevant theories. The first of which is egoism. Normative egoism posits that the only actions that should be taken morally are those actions that maximize the individuals self-interest. An egoist only considers the benefit and detriments other people experience in so far as those experiences will influence the self-interest of the egoist. Although it may not seem like it, egoism is very similar to utilitarianism, except that utilitarianism focusses on the maximum happiness of all people involved, and egoism only focusses on the maximum happiness of the individual.
The last ethical theory that can be applied to the adapted trolley problem is (social) contractualism. Contractualism does not make any claims about the inherent morality of actions, but rather posits that a moral action is one that is mutually agreed upon by all parties affected by the action. What this agreement should look like exactly differs per version of contractualism: in some versions there must be unanimous consent, while in other versions there must be a simple or a supermajority. A good action is therefore one that can be justified by other relevant parties, and a wrong action is one that cannot be justified by the same.
-ethical plurism -ethical absolutism -ethical relativism
Ethical theories applied to adapted trolley problem.
Responsibility
Whereas automated vehicles were a distant future a mere twenty years ago, they are a reality right now. For some years companies like for example Google have run trials with automated vehicles in actual traffic situations. Though this research has been ground-breaking, there have also been some incidents in the past years. In 2016 a Tesla driver was killed while using the car’s autopilot because the vehicle failed to recognize a white truck (Yadron & Tynon, 2016).
One very important factor in the development and sale of automated vehicles is the question of who is responsible when things go wrong. In this section we will look in detail at all factors involved and come up with certain solutions. As brought up by Marchant and Lindor (2012), there are three questions that need to be analysed. Firstly, who will be liable in the case of an accident? Secondly, how much weight should be given to the fact that autonomous vehicles are supposed to be safer than conventional vehicles in determining who of the involved people should be held responsible? Lastly, will a higher percentage of crashes be caused because of a manufacturing ‘defect’, compared to crashes with conventional vehicles where driver error is usually attributed to the cause (Marchant & Lindor, 2012)?
Current legislation If we take a look at how responsibility works for conventional vehicles, we find that responsibility is usually addressed to the drive due to failure to obey to the traffic regulations (Pöllänen, Read, Lane, Thompson, & Salmon, 2020). This can be as small and common as driving too fast or losing attention for a fraction of a moment, something nearly everyone is guilty of doing at some point. Where this usually doesn’t matter, sometimes it can lead to catastrophically results. This moment of misfortune still holds the driver responsible. As Nagel (1982) theorized, between driving a little too fast and killing a child than crosses the street unexpected and there being no child, there is only bad luck. The consequence, however, is vast for the child, but also for the driver (Nagel, 1982). This reasoning could also be applied to automated vehicles. If an accident happens it is just bad luck for the driver, and he will without doubt be liable. However, looking at the fact that this depends on luck, and the fact that most autonomous vehicles allow for restricted to no control, this option is not considered as a plausible one (Hevelke & Nida-Rümelin, 2015).
Blame attribution A couple of studies have shown that the level of control is crucial in blame attribution. McManus and Rutchick (2018) showed that people attribute less blame to a driver in a fully automated vehicle in comparison to a situation where the driver selected a different algorithm (e.g. to behave selfishly) or drove manually (McManus & Rutchick, 2018). Another study (Li, Zhao, Cho, Ju, & Malle, 2016) investigated blame attribution between the manufacturer, government agencies, the driver and pedestrians. They found that blame is reduced for drivers when the vehicle is fully autonomous, whereas the blame for the manufacturer or government agencies increased.
The manufacturer It would be obvious to say the manufacturer of the car is responsible. They designed the car, so if it makes a mistake, they are to blame. However, there are different types of defects in the manufacturing process. Firstly, there is a defect in manufacturing itself, where the product did not end up as it was supposed to, even though rules are followed with care. This error is very rare, since manufacturing these days is done with a very low error rate (Marchant & Lindor, 2012). A second defect lies in the instructions. When it is failed to adequately instruct and warn, this could result in a consumer defect. A third defect, and the most significant for autonomous vehicles, is that of design. This holds that the risks of harm could have been prevented or reduced with an alternative design (Marchant & Lindor, 2012).
Any flaw in the system that might cause the car to crash, the manufacturers could have known or did know beforehand. If they then sold the car anyway, there is no question in that they are responsible. However, by holding the manufacturer responsible in every case, it would immensely discourage anyone to start producing these autonomous cars. Especially with technology as complex as autonomous driving systems, it would be nearly impossible to make it flawless (Marchant & Lindor, 2012). In order to encourage people to manufacture autonomous vehicles and still hold them responsible, a balance needs to be found between the two. This is necessary, because removing all liability would also result in undesirable effects (Hevelke & Nida-Rümelin, 2015). In short, there needs to be found a way to hold the manufacturer liable enough that they will keep improving their technology.
Semi-autonomous vehicles As stated above, there have been studies on blame attribution in fully autonomous vehicles, and those with certain pre-selected algorithms. A semi-autonomous vehicle (with a duty to intervene) has not been discussed yet. A good analogy for a semi-autonomous vehicle would be that of an auto-piloted airplane. The plane flies itself, though it is the responsibility of the pilot to intervene when something goes wrong (Marchant & Lindor, 2012). So, it could be suggested in the question of responsibility in case of an accident to hold the driver of the vehicle responsible. If the car is designed in such a way that the driver has the ability to take over and intervene, this could really be used in an argument against the driver. There is an argument in what the utility of the automated vehicle will be if they are designed like this. After all, when the driver has a duty to intervene, the vehicle can no longer be summoned when needed, it can no longer be used as a safe ride home when drunk, or when tired (Howard, 2013). However, as long as the vehicles will still reduce accidents overall, saying the driver has a duty to intervene or not would still be a better option than using conventional vehicles (Hevelke & Nida-Rümelin, 2015). It could be that the accident rate is dropped even more when the driver actually does have a duty to intervene, due to the fact that it can now intervene when it for example sees something the car doesn’t see. It would also mean that there is more of a transitioning phase when introducing the automated vehicles, instead of them suddenly being fully automatic.
On the other hand, asking the driver to intervene in a fully automated vehicle is questionable. It would assume that the driver can intervene at all times, and this is not always the case due to human error in reaction time or danger anticipation (Hevelke & Nida-Rümelin, 2015). It would be difficult to recognize whether or not the automated vehicle will fail to respond correctly, and thus unclear when the driver needs to intervene. In this case it would be unrealistic to expect the driver to predict a dangerous situation. When implementing this reasoning, another problem is possible to arise: the driver might intervene when it shouldn’t have, resulting in an accident (Douma & Palodichuk, 2012). Next to that, as argued by Hevelke & Ninda-Rümelin (2015), it seems impossible to ask a driver to pay attention all the time to be possible to intervene, while an actual accident is quite rare. All in all, it would be unreasonable to put responsibility on a driver that did not – or could not – intervene.
Shared liability
As is previously discussed, the responsibility of an accident can be placed on the individual driving the autonomous vehicle. For a number of reasons this was not ideal. An alternative would be to create a shared liability. People that drive cars everyday (especially when not necessary) take the risk of possibly causing an accident. They still make the choice to drive the car (Husak, 2004). You can extrapolate this thinking to the use of automated vehicles. If people choose to drive an automated vehicle, they in turn participate in the risk of an accident happening due to the autonomous vehicle. The responsibility of an accident is therefore shared with everyone else in the country also using the automated vehicle. In that sense the driver itself did not do something wrong, it did not intervene too late, it simply shoulders the burden with everyone else. A system that could work with this line of thinking is the entering of a tax or mandatory insurance (Hevelke & Nida-Rümelin, 2015).
So, it seems there are a couple of options. The manufacturer can be fully responsible; however, this could result in the intermittence of autonomous vehicle manufacturing. On the other hand, it is desirable that the manufacturer does have some sort of liability, so they keep investing to improve the vehicle. At the same time, giving the driver full responsibility only seems to be able to work in the beginning phase of autonomous vehicles. When they are still in development, and drivers really do have a duty to intervene. When the vehicles are more sophisticated and able to fully drive autonomously, the responsibility can be shared with all people through a tax or insurance.
Safety
One of the main factors deciding whether self-driving cars will be accepted is the safety of them. Because who would leave their life in the hands of another entity, knowing it is not completely safe. Though almost everyone gets into buses and planes without doubt or fear. Would we be able to do the same with self-driving cars? Cars have become more and more autonomous over the last decades. Furthermore, self-driving cars will operate in unstructured environments, this adds a lot of unexpected situations. (Wagner et al., 2015)
Software
Traffic behaviour The cars safety will be determined by the way it is programmed to act in traffic. Will it stop for every pedestrian? If it does pedestrians will know and cross roads wherever they want. Will it take the driving style of humans? How does the driving behaviour of automated vehicles influence trust and acceptance?
In a research two different designs were presented to a group of participants. One was programmed to simulate a human driver, whilst the other one is communicating with it’s surroundings in a way that it could drive without stopping or slowing down. The research showed no significant different in trust of the two automated vehicles. However, it did show that the longer the research continued the trust grew. (Oliveira et al., 2019) It is therefore to say that the driving behaviour does not necessarily influence the acceptance. But the overall safety of the driving behaviour determines this.
Errors
Despite what we think, humans are quite capable of avoiding car crashes. It is inevitable that a computer never crashes, think about how often your laptop freezes. A slow response of a mini second can have disastrous consequences. Software for self-driving vehicles must be made fundamentally different. This is one of the major challenges currently holding back the development of fully automated cars. On the contrary automated air vehicles are already in use. However, software on automated aircraft is much less complex since they have to deal with fewer obstacles and almost no other vehicles. (Shladover, 2016)
Cybersecurity
The software driving fully AV will have more than 100 million lines of code, so it is impossible to predict the security problems. Windows 10 is made of 50 million lines of code and there have been lots of bugs. Double the amount of code will result in an even higher probability of unknown vulnerabilities. (Parkinson et al., 2017)
Vs humans
Self-driving cars hold the potential of eliminating all accidents, or at least those caused by inattentive drivers. (Wagner et al., 2015) In a research done by Google it is suggested that the Google self-driving cars are safer than conventional human-driven vehicles. However, there is insufficient information to fully take a conclusion on this. But the results lead us to believe that highly-autonomous vehicles will be more safe than humans in certain conditions. This does not mean that there will be no car-crashes in the future, since these cars will keep on being involved in crashes with human drivers (Teoh et al., 2017)
The city
The city is probably one of the most complicated locations for a self-driving car to operate in. It is filled with vulnerable road users, such as pedestrians and bikers which are relatively hard to track. Therefore, freeways are likely to be the first spaces in which the automated cars will be able to operate. This is a much more structured environment met simple rules and less unexpected situations. However, this will not solve the issue of traffic jam at popular destinations. Some might say the ambition is to allow cars, bikes and pedestrians to share road space much more safely, with the effect that more people will choose not to drive. ‘But, if a driverless car or bus will never hit a jaywalker, what will stop pedestrians and cyclists from simply using the street as they please?’ (Duranton, 2016)
Google acknowledges this problem and states that when Google cars cannot operate in existing cities, perhaps new cities need to be created. And the truth is, this sounds silly, but it has happened in the past. The first suburb of America was developed by rail entrepreneurs who realized that developing suburbs was much more profitable than operating railways. (Cox, 2016)
We might need to look at alternative technologies that we need in urban transport. Rather than developing individualist self-driving cars, let’s look at the ‘technology of the network’. How can we connect more people without consuming the space we live in. (Duranton, 2016)
Trust
For decades, we have trusted safe operation of automated mechanisms around and even inside us. However, in the last few years the autonomy of these mechanisms have drastically increased. As mentioned above, this brings along quite a few safety risks. Questions of whether or not to trust a new technology are often answered by testing. (Wagner et al., 2015)
There has been a survey about the trust in fully automated vehicles. Trust was defined as “the attitude that an agent will help achieve an individual’s goal in a situation characterised by uncertainty and vulnerability” (Lee & See, 2004, p.51) Within this survey 60% of the respondents mentioned to have difficulties trusting automated vehicles. Trust in this context can be seen as the driver’s belief that the computer drives at least as good as a human. However, with some uncertaintity of the driver might get involved in accidents because of failures. (Bock, German & Sippl, 2017 cited in Johsen et al., 2017)
The trust to be able to fully implement these technologies is not where it is supposed to be. We know that trust can built-up over time and this is also the case with trusting self driving cars. The hesitation is the greatest amongst the elderly, whom are also the generation that gain a lot of benefits as well. The good news of this research is that 50% of the older adults reported back that they are comfortable with the concept of tools that can help the driver. The amount of tools can grow, whilst the driver/passenger can get used to the idea of a completely self-driving car (Abraham et al., 2016)
Privacy
Perspective of private end-user
Additional features important for users
While many people look positively towards the implementation of SDC’s, less people are willing to buy one. Also, many people don’t want to invest more money in SDC’s than they do in conventional cars right now. Therefore, a car sharing scheme is preferred by many. Also many people say that they will still have concerns riding a SDC and they prefer to be able to intervene manually whenever they want or need to. Additionally, people like to take over full control when they like to.
Most important benefits or concerns (in order of relevance).
- An SDC could solve transport-problems for older or disabled people.
- People are able to do other things while driving an SDC.
- People are concerned of legal issues caused by SDC’s.
- People are concerned of hackers’ attacks at SDC’s.
Strategic implications (in order of relevance).
- A feature making the user able to take over full control should be implemented. Female and old users showed the highest agreement. Pros: People are still able to enjoy the pleasures of manually driving and they don’t lose the emotion of freedom. Cons: The total efficiency of driving will decrease. People will most likely drive less efficient, if they don’t speed. If every car drives autonomously, the cars can communicate better and adapt earlier and better to each other. Other SDC’s can’t predict what a human driver will do. It is likely that more accidents will take place, because SDC’s will most likely be safer.
- Free test rides should be offered to people.
- Salesmen should offer comprehensive information in the showroom.
(König, M., & Neumayr, L. 2017b)
As already said above, more people are willing to accept SDC’s when they don’t have to buy a car themselves. This means that sharing cars will be the new normal. The idea is that you can order one with a mobile app or something like that and it will drive to you by itself. This is only possible if SDC’s become autonomous at the highest level. If they are autonomous, but require a person to intervene when things go wrong, they may not drive without passenger. As also mentioned, people don’t accept fully autonomous cars as much as cars with a possibility to intervene. The problems posed by ridesharing are that not all passengers, who don’t know each other, may travel from the same point to the same point. Also, people may not always feel to comfortable when they travel with strangers. Therefore, people are willing to accept this idea more when they can order a ride for themselves and when it doesn’t stop to pick up others. That way, it will become available again when the ride is finished. This will require more cars on the road in total than when rides are shared, so it only solves part of the traffic problem. The same amount of people need to move themselves at the same time as now and buses or trains will be made less use of, because cars will be more accessible. As world population also increases, ridesharing may be necessary. A solution would be that ordering a private ride will be more expensive. Then, only a part of the population (wealthy businessmen etcetera), would make use of this option and the majority of the people would have to ride with others. Only the existence of this option and the possibility of enjoying a private ride when you really need to, could make it easier for people to accept. One benefit of not owning cars, will be that parking spots within cities won’t be needed anymore. The cars could be deployed from a base outside the city and they can be parked there when not needed.
Additional features important for users While many people look positively towards the implementation of SDC’s, less people are willing to buy one. Also, many people don’t want to invest more money in SDC’s than they do in conventional cars right now. Therefore, a car sharing scheme is preferred by many. Also many people say that they will still have concerns riding a SDC and they prefer to be able to intervene manually whenever they want or need to. Additionally, people like to take over full control when they like to. Most important benefits or concerns (in order of relevance). - An SDC could solve transport-problems for older or disabled people. - People are able to do other things while driving an SDC. - People are concerned of legal issues caused by SDC’s. - People are concerned of hackers’ attacks at SDC’s. Strategic implications (in order of relevance). - A feature making the user able to take over full control should be implemented. Female and old users showed the highest agreement. Pros: People are still able to enjoy the pleasures of manually driving and they don’t lose the emotion of freedom. Cons: The total efficiency of driving will decrease. People will most likely drive less efficient, if they don’t speed. If every car drives autonomously, the cars can communicate better and adapt earlier and better to each other. Other SDC’s can’t predict what a human driver will do. It is likely that more accidents will take place, because SDC’s will most likely be safer. - Free test rides should be offered to people. - Salesmen should offer comprehensive information in the showroom.
(König, M., & Neumayr, L. 2017b)
As already said above, more people are willing to accept SDC’s when they don’t have to buy a car themselves. This means that sharing cars will be the new normal. The idea is that you can order one with a mobile app or something like that and it will drive to you by itself. This is only possible if SDC’s become autonomous at the highest level. If they are autonomous, but require a person to intervene when things go wrong, they may not drive without passenger. As also mentioned, people don’t accept fully autonomous cars as much as cars with a possibility to intervene. The problems posed by ridesharing are that not all passengers, who don’t know each other, may travel from the same point to the same point. Also, people may not always feel to comfortable when they travel with strangers. Therefore, people are willing to accept this idea more when they can order a ride for themselves and when it doesn’t stop to pick up others. That way, it will become available again when the ride is finished. This will require more cars on the road in total than when rides are shared, so it only solves part of the traffic problem. The same amount of people need to move themselves at the same time as now and buses or trains will be made less use of, because cars will be more accessible. As world population also increases, ridesharing may be necessary. A solution would be that ordering a private ride will be more expensive. Then, only a part of the population (wealthy businessmen etcetera), would make use of this option and the majority of the people would have to ride with others. Only the existence of this option and the possibility of enjoying a private ride when you really need to, could make it easier for people to accept. One benefit of not owning cars, will be that parking spots within cities won’t be needed anymore. The cars could be deployed from a base outside the city and they can be parked there when not needed.
Cox, W. (2016)
According to Chaiwoo Lee et al. there are three important determinants of self-driving car acceptance. These are perceived usefulness, affordability, social support, lifestyle fit and conceptual compatibility. This means that factors like responsibility and security mean less to people who maybe want to buy a SDC. Younger people are more likely to consider making use of SDC’s and this is caused, for a big part, by the fact that they have less experience with technology. They may have fear for new things like this, because they are unfamiliar with them and they don’t know how it works. People who reported to have some knowledge about technology, were more likely to purchase a SDC. This even makes trust towards technology the most important factor in making the decision to buy a SDC. Therefore, it may be an idea to give extensive information in car shops or on tv for example. It is hard to solve this problem, for they may not be interested in this information at all, but it could solve a part of the problem. History proves that human beings in general have always been afraid of new technologies, like trains, wireless network systems or even normal cars. Another big factor in buying an SDC is the possession of a driver’s license. This implies that way more people will buy/use an SDC when they become fully autonomous. If they aren’t fully autonomous, it would not be responsible to let a human person without driver’s license drive of course.
Lee C., Ward C., Raue M., D’Ambrosio L., Coughlin J.F. (2017)
According to Martina Raue et al. positive and negative unconscious feelings are more important in making the decision to buy an SDC than strong arguments. People can recall more negative emotion words than positive, which is caused by the fact that they are more likely to recall a negative event than a positive. This implies that people’s attitude is heavily affected by the event of an SDC-crash. It would therefore of course, be better to make such a crash not a headliner in the newspapers. But this is of course not in the interests of newspaper companies, for they only want to sell as much papers as they can. It would be better to write more positive articles towards SDC’s, but as already said, bad news sticks to your mind. Therefore either newspaper companies write more bad news-articles and they remain more important in people’s minds.
Raue, M., D’Ambrosio, L. A., Ward, C., Lee, C., Jacquillat, C., & Coughlin, J. F. (2019)
References used in report
Sven Nyholm, Jilles Smids. (2016). The Ethics of Accident-Algorithms for Self-Driving Cars: an Applied Trolley Problem? Ethical Theory and Moral Practice, 1275–1289.
Hevelke, A., & Nida-Rümelin, J. (2015). Responsibility for Crashes of Autonomous Vehicles: An Ethical Analysis. Science and Engineering Ethics, 21(3), 619–630. https://doi.org/10.1007/s11948-014-9565-5
Marchant, G. E., & Lindor, R. A. (2012). Santa Clara Law Review The Coming Collision Between Autonomous Vehicles and the Liability System THE COMING COLLISION BETWEEN AUTONOMOUS VEHICLES AND THE LIABILITY SYSTEM. Number 4 Article, 52(4), 12–17. Retrieved from http://digitalcommons.law.scu.edu/lawreview
Wagner M., Koopman P. (2015) A Philosophy for Developing Trust in Self-driving Cars. In: Meyer G., Beiker S. (eds) Road Vehicle Automation 2. Lecture Notes in Mobility. Springer, Cham. https://doi.org/10.1007/978-3-319-19078-5_14
Oliveira, L., Proctor, K., Burns, C. G., & Birrell, S. (2019). Driving Style: How Should an Automated Vehicle Behave? Information, 10(6), 219. MDPI AG. Retrieved from http://dx.doi.org/10.3390/info1006021
Shladover, S. (2016). THE TRUTH ABOUT “SELF-DRIVING” CARS. Scientific American, 314(6), 52-57. doi:10.2307/26046990
26 References
Greenblatt, N. A. (2016). Self-driving cars and the law. IEEE Spectrum, 46-51. doi:10.1109/MSPEC.2016.7419800
Holstein, T., Dodic-Crnkovic, G., & Pellicione, P. (2018). Ethical and Social Aspects of Self-Driving Cars. Retrieved from https://arxiv.org/abs/1802.04103
Nielsen, T. A., & Haustein, S. (2018). On sceptics and enthusiasts: What are the expectations towards self-driving cars? Transport Policy, 49-55. Retrieved from https://doi.org/10.1016/j.tranpol.2018.03.004
Stilgoe, J. (2018). Machine learning, social learning and the governance of self-driving cars. Social Studies of Science, 25-56. Retrieved from https://doi.org/10.1177/0306312717741687
Wagner, M., & Koopman, P. (2015). A Philosophy for Developing Trust in Self-driving Cars. Road Vehicle Automation 2, 163-171. Retrieved from https://link.springer.com/chapter/10.1007/978-3-319-19078-5_14
Sven Nyholm, Jilles Smids. (2016). The Ethics of Accident-Algorithms for Self-Driving Cars: an Applied Trolley Problem? Ethical Theory and Moral Practice, 1275–1289.
Nyholm, S. R. (2018). The ethics of crashes with self-driving cars: a roadmap I.
Chandiramani, J. R. (2017). Decision Making under Uncertainty for Automated Vehicles in Urban Situations. Master of Science Thesis.
Ibo van de Poel, Lambèr Royakkers. (2011). Ethics, Technology, and Engineering an introduction. Wiley-Blackwell.
Sam Levin, Nicky Woolf. (2016). Tesla driver killed while using autopilot was watching Harry Potter, witness says. The Guardian. https://www.theguardian.com/technology/2016/jul/01/tesla-driver-killed-autopilot-self-driving-car-harry-potter
Alexander Hevelke, Julian Nida-Rümelin. (2014). Responsibility for Crashes of Autonomous Vehicles: An Ethical Analysis. Science and Engineering Ethics Joshua Greene. (2013). Moral Tribes.
Noah J. Goodall. (2016). Ethical Decision Making During Automated Vehicle Crashes
Bonnefon, J.-F., Shariff, A., & Rahwan, I. (2016). The Social Dilemma of Autonomous Vehicles. Science, 1573-1576.
Katarzyna de Lazari-Radek, Peter Singer. Utilitarianism: A Very Short Introduction (2017), p.xix, ISBN 978-0-19-872879-5.
Shladover, S. (2016). THE TRUTH ABOUT “SELF-DRIVING” CARS. Scientific American, 314(6), 52-57. doi:10.2307/26046990
Duranton, G. (2016). Transitioning to Driverless Cars. Cityscape, 18(3), 193-196. Retrieved February 7, 2021, from http://www.jstor.org/stable/26328282
Cox, W. (2016). Driverless Cars and the City: Sharing Cars, Not Rides. Cityscape, 18(3), 197-204. Retrieved February 7, 2021, from http://www.jstor.org/stable/26328283
Stone, J. (2017). Who’s at the wheel: Driverless cars and transport policy. ReNew: Technology for a Sustainable Future, (139), 38-41. Retrieved February 7, 2021, from https://www.jstor.org/stable/90002086
Frey, T. (2012). DEMYSTIFYING THE FUTURE: Driverless Highways: Creating Cars That Talk to the Roads. Journal of Environmental Health, 75(5), 38-40. Retrieved February 7, 2021, from http://www.jstor.org/stable/26329536
Focussed on acceptance of the technology:
König, M., & Neumayr, L. (2017b). Users’ resistance towards radical innovations: The case of the self-driving car. Transportation Research Part F: Traffic Psychology and Behaviour, 44, 42–52. https://doi.org/10.1016/j.trf.2016.10.013
Nees, M. A. (2016). Acceptance of Self-driving Cars. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 60(1), 1449–1453. https://doi.org/10.1177/1541931213601332
S. Karnouskos, "Self-Driving Car Acceptance and the Role of Ethics," in IEEE Transactions on Engineering Management, vol. 67, no. 2, pp. 252-265, May 2020, doi: 10.1109/TEM.2018.2877307.
Lee C., Ward C., Raue M., D’Ambrosio L., Coughlin J.F. (2017) Age Differences in Acceptance of Self-driving Cars: A Survey of Perceptions and Attitudes. In: Zhou J., Salvendy G. (eds) Human Aspects of IT for the Aged Population. Aging, Design and User Experience. ITAP 2017. Lecture Notes in Computer Science, vol 10297. Springer, Cham. https://doi.org/10.1007/978-3-319-58530-7_1
Raue, M., D’Ambrosio, L. A., Ward, C., Lee, C., Jacquillat, C., & Coughlin, J. F. (2019). The Influence of Feelings While Driving Regular Cars on the Perception and Acceptance of Self-Driving Cars. Risk Analysis, 39(2), 358–374. https://doi.org/10.1111/risa.13267
Wagner M., Koopman P. (2015) A Philosophy for Developing Trust in Self-driving Cars. In: Meyer G., Beiker S. (eds) Road Vehicle Automation 2. Lecture Notes in Mobility. Springer, Cham. https://doi.org/10.1007/978-3-319-19078-5_14
Oliveira, L., Proctor, K., Burns, C. G., & Birrell, S. (2019). Driving Style: How Should an Automated Vehicle Behave? Information, 10(6), 219. MDPI AG. Retrieved from http://dx.doi.org/10.3390/info1006021
Shladover, S. (2016). THE TRUTH ABOUT “SELF-DRIVING” CARS. Scientific American, 314(6), 52-57. doi:10.2307/26046990
S. Parkinson, P. Ward, K. Wilson and J. Miller, "Cyber Threats Facing Autonomous and Connected Vehicles: Future Challenges," in IEEE Transactions on Intelligent Transportation Systems, vol. 18, no. 11, pp. 2898-2915, Nov. 2017, doi: 10.1109/TITS.2017.2665968.
Teoh, E. R., & Kidd, D. G. (2017). Rage against the machine? Google’s self-driving cars versus human drivers. Journal of Safety Research, 63, 57–60. https://doi.org/10.1016/j.jsr.2017.08.008
Lee, J. D., & See, K. A. (2004). Trust in Automation: Designing for Appropriate Reliance. Human Factors: The Journal of the Human Factors and Ergonomics Society, 46(1), 50–80. https://doi.org/10.1518/hfes.46.1.50_30392
Abraham, H., Lee, C., Brady, S., Fitzgerald, C., Mehler, B., Reimer, B. & Coughlin, J.F. (2016). Autonomous Vehicles, Trust, and Driving Alternatives: A survey of consumer preferences. Agelab Life tommorow, 0. Geraadpleegd van https://bestride.com/wp-content/uploads/2016/05/MIT-NEMPA-White-Paper-2016-05-30-final.pdf
Responsibility
Douma, F., & Palodichuk, S. A. (2012). Criminal Liability Issues Created by Autonomous Vehicles. Santa Clara Law Review, 52(4), 1157–1169. Retrieved from http://digitalcommons.law.scu.edu/lawreview
Hevelke, A., & Nida-Rümelin, J. (2015). Responsibility for Crashes of Autonomous Vehicles: An Ethical Analysis. Science and Engineering Ethics, 21(3), 619–630. https://doi.org/10.1007/s11948-014-9565-5
Howard, D. (2013). Robots on the Road: The Moral Imperative of the Driverless Car. Retrieved March 7, 2021, from Science Matters website: http://donhoward-blog.nd.edu/2013/11/07/robots-on-the-road-the-moral-imperative-of-the-driverless-car/#.U1oq-1ffKZ1
Marchant, G. E., & Lindor, R. A. (2012). Santa Clara Law Review The Coming Collision Between Autonomous Vehicles and the Liability System THE COMING COLLISION BETWEEN AUTONOMOUS VEHICLES AND THE LIABILITY SYSTEM. Number 4 Article, 52(4), 12–17. Retrieved from http://digitalcommons.law.scu.edu/lawreview
Summaries References
Self-driving cars and the law
The law assumes that a human being is in the driver’s seat of a car. This poses a problem for the inevitable futuristic implementation of self-driving cars. No current laws state who is responsible when an accident happens. Also, roads aren’t adapted to the needs of SDC’s. The biggest part of testing the cars takes place in the United States. A human being must be behind the wheel to intervene before possible accidents there. New laws in favor of these cars must be made soon, because companies won’t fully invest before they know the necessary regulations exist. Car companies are afraid for lawsuits, because they will be extremely expensive, the verdict will be hard to predict because no laws exist and a single lawsuit can lead to a recall of all cars. Finally, car companies are afraid for high punitive damage awards.
The solution, according to the writer, would be that computer drivers should be treated equally as human drivers. Only its conduct needs to be considered, not the thoughts, just like a judge can’t see what a human driver thought when he caused an accident. This means that a computer driver would be found liable when it runs a red light for example. Not when it drives as safe as it can and still cause an accident. The carmaker would be responsible, because they are the ones determining the actions of the car. Afterwards, the carmaker would invest much money in improving safety, because of bad publicity reasons. Judges have much precedent, because cases in which human beings were involved, can be used when a computer is involved and insurances would be lower than for a normal car.
Changes in public policy have to be made as well. A human can see traffic lights and signs et cetera. Camera’s of a car would be able to see and detect them as well, but it would be much easier if radio frequency transmitters were to be implemented. That way, a car can just receive a signal without the chance of not visually detecting it. The rollout of autonomous vehicles has to be speeded, because much oil could be saved, for they drive more efficiently. Also many accidents can be prevented, because they are more reliable than a human being. They rely on electronic signals, which can be processed much faster. Also, SDC’s can learn from mistakes that another car has made, through updates. Human beings can only learn from their own mistakes. Probably when they are implemented, we wouldn’t own cars anymore, but just order one when you need one. Privacy will be a big concern, since manufacturers will know your exact location and your destination. Camera’s are likewise to be installed internally to prevent vandalism. At last, parking spaces in central areas would not be needed anymore, so we have more space.
Ethical and Social Aspects of Self-Driving Cars
It is hard to say which choices a SDC has to make when dangerous situations occur. The trolley-problem is a commonly used model to describe which choices it can make. There exist a few problems why we cannot fully depend on this problem. There are a few ethical theories which we can use, but it’s hard to say which one is correct and they all have different conclusions. Design choices and ethical programming influence each other. A cheaper camera has a negative effect on accurate quick decision-making for example. Self-driving cars are cars which are able to operate without the presence of a human being. This is the highest level of autonomous cars. Self-adaptive software can make sure that a car learns all the time and is not dependent on slow updates. Most of the functionality in the automotive domain is based on software. This relies on computer vision, machine learning and parallel computing. A problem is that calculations are based on an abstract representation of the real world, formed by all things sensed by camera’s etcetera. Engineers have to choose which data they will use, as a camera can see an obstacle when a radar doesn’t.
Safety is the most important requirement of SDC’s. A drivers license for SDC’s is a suggestion. Also an independent organization should be able to check the code. Testing is very important for making sure it is safe enough. Economic aspects happen to be the highest priority of companies and cheap equipment could lead to wrong decision-making. Security is also very important, because when hackers hack into the device, safety will be affected crucially. There exist eight basic principles for security. Should there be a threshold for safety? Should the vehicle be connected to the network or not? Connection makes it easier to prevent accidents and operate more efficiently whereas no connection makes it almost impossible to hack.
Privacy is another requirement. There already exists much legislation on privacy and these cars have to meet them as well. Trust and transparency are other requirements. It is hard to determine to which organizations/people data has to be disclosed. More data-sharing can make it easier for companies to learn from others. Reliability, responsibility and accountability and quality assurance are the final requirements. There are public interests which manufacturers have to take into account. Giving more choice to people can make the people more responsible for the cars actions. New selling points have to be thought of. Will exterior be just as important as it is right now? There are no simple answers to each safety question, but this was the same when normal cars were introduced. Also with those, safety couldn’t be guaranteed. The unsolvable trolley problem must not be tried to be solved.
On sceptics and enthusiasts: What are the expectations towards self-driving cars?
Willingness to accept SDC’s differs between groups of age, gender, country etc. It is unknown what makes this difference and therefore more research on acceptance is needed. This paper studies the acceptance of automated driving and related expectations in the Danish population. The parts of the questionnaire were: car access and travel patterns, interest and attitudes towards SDC’s, expectations towards fully automated vehicles and personal background information. People were divided into the groups of scepticism, enthusiasm and car stress for each part of the questionnaire. Most people are sceptic, followed by indifferent stressed and enthusiasts form the smallest group by extent. Enthusiasts are younger people who live in more urban areas, whereas sceptics are often older and live in more rural areas. Powerlessness and freedom are emotions related to not accepting a SDC.
Enthusiasts live more in urban areas because they are more familiar with driving in congested areas and therefore feel the need towards more efficient driving methods. Although old people might not be able to use conventional driving methods, they are still not willing to accept SDC’s. The differences must be studied and conclusions must be drawn about how we can implement SDC’s so that more people will enjoy them. For example, we can consider to keep manual options, so sceptics won’t lose the joy of driving. For all the results, see the paper.
Machine learning, social learning and the governance of self-driving cars
Developers of SDC’s should aim to make them safer. It is hard to do that, because innovation is very unsure. You don’t know exactly what your product will be, so it is hard to know exactly what you want to be safe to do what. The final design can be different from the current design so you’re testing a different thing. The focus should be on social learning. The system needs to learn from the society and the society needs to learn from the system. Much can also be learned from historical cases. The algorithmic architecture of the programming begins with if-else rules, but situations are too complex to use only this approach. It should learn from vast datasets out of the real world. There could be problems because regulations aren’t necessarily based on needs in the real world and may be arbitrary. Developers aren’t even capable of seeing how the system is learning from the data. Therefore explicit problems need to be defined beforehand. Self-driving and autonomous cars are misnomers, because they are never autonomous. They are driven by social goals. Technology can never have an own will. You must make people aware of the limits. This is why the German government has asked Tesla to rename the Autopilot-function due to failure. It’s dangerous for people to think it’s completely safe. Tesla never connected the failure to their own shortages. But they did install technological alternatives when they noticed some weren’t good enough. Autonomous cars aren’t as independent as people tend to believe. They should be well-trained and therefore it would be positive to democratize the learning, so that every company can maximize the outcome and the safety.
A Philosophy for Developing Trust in Self-driving Cars
Cars become more automated and this will reduce the rates of accidents. They can eliminate accidents due to inattentive drivers. However, humans are able to react way better to situations they are not explicitly trained for. The world contains is an unstructured network and even thousand test-miles cannot eliminate some failures. Inductive inference is crucial in building solid software. This is for example machine learning. A computer can learn for itself what the clearest feature of pedestrians are and how to react to them.
Situations which occur not very often are hard to take into account for programmers and it is not easy to learn for them through experience. According to Popper a theory is only meaningful when it is falsifiable, because one needs only one negative example to falsify a theory. According to the author, one single accident makes the safety case more meaningful. No confirmatory tests should be executed. Rather, the goal should be a negative test result, so we know what to improve. Field testing costs too much money to do for a long time and simulation testing doesn’t fit either, because one will never simulate situations he doesn’t expect to take place. Fuzz testing is a well-fitting alternative, but it is not very efficient, because it uses random values of which a great part aren’t very interesting to test. The Ballista project uses dictionaries of interesting values to test and therefore is more likely to find big vulnerabilities. The conclusion is that the tester should aim to find flaws, instead of never-ending evidence the system works at all times.
The ethics of crashes with self‐driving cars: A roadmap, I
Self‐driving cars hold out the promise of being much safer than regular cars. Yet they cannot be 100% safe. Accordingly, they need to be programmed for how to deal with crash scenarios. Should cars be programmed to always prioritize their owners, to minimize harm, or to respond to crashes on the basis of some other type of principle? The article first discusses whether everyone should have the same “ethics settings.” Next, the oft‐made analogy with the trolley problem is examined. Then follows an assessment of recent empirical work on lay‐people's attitudes about crash algorithms relevant to the ethical issue of crash optimization. Finally, the article discusses what traditional ethical theories such as utilitarianism, Kantianism, virtue ethics, and contractualism imply about how cars should handle crash scenarios.
It might seem like a good idea to always hand over control to a human driver in any accident scenario. However, typical human reaction‐times are too slow for this to always be a good idea (Hevelke & Nida‐Rümelin, 2015) Jason Millar argues that a person's car should function as a “proxy” for their ethical outlook. People should therefore be able to choose their own ethics settings (Millar, 2014; see also Sandberg & Bradshaw‐Martin, 2013). Similarly, Giuseppe Contissa and colleagues argue that self‐driving cars should be equipped with an “ethical knob,” so that whoever is currently using the car can set it to their preferred settings. (Contissa, Lagioia, & Sartor, 2017) Jan Gogoll and Julian Müller, in contrast, argue that we all have self‐interested reasons to want everyone's cars to be programmed according to the same settings. (Gogoll & Müller, 2017). One advantage to giving people a certain degree of choice here is that this might make it easier to hold them responsible for any bad outcomes that crashes involving their vehicles might give rise to (Sandberg & Bradshaw‐Martin, 2013; cf. Lin, 2014).
One of the questions this raises is whether the vast literature on the trolley problem might be a useful source of ideas about how to deal with the ethics of crashing self‐driving cars. Together with Jilles Smids, I have put forward three reasons for being skeptical about relying very heavily on the trolley problem literature here (Nyholm & Smids, 2016). Firstly, in the trolley literature, we are typically asked to imagine that the only morally relevant factors are a very small set of factors. . Any bigger and more complex sets of considerations are imagined away. Secondly, in most trolley discussions, we are asked to set all questions of moral and legal responsibility aside, and only focus on the choice between the one and the five. In actual traffic ethics, we cannot ignore questions about responsibility. Thirdly, in trolley discussions, a fully deterministic scenario is imagined. It is assumed that we know with certainty what the outcomes of our available choices would be. In contrast, when we are prospectively programming self‐driving cars for how to deal with accident scenarios, we do not know what scenarios they will face. We must make risk‐assessments. (Nyholm & Smids, 2016). Emperical ethics: minimize overall harm. . However, when surveyed about what kinds of cars they themselves would want to use, people tend to favor cars that would save them in an accident scenario. People appear to have inconsistent or paradoxical attitudes. In the finding mentioned above, many people want others to have harm‐minimizing cars, while themselves wanting to have cars that would favor them.
“Top‐down” approach. That is, we can consider what utilitarians (or consequentialists more broadly), Kantians (or deontologists more broadly), virtue ethicists, or contractualists would recommend regarding this topic. Utilitarian ethics is about maximizing overall happiness, while minimizing overall suffering. Kantian ethics is about adopting a set of basic principles (“maxims”) fit to serve as universal laws, in accordance with which all are treated as ends‐in‐themselves and never as mere means. Virtue ethics is about cultivating and then fully realizing a set of basic virtues and excellences. Contractualist ethics is about formulating guidelines people would be willing to adopt as a shared set of rules, based on nonmoral or self‐interested reasons, in a hypothetical scenario where they would be making an unforced agreement about how to live together. A utilitarian would be mindful of the fact that people might be scared of taking rides in “utilitarian” cars, instead preferring cars programmed to prioritize their passengers. . The lesson from Kantian ethics might be that we should choose rules we would be willing to have as universal laws applying equally to all—so as to make everything fair, and not give some people an unjustified advantage in crash‐scenarios. ? It is hard to come up with any virtue ethical ideas about how self‐driving cars should crash (cf. Gurney, 2016). But virtue ethics might help when we think about the ethics of automated driving more generally. Perhaps a lesson from a virtue ethical perspective is that we should try to design and program cars in ways that help to make people act carefully and responsibly when they 6 of 10 NYHOLM use self‐driving cars.
The Ethics of Accident-Algorithms for Self-Driving Cars: an Applied Trolley Problem?
We identify three important ways in which the ethics of accidentalgorithms for self-driving cars and the philosophy of the trolley problem differ from each other. These concern: (i) the basic decision-making situation faced by those who decide how selfdriving cars should be programmed to deal with accidents; (ii) moral and legal responsibility; and (iii) decision-making in the face of risks and uncertainty.
According to Frances Kamm, the basic philosophical problem is this: why are certain people, using certain methods, morally permitted to kill a smaller number of people to save a greater number, whereas others, using other methods, are not morally permitted to kill the same smaller number to save the same greater number of people? (Kamm 2015) The morally relevant decisions are prospective decisions, or contingency-planning, on the part of human beings. In contrast, in the trolley cases, a person is imagined to be in the situation as it is happening, split-second decision-making. It is unlike the prospective decision-making, or contingency-planning, we need to engage in when we think about how autonomous cars should be programmed to respond to different types of scenarios we think may arise. The decision-making about self-driving cars is more realistically represented as being made by multiple stakeholders – for example, ordinary citizens, lawyers, ethicists, engineers, risk-assessment experts, car-manufacturers, etc. These stakeholders need to negotiate a mutually agreed-upon solution. . In one case, the morally relevant decision-making is made by multiple stakeholders, who are making a prospective decision about how a certain kind of technology should be programmed to respond to situations it might encounter. And there are no limits on what considerations, or what numbers of considerations, might be brought to bear on this decision. In the other case, the morally relevant decision-making is done by a single agent who is responding to the immediate situation he or she is facing – and only a very limited number of considerations are taken into account.
Responsibility: Suppose, for example, there is a collision between an autonomous car and a conventional car, and though nobody dies, people in both cars are seriously injured. This will surely not only be followed by legal proceedings. It will also naturally – and sensibly – lead to a debate about who is morally responsible for what occurred. Forward-looking responsibility is the responsibility that people can have to try to shape what happens in the near or distant future in certain ways. Backward-looking responsibility is the responsibility that people can have for what has happened in the past, either because of what they have done or what they have allowed to happen. (Van de Poel 2011) Applied to riskmanagement and the choice of accident-algorithms for self-driving cars, both kinds of responsibility are highly relevant.
Uncertainties: the self-driving car cannot acquire certain knowledge about the truck’s trajectory, its speed at the time of collision, and its actual weight. Second, focusing on the self-driving car itself, in order to calculate the optimal trajectory, the self-driving car needs (among other things) to have perfect knowledge of the state of the road, since any slipperiness of the road limits its maximal deceleration. Finally, if we turn to the elderly pedestrian, again we can easily identify a number of sources of uncertainty. Using facial recognition software.
Responsibility for Crashes of Autonomous Vehicles: An Ethical Analysis
Autonomous cars are involved around legal, but also moral questions. Patrick Lin is concerned that any security gain will constitute a trade-off with human lives. The second question is whether it would be morally okay to put liability on the user based on a duty to pay attention to the road and traffic and to intervene when necessary to avoid accidents. It should depend on whether or not the driver would ever have a chance to intervene. In this article, two options are discussed: driver with a duty to intervene, or a driver with no duty (and thus no control). For the first option, if the driver never had a real chance of intervening, he should not be held responsible. However this holds only for the new cars, and they would still not be accessible to blind etc. For the second option where the driver has no control, it makes more sense to hold them accountable. However, this would make more sense in some kind of tax or insurance. Manufacturers should not be freed of their liability completely (take the Ford Pinto case as an example).
Ethical decision making during automated vehicle crashes
Three arguments were made in this paper: automated vehicles will almost certainly crash, even in ideal conditions; an automated vehicle’s decisions preceding certain crashes will have a moral component; and there is no obvious way to effectively encode human morality in software. A three-phase strategy for developing and regulating moral behavior in automated vehicles was proposed, to be implemented as technology progresses. The first phase is a rationalistic moral system for automated vehicles that will take action to minimize the impact of a crash based on generally agreed upon principles, e.g. injuries are preferable to fatalities. The second phase introduces machine learning techniques to study human decisions across a range of real-world and simulated crash scenarios to develop similar values. The rules from the first approach remain in place as behavioral boundaries. The final phase requires an automated vehicle to express its decisions using natural language, so that its highly complex and potentially incomprehensible-to-humans logic may be understood and corrected.
The social dilemma of autonomous vehicles
When it becomes possible to program decision-making based on moral principles into machines, will self-interest or the public good predominate? In a series of surveys, Bonnefon et al. found that even though participants approve of autonomous vehicles that might sacrifice passengers to save others, respondents would prefer not to ride in such vehicles (see the Perspective by Greene). Respondents would also not approve regulations mandating self-sacrifice, and such regulations would make them less willing to buy an autonomous vehicle.
The truth about ‘self-driving’ cars
They are coming but not in the way you may have been led to think. Selfdriving cars have many issues: taking save turns, changing road surfaces, snow and ice and avoid traffic cops, crossing guards & emergency vehicles. And automatic stopping for pedestrians will make us people rather walk or take the subway.
We have a very unrealistic expectation of self driving cars. They will not happen the way you have been told.
We are currently only arriving at level 3 cars. CEO of Nissan said fully automated cars (level 5) will be on the road by 2020. This isn’t true, level 4 cars may arrive in the next decade. Defining automated driving: much more complex than we think. Despite the popular perception, human drivers are remarkably capable of avoiding crashes. Mind how often your laptop freezes / is slow. This will inevitably lead to crashes, so there is a major software problem.
Software on aircraft is much less complex, since they have to deal with less obstacles and other vehicles. Also, the testing of the automated cars will have lots of problems. A lot of people will have to be subject of crashes statistically over a long period of time. Also, there is boundary money-wise, since the cars must stay affordable for the public. Some people think AI will give us self-driving cars. However, the problem with that is that it is non-deterministic. The possibility of having 2 cars with the same assembly but after a year automation systems will have different behaviour. It is out of our control.
Writer: Fully automated cars will not be here until 2075. In level 3 cars there is a problem with the driver zoning out. This problem so hard, some car manufacturers will not even try level 3. So outside of traffic jam assistants level 3 will probably never happen. Level 4 will happen eventually, but on certain parts of roads and with certain weather conditions. These scenarios might not sound as futuristic as having your own personal electronic chauffeur, but they have the benefit of being possible and soon.
Transitioning to driverless cars
Despite some nuances, the future looks mostly bright. The questions are how to get there, and what the transition to a full system of driverless cars will look like. A lot of the discussion so far has focused on insurance and ethical issues. Who is responsible in case of accidents? If the computer has to choose a victim in a collision, who will it be, its own passenger or a passenger in another car? These questions are interesting, but it is hard to imagine they will be major stumbling blocks. New technologies have brought new risks for many years, and ways have been found to spread those risks and define new forms of protection and liability. The ethical question probably makes for interesting debates in an introduction to ethics class at a university, but it is unlikely to have much practical relevance. Driverless cars will be much safer than cars are now.
A good case can be made that the key transitional problems will be instead about the political economy of the regulation of driverless cars and the cohabitation between driverless cars and cars driven by human beings. For car producers or would-be car producers, two strategies are possible. The first is incremental and consists of making cars gradually less reliant on drivers. That has been the strategy of most incumbent car producers. The incremental strategy presents one major problem, however. Partially driverless cars may be safer, but the true timesaving benefits of driverless cars will occur only when cars become completely driverless. With this scenario, the transition is likely to be extremely long, and how the last step about getting rid of the wheel will take place is unclear.
The alternative strategy is rupture and the direct development of cars without a steering wheel; that is the Google, Inc. strategy. It is an appealing but difficult proposition on several counts. It will require maximum software sophistication right from the start. If anything, processes will get easier with more driverless cars. Some technical issues seem extremely tricky to resolve. Incumbent car manufacturers that are betting on incremental change, not cars without wheels right from the start, will probably do everything they can to prevent fully driverless cars from being able operate.
Realizing that its radical innovation will be a hard sell, Google appears to want to make it even more radical. If Google cars cannot operate in existing cities, perhaps new cities need to be created for them. That probably sounds like a mad idea to many, but history teaches us that it may not be as crazy as it sounds. What was possibly the first suburb of America, the Main Line of Philadelphia, Pennsylvania, was developed by rail entrepreneurs who realized that developing suburbs was much more profitable than operating railways.
Driverless Cars and the City: Sharing Cars, Not Rides
The world of driverless cars heralds revolutionary changes, but for cities (metropolitan areas) the process will be evolutionary. No “Big Bang” will happen, but it will slowly evolve. Driverless cars will not significantly impact urban form, but will expand opportunity and quality of life for the disabled and other people who are unable to drive.
Who’s at the wheel: Driverless cars and transport policy Many of the claims for the benefits of driverless technologies rely on the complete transformation of the existing vehicle fleet. But the transition will not be smooth or uniform: winners and losers in the competition between the different interest groups will depend on many factors.
Freeways are likely to be the first spaces in which the new vehicles will be able to operate. In any case, problems of congestion and competition for space at any popular destination will not be resolved. The ambition is to allow cars, bikes and pedestrians to share road space much more safely than they do today, with the effect that more people will choose not to drive. But, if a driverless car or bus will never hit a jaywalker, what will stop pedestrians and cyclists from simply using the street as they please?
Some analysts are even predicting that the new vehicles will be slower than conventional driving, partly because the current balance of fear will be upset. While this might be attractive to cyclists, will it affect the marketability of Google’s new products? With huge reserves of cash and consequent lobbying power, Google and its ilk will be in a strong position to demand concessions from governments and road authorities. You can just imagine the pitch: we can save you billions on public transport operations, but we need fences to keep bikes and pedestrians out of the way of our vehicles in busy urban centres. Lost in the enthusiasm for the new, is the simple reality of the limited availability of urban space. New technologies of driverless trains may reduce costs and allow us to improve the quality of the service, but only if that is the focus of investment and innovation.
I would urge readers of ReNew to turn their minds to the real alternative technologies we need in urban transport. Rather than follow the individualist model which directs our attention to the technology of the vehicle, let’s turn our attention to the ‘technology of the network’. How can we build on the insights of the Europeans and Canadians and use the potentials of IT and electronics to build better collective transport systems that connect all of us to the life of the city without consuming all the space we need to live and grow.
Driverless Highways: Creating Cars That Talk to the Roads
The art of road building has been improving since the Roman Empire. The highways today remain as little more than dumb surfaces with no data flowing between vechicles and the road. China already has restrictions on the limit of vehicles that can be licensed in Shanghai and Beijing. Going driverless brings some exciting new options. Driverless cars will be a very disruptive technology. To compensate for the loss of a driver, vehicles will need to become more aware of their surroundings. With cameras you create a symbiotic relationship that is far different than human-to-road relationship, which is largely emotion based. An intelligent car coupled with an intellegent road is a powerfull source.
- Lane compression
- Distance compression
- Time compression
On-demand transportation. All car parts and component need to be designed to be more durable and longer lasting. Shifting from driver to rider. More fancy dashboards, movies, music and massage interfaces. China doesn’t need more cars, it needs more transportation.
Conclusion: We all love to drive, but humans are the inconsistent variable in this demanding area of responsibility. Driving requires constant vigilance, constant alertness, and constant involvement. Once we take the driver out of the equation, however, we solve far more problems than the wasted time and energy needed to pilot the vehicle. But vehicle design is only part of the equation. Without reimagining the way we design and maintain highways, driverless cars will only achieve a fraction of their true potential.
Users’ resistance towards radical innovations: The case of the self-driving car
The advent of self-driving cars could eliminate the driver from the driving equation, having the potential to substantially improve safety, time and fuel efficiency as well as mobility in general. The introduction of such a radically new technology is surrounded by a high degree of uncertainty and possibly not all stakeholders would welcome the change. As a result, the wide-spread acceptance and hence adoption of this new technology is far from certain and will thus be analyzed comprehensively in this paper. Given that it will be the end-consumers (the actual drivers) who will eventually decide whether self-driving cars will successfully materialize on the mass market the lack of wider empirical evidence for the user perspective forms the rationale for our research.
User resistance to change has been found to be a crucial cause for many implementation problems. The assumption that a possibly disruptive innovation such as the self-driving car could lead to major resistance on behalf of the public is based on the fact that people regularly react with caution and wariness to ‘new things’ and ‘change’ or, in extreme cases, even fight them.
Possible causes of resistance: Regarding the desired level of automation, Khan, Bacchus, and Erwin (2012, p. 88) hypothesize that “it is likely that a significant percentage of drivers may not be comfortable with full autonomous driving.”
1. People might experience driving to be “adventurous, thrilling and pleasurable” (Steg, 2005, p. 148). Mokhtarian and Salomon (2001, p. 695) argue that travel “is not only derived demand”, but may be “desired for its own sake”. While self-driving cars might post significant advantages for many segments of the population, driving enthusiasts might not be among the people adopting this new technology.
2. Similarly, analyzing reasons why people do not use public transportation, Böhm et al. (2006, p. 4) make a distinction between “moving” and “being moved”, highlighting the latter as “dependent”. This poses the question whether self-driving cars could be seen as providing the ultimate level of autonomy, as people are free to engage in any activity once relieved from the task of driving or, psychologically, making people dependent on technology.
3. Further, as people regularly view their cars as source of power and similar attributes, “it is uncertain whether this close identification of personal autonomy with a person’s vehicle may be different with regard to use of autonomous vehicles” (Glancy, 2012, p. 1188).
4. Other users might resist self-driving technology not because they value the driving task but because they simply do not trust “a machine making decisions for them” (Rupp & King, 2010, p. 3).
5. There are also privacy issues.
6. Another potential cause for barriers towards self-driving technology is the risk of a “misbehaving computer system” (Douma & Palodichuk, 2012, p. 1164). With autonomous vehicles, criminals or terrorists might be able to hack into and use their cars for illegal purposes such as drug trafficking or, even worse, terroristic attacks (Douma & Palodichuk, 2012).
7. Further, the unavoidable rate of failure (or crashes), no matter how small, could foster initial mistrust, especially as people tend to underestimate the safety of technology while putting excessive trust in human capabilities like their own driving skills
Results: This study is an explorative study, since scientific research about self-driving cars is still in its infancy. A non-probability convenience sampling method was applied. Data were collected over a two-week time frame in July 2015 using a quantitative self-completion online questionnaire. Discussion: there were considerable differences between sub-groups with older respondents to be more worried about self-driving cars than younger respondents, females to have more concerns than males, and rural respondents to value self-driving cars less than urban participants. Surprisingly, people who used a car more often tended to be less open. Correspondingly, and across all sub-groups, the most pronounced desire of respondents was to have the possibility to manually take over control of the driving task whenever wanted, which entails the necessity to keep the steering wheel. It is thus seen as crucial to include an overriding function in the initial versions of self-driving cars. It stood out that the more participants knew about self-driving cars, the more positive their attitude towards these vehicles tended to be. Thus, a lack of knowledge about the functioning of the product will most certainly lead to non-adoption.
Acceptance of Self-driving Cars
One study (Payre, Cestac, & Delhomme, 2014) reported that driving while impaired from alcohol, drugs, or medications was a major dimension of acceptance of self-driving vehicles, and other studies have suggested that people expect to be able to engage in a wide variety of secondary tasks in self-driving cars (Kyriakidis, Happee, & De Winter, 2014; Pettersson & Karlsson, 2015).
These emerging expectations may reflect overconfidence in our ability to automate the driving task. The implementation of autonomous vehicles faces considerable unresolved challenges. Unless automation of driving can be implemented with perfect or near-perfect reliability—an outcome that seems implausible, especially during anticipated transitional phases of deployment during which self-driving cars will share roads with traditional vehicles (Sivak & Schoettle, 2015)—the human likely will retain a supervisory role during automated driving. Human operators of autonomous vehicles seem to be in danger of being allocated an especially mundane function: to continuously maintain awareness of the driving scenario in anticipation of very infrequent occasions when human intervention will be necessary. Even if appropriate interfaces can be designed to keep drivers in the loop, it remains unclear whether consumers would accept an automated vehicle that could perform all driving tasks, did perform most driving tasks, yet demanded a high amount of monitoring workload.
Highly idealized portrayals have begun to foster expectations that self-driving cars will require little or no human intervention and will create a windfall of work, leisure, or social time during transit. Initial deployment of self-driving cars could be slowed or harmed if the technology is received with disappointment. Trust in automation is influenced by expectations and attitudes that develop before a person uses a system (Hoff & Bashir, 2015), thus it will be important to understand acceptance before the arrival of self-driving cars on markets (see Payre et al., 2014). To the extent that idealized portrayals of vehicle automation already have begun to influence acceptance, they may also be encouraging unrealistic expectations about automation performance that could be counterproductive to acceptance in the long run.
In this experiment, an online sample of participants read either a realistic or an idealized description of a close friend or family member’s experiences during the first six months of ownership of a self-driving car. The realistic vignette emphasized that the driver felt the need to monitor the vehicle during automated operations and occasionally needed to resume manual control to prevent accidents. The idealistic scenario described a vehicle with perfect reliability that did not require human monitoring or intervention and had won the driver’s trust. A novel, 24-item scale assessed acceptance of self-driving cars in both vignette conditions and a control condition. The idealized portrayal was hypothesized to increase overall acceptance of self-driving cars.
Participants completed an instrument created for this experiment, the Selfdriving Car Acceptance Scale (SCAS). The SCAS featured 24 statements that were written to assess the extent to which participants were accepting of self-driving cars. Responses were made on a 7-point Likert scale with the anchors “strongly disagree” and “strongly agree.” People may be more accepting of self-driving cars under idealized rather than (arguably more) realistic scenarios during the initial deployment of the technology. The effect of the idealized depiction was small, but it suggested that idealized descriptions may be able to affect acceptance of self-driving cars before people interact with them.
Self-Driving Car Acceptance and the Role of Ethics
In the scope of unavoidable accidents, what is the effect of different ethical frameworks governing self-driving car decision-making, on their acceptance? Research question: In the scope of unavoidable accidents, what is the effect of different ethical frameworks governing self-driving car decision-making, on their acceptance? To exemplify the ethics impact on the acceptance of self-driving cars, one has to consider the situation of an eminent fatal accident involving pedestrians and car passengers. One could argue that innocent passengers ought to be spared, and hence the car passengers should bear the risk of being fatally injured. This most probably would be seen positively by the majority of the people in a city, especially the nondrivers. However, the question that is raised is if anyone would then buy such a car if s/he knows s/he is in high danger; probably not. Subsequently, that may result in a decrease in the sales of self-driving cars, and they will never reach a critical mass. Hence, the envisioned benefits coupled with their existence (e.g., overall reduction of accidents) would also not be materialized as expected.
The ethics embedded in the decisionmaking of a self-driving car, especially in the case of unavoidable accidents, would most probably impact their acceptance by the public. Also, the nature of the ethics, i.e., the ethical framework utilized may also play a role, something that is not sufficiently investigated. In this work quantitative positivist research is carried out, and the empirical data is collected via a questionnaire. With respect to the process followed, first, the ethical frameworks are selected and described. Ethical frameworks are posed in the unavoidable accident context and a model that hypothesizes their link to the acceptance of self-driving cars is proposed. Subsequently, a survey with questions that capture the identified factors (ethical frameworks) is constructed and empirical data is collected. The sampling frame is general, the initial scope is university students (at Master’s level) as they pose a good mix of technology savviness and will be able to easily understand the context in which self-driving cars will have to operate. The following frameworks were selected as representative: Utilitarianism, Deontology, Relativism, Absolutism (monism), and Pluralism. Utilitarianism is a normative ethical framework that considers as the best action, the one that maximizes a utility function by considering the positive and negative consequences of the choices pertaining to the decision.
Deontology is a normative ethical framework and considers that there are rules that have an absolute quality in them, which means that they cannot be overridden. As such, deontologists reject that what matters are the consequences of an action, and focus that what matters is the kind of action to be taken. Ethical Relativism is a meta-ethical framework where it is argued that “all norms, values, and approaches are valid only relative to (i.e., within the domain of) a given culture or group of people”. Hence, in this framework, it is proposed that a society’s practices can be judged only by its own moral standards. Ethical absolutism or ethical monism is a meta-ethical framework that is on the antipodal point of the ethical relativism. This framework, also referred to as “doctrine of unity”, can be described as follows: “There are universally valid moral rules, norms, beliefs, practices, etc. [. . . that] define what is right and good for all at all times and in all places – those that differ are wrong”.
Ethical pluralism is a meta-ethical framework that rejects absolutism (that there is only one correct moral truth) and relativism (that there is no correct moral truth) as unsatisfactory and proposes that there is a plurality of moral truths. It is sometimes referred to as “doctrine of multiplicity”. The ethical pluralist argues that indeed there are universal values (as indicated in absolutism) however, instead of considering that there is only a single set always applicable, it considers that there are many which can be interpreted, understood and applied in diverse contexts (as indicated in ethical relativism).
A closer look at Utilitarianism results shown in Figure 3 reveals that most people consider that an assessment of some kind ought to be done by the self-driving car and be integrated into its decision algorithms.
Deontology implies that there is an expectation that the self-driving cars carry out their duties with good intentions independent of consequences. As seen in Figure 4, the prevalent view is that cars should treat all people on an equal basis (hence not assigning values to individual people as utilitarianism suggests), as well as trying to protect the innocent pedestrians.
Absolutism (monism) propagates the existence of global moral values, norms, beliefs, and practices that are praised by the those who agree while they are condemned by those who disagree. Such views propagate group beliefs and may create tensions in society, as shown in the wide-spread of replies in question A4 in Figure 5, whether life is sacred and knowingly killing people by a machine would be acceptable. As shown in figure 5 there is a strong positioning that the car should have such ethics, and take life and death decisions independently if its owner agrees to it or not. This has several implications, as it would mean that self-driving cars would behave differently than their owners might wish, and raises concerns if cars that do so would actually be bought by people who disagree with their car’s decisions in critical situations.
Relativism affirms tolerance and is bound to culture, time, society, which may ease the acceptance of decisions taken by self-driving cars in critical situations. As shown in Figure 6, people consider that the self-driving car ought to take into account such ethics in its decisions. Such considerations may reflect the diversity of cultures and philosophies found in the world, but may also create “deadlocks” where specific decisions of the self-driving car, cannot be praised or condemned.
Pluralism, propagating the plurality of moral truths, provides a balance among the highly heterogeneous world, tolerance and basic human values such as human rights. Hence, ethical differences may be approached at a global scale. This is also reflected in the views captured in Figure 7, where a mix of aspects is shown, e.g., the owner’s or society’s moral views should be considered, while law and global ethical values are ought also to be respected. Therefore, the pluralism framework is seen as a good candidate for decision-making in self-driving cars. However, due to the multiple perspectives that need to be incorporated, it is also a highly complex one, and hence not easy to realize it.
Finally, the survey also measured some aspects of the self-driving car acceptance as shown in Figure 8, from which it is evident that there is a need for ethics to be embedded in self-driving cars. People seem to trust self-driving cars, and therefore they would opt to buy them once they are available, and may prefer them over the normal (non-self-driving) ones. Overall there is a very strong view, that the society needs self-driving cars, as their benefits for a safer and more inclusive society cannot be overseen. the overall strong support for all frameworks means that there is no clear suggestion, at least from this research, that there should be a preference for a specific framework in the self-driving cars, and no one-size-fits-all solution can be proposed. On the contrary, since all of them seem to have an impact, different parts of the society and people may have different needs and preferences. One thing is clear; that the ethical frameworks considered in this research need to be investigated in-depth, not only qualitatively, but also with mass-scale quantitative surveys as part of the overall research priorities set for AI.
Future directions: It is high time to investigate in detail the ethical angle of issues that pertain to the acceptance of self-driving cars, especially from the diverse viewpoints of the multiple stakeholders involved in their lifecycle. As such, an intersectional analysis pertaining law, society, economy, culture, etc. may be the proper way to move forward and tackle issues raised in this work.
Some challenges are: - will people adjust their road behaviour because of reliance on automation?
- If the ethics of the car conflict with the ethics of the buyer, will they actually buy/use the car?
- Is there bias in learning algorithms for self-driving cars, especially in regard to ethics?
- Should all cars have the same ethical setting?
- How do we stop ‘’hackers’’ from making their own preferential ethical setting?
- How do we stop the fact that it is likely that the cost of the car implies better ethical software?
- How do we tackle privacy concerns?
- Who is liable for the ethical decisions of the car?
- How would two cars with different ethical settings negotiate their outcome?
Differences in Acceptance of Self-driving Cars: A Survey of Perceptions and Attitudes
Introduction: There is a significant body of research around technology acceptance across various domains. Numerous studies have built on to earlier models such as the Technology Acceptance Model (TAM) [1] and the Diffusion of Innovations Theory [2]. In TAM, perceived usefulness and perceived ease-of-use are main factors that affect a user’s attitudes toward using technology, which then influences the user’s behavioral intentions and actual usage, as illustrated in Fig. 1. In the Diffusion of Innovations Theory, five characteristics – relative advantage, compatibility, complexity, trialability and observability – are the key factors that underlie adoption.
Age-related changes in physical and cognitive capabilities, however, can lead to declines in mobility and driving abilities [14, 15], leading many older adults to stop driving altogether. For this reason, they may be the primary beneficiaries of self-driving cars. Older adults, however, have knowledge of and experiences with technology that may differ from younger generations, which may cause them to perceive and accept self-driving cars differently.
While research on technology adoption and transportation safety has begun to explore determinants of acceptance and age effects with regards to new automotive technologies, how different generations perceive and accept self-driving cars is not yet fully understood. In this study, a large-scale survey was conducted to investigate older adults’ perceptions of and attitudes toward self-driving cars, and how their perspectives differ from other generations.
Results: The following factors were significant predictors of self-driving car acceptance: perceived usefulness, affordability, social support, lifestyle fit and conceptual compatibility. Across ages, those who perceived self-driving cars to be more practical, affordable, accepted by peers, and compatible with their lifestyles and conceptual mental models were more interested in getting and using them. Furthermore, attitudinal interest in self-driving cars strongly predicted behavioral intentions to use them.
Age was negatively associated with perceptions, attitudes and behavioral intentions toward the acceptance and use of self-driving cars. Older participants perceived self-driving cars as significantly less useful and more difficult to use compared to younger participants. Older adults were also more likely to think that self-driving cars would be more expensive and more difficult to find where to purchase or access. Older adults indicated that they believed self-driving cars were less likely to be backed up with technical support, less likely to provide emotional benefits, less likely to be approved by their peers, less reliable, less likely to work with other technologies they have, and less likely to fit with their lifestyles and mental models, compared to younger participants. Strong inverse relationships with age were also found for overall level of interest in using a self-driving car and likelihood of purchasing one in the future, indicating that older adults are currently less interested in self-driving cars and less likely to use one when it becomes available. Millennials were most favorable toward the use of self-driving cars. The silent generation (born before 1945) said they were not likely to consider using a self-driving car in any case.
Across ages, however, participants indicated that they would be more likely to use a self-driving car if they were no longer able to drive and less likely to use one if they were capable of driving.
In addition to age, experience with technology in general was strongly associated with self-driving car acceptance. Participants who self-reported greater experience with technology in general and higher confidence in use of new technologies were significantly more interested in self-driving cars and more likely to purchase one in the future. hose who self-reported being more knowledgeable of new technologies were significantly more likely to purchase a self-driving car in the future if they were no longer able to drive. The findings suggest that while self-driving car acceptance varies across generations, as shown in Table 4, age may have an indirect effect on acceptance through experience with technology in general. Additionally, current drivers and non-drivers showed minor differences in their attitudes toward using self-driving cars. Participants who did not have a valid driver’s license were significantly more likely to be interested in using a self-driving car than those who currently had a valid driver’s license. No significant interaction effects were observed between age and possession of a driver’s license.
The Influence of Feelings While Driving Regular Cars on the Perception and Acceptance of Self-Driving Cars
Introduction: Negative emotions that driving may engender in some people have also been found to been connected to a greater likelihood of crashes. Removing human error from driving is one of the greatest potential benefits of self-driving cars, as driver error could be directly or indirectly responsible for as many as 94% of all traffic accidents. The rapidly growing population of older drivers may especially benefit from self-driving cars.
Previous work has found that people’s risk and benefit perceptions as well as trust in the technology are related to its acceptance. . In this research, we specifically examine how people’s feelings around driving traditional cars may affect their perceptions of risk and benefit of and trust in self-driving cars. Further, we investigate how these feelings, perceptions, and trust in turn influence people’s acceptance of the technology.
Laypeople often evaluate the risks and benefits of new technologies differently than experts, and their perceptions of risk are also shaped by their perceptions of benefits the technology may offer. For laypeople, risk perception tends to decrease when benefit perception increases, and vice versa. The characteristics of the technology itself can be captured by two orthogonal dimensions: dread risk and unknown risk. Dread risks include people’s perceptions of the potential for lack of control, catastrophic outcomes, and fatalities. Unknown risks include perceived newness, lack of scientific knowledge, unobservable consequences, and delay of effects. Individual-level factors that affect people’s perceptions of risk include knowledge and affective associations. People’s levels of knowledge about a technology should affect the extent to which they understand both its risks and benefits.
As noted above, affect, in the form of a subtle feeling of positivity or negativity, can serve as a decision heuristic that people use in situations of uncertainty and limited knowledge, known as the affect heuristic. The basis of these feelings is often prior experiences or thoughts related to the decision at hand but it could also be a less relevant emotional state such as current mood.
The nature or valence of the affect plays a role in how it is weighed in judgments. In particular, people tend to attend to or weight negative information or emotions more heavily than positive ones when making evaluations. The affect heuristic also serves as one explanation for the inverse relationship between risk and benefit assessments. If people’s emotional responses are more positive, they tend to judge risks to be lower and benefits to be higher; the more negative people’s affective reactions are, the more likely they are to judge risks to be higher and benefits to be lower. People may be particularly more likely to rely on their affective reactions as a common source to generate both their risk and benefit evaluations when they lack expertise within a given domain.
The affect heuristic suggests that affect shapes people’s willingness to adopt new technologies to the extent that the technology is novel, its performance is uncertain, and its impacts are unknown. Perceived usefulness or the perceived potential benefits has been shown in some empirical work to be a more significant factor in explaining adoption than ease of use. Other factors that have been identified as significant for understanding technology adoption include the relevance of people’s previous experiences (including with the technology) and system reliability—the ability of the system to work without failure. Emotion is also a factor. Further, individual characteristics such as age, gender, lifestyle, and comfort levels with different technologies may also affect people’s willingness to adopt new technologies.
Studies have found that people’s degree of acceptance varies by individual characteristics, with younger, male, or more tech savvy people generally more interested in using self-driving cars than older, female, or less tech savvy people. People’s hesitations around the acceptance of automated vehicle technologies may also be tied to their feelings around driving itself, and many people report driving to be positive for them. For example, in a study that compared all levels of automation (from manual [fully human controlled] to fully automated), participants found manual driving the most enjoyable. Yielding control was a major barrier to adoption of self-driving cars among regular commuters (Howard & Dai, 2013). Because self-driving cars represent a fundamental change in the driving task, people’s current feelings about driving traditional vehicles may shape how they assess changes or alternatives to it.
The present study focuses on how feelings experienced while driving influence risk and benefit perceptions as well as trust in self-driving cars and how, in turn, these perceptions affect the acceptance of these vehicles. We approached this question in an exploratory manner and formulated the following research question: How do feelings related to human-operated driving influence risk and benefit perceptions of, as well as trust in, self-driving cars? Note: For participant and exact questions, see the paper itself.
Results: Higher risk perception was predicted by less experience with vehicle automation technologies, higher levels of positive affect (control), higher levels of negative affect experienced while driving, and being female. Higher benefit perception was related to having fewer years as a driver, greater self-reported knowledge of self-driving cars, more experience with vehicle automation technologies, lower levels of positive affect (control), higher levels of positive affect (enjoyment), higher levels of negative affect, and being male. Trust in self-driving cars was related to having fewer years as a driver, greater self-reported knowledge of self-driving cars, more experience with advanced vehicle technologies, no knowledge of any accidents involving a self-driving car, positive affect (enjoyment) experienced while driving, and being male.
As for interest in using a self-driving car, risk perception, benefit perception and trust were all significant predictors, but benefit perception had the largest effect size among the three.
Discussion: Our results indicate that feelings experienced while driving regular cars inform people’s risk and benefit perceptions of as well as their trust in self-driving cars. We asked about people’s affective experiences driving traditional vehicles—not self-driving cars; nevertheless, people’s feelings about the more familiar driving of current vehicles carried over to their assessments of self-driving cars. Also, one’s attitudes about the status quo should inform perceptions of change to it. People who experienced high levels of negative affect had both higher risk and higher benefit perceptions of self-driving cars. This is contrary to what we would expect from research on the affect heuristic. Because positive affect is associated with more automatic processing, people who have more positive associations with driving may also be less inclined to deliberate about potential risks associated with self-driving cars.
Our results further underscore the significance of benefit perception for understanding technology acceptance. As self-driving cars are still more conceptual than tangible, their usefulness may not be 14 Raue et al. obvious to many, but so too may the risks of such technologies not be fully understood. As the technology continues to mature and becomes more widely adopted, it may be especially important to communicate to the public about its benefits and risks, so that communities can make better decisions about how they want to use and interact with the technology.
Planning
| Week | Task 1 | Task 2 | Task 3 | Task 4 | Objectives (end of the week) |
|---|---|---|---|---|---|
| Week 1 | Choose subject | Make a planning | Collect information | Update the wiki-page | Subject chosen |
| Week 2 | Define research question | Literature research | Concrete planning | Update the wiki-page | Research question specified |
| Week 3 | Literature review | Define subtopics | Literature study | Update the wiki-page | Subtopics defined |
| Week 4 | Make survey | Plan meetings in smaller groups | Write hypothesis | Update the wiki-page | Survey started |
| Week 5 | Send out survey | Contact professors | Switch subtopics | Update the wiki-page | Contact made |
| Week 6 | Analysing survey | Make final report | Write conclusion/recommendation | Update the wiki-page | Final report finished |
| Week 7 | Begin filming the presentation | Edit the film for demonstration | Update the wiki-page | Film for demonstration finsihed | |
| Week 8 | Peer review | Last preparations for demonstration | Finalize the wiki-page | Presentation/demonstration |
Planning per week
Week 1
| Name | Total [h] | Break-down |
|---|---|---|
| Laura Smulders | 8.5 | Meetings [3h], Starting lecture [1h], Research [1h], 5 relevant references [2h], Start/discuss problem statement & objectives [1.5h] |
| Sam Blauwhof | 8.5 | Meetings [3h], Starting lecture [1h], Research [1h], 5 relevant references [2h], Start/discuss Approach, Milestones and deliverables [1.5h] |
| Joris van Aalst | 9 | Meetings [3h], Starting lecture [1h], Research [2h], 5 relevant references [2h], Start/discuss User part [1h] |
| Roel van Gool | 8 | Meetings [3h], Starting lecture [1h], Research [1.5h], 5 relevant references [2h], Check references [0.5h] |
| Roxane Wijnen | 8 | Meetings [3h], Starting lecture [1h], Research [1h], 5 relevant references [2h], Start/discuss user requirements [1h] |
Week 2
| Name | Total [h] | Break-down |
|---|---|---|
| Laura Smulders | 7 | Meetings [3h], Summarize 5 relevant articles [4h] |
| Sam Blauwhof | 7.5 | Meetings [3h], Summarize 5 relevant articles [4.5h] |
| Joris van Aalst | 8 | Meetings [3h], Summarize 5 relevant articles [5h] |
| Roel van Gool | 8 | Meetings [3h], Summarize 5 relevant articles [5h] |
| Roxane Wijnen | 7.5 | Meetings [3h], Summarize 5 relevant articles [4.5h] |
Week 3
| Name | Total [h] | Break-down |
|---|---|---|
| Laura Smulders | 7 | Meetings [3h], Problem statement [3h], Update Wiki [1h] |
| Sam Blauwhof | 7.5 | Meetings [3h], Safety - traffic behaviour [4.5h] |
| Joris van Aalst | 7.5 | Meetings [3h], Perspective of private end-user [4.5h] |
| Roel van Gool | 8 | Meetings [3h], Ethical theories [5h] |
| Roxane Wijnen | 7 | Meetings [3h], Responsibility [4h] |
Week 4
| Name | Total [h] | Break-down |
|---|---|---|
| Laura Smulders | 12.5 | General meetings [2h], Meeting with Sam & Roel [2.5h], Update Wiki [1h], Hypothesis [2h], Planning [1.5h], Literature study [3.5h] |
| Sam Blauwhof | 11 | General meetings [2h], Meeting with Laura & Roel [2.5h], Survey with Joris [2.5h], Literature study [4h] |
| Joris van Aalst | 10.5 | General meetings [2h], Meeting with Roxane [2h], Survey with Sam [2.5h], Literature study [4h] |
| Roel van Gool | 10.5 | General meetings [2h], Meeting with Laura & Sam [2.5h], Research platforms survey [0.5h], Literature study [5.5h] |
| Roxane Wijnen | 8 | General meetings [2h], Meeting with Joris [2h], Literature study [4h] |
Week 5
| Name | Total [h] | Break-down |
|---|---|---|
| Laura Smulders | 9 | General meetings [2h], Meeting with Sam & Roel [2h], Survey with Roxane & Roel [2.5h], Define relevant factors & Literature study [2h], Update Wiki & Planning [0.5h] |
| Sam Blauwhof | 8 | General meetings [2h], Meeting with Laura & Roel [2h], Literature study [4h] |
| Joris van Aalst | 7 | General meetings [2h], Meeting with Roxane [1.5h], Literature study [3.5h] |
| Roel van Gool | 10 | General meetings [2h], Meeting with Laura & Sam [2h], Survey with Roxane & Laura [2.5h], Contact with Lambèr Royakkers [0.5h], Literature study [3h] |
| Roxane Wijnen | 9.5 | General meetings [2h], Meeting with Joris [1.5h], Survey with Laura & Roel [2.5h], Contact with Raymond Cuijpers survey [0.5h], Literature study [3h] |
Week 6
| Name | Total [h] | Break-down |
|---|---|---|
| Laura Smulders | Meetings [], | |
| Sam Blauwhof | Meetings [], | |
| Joris van Aalst | Meetings [], | |
| Roel van Gool | Meetings [], | |
| Roxane Wijnen | Meetings [], |
Week 7
| Name | Total [h] | Break-down |
|---|---|---|
| Laura Smulders | Meetings [], | |
| Sam Blauwhof | Meetings [], | |
| Joris van Aalst | Meetings [], | |
| Roel van Gool | Meetings [], | |
| Roxane Wijnen | Meetings [], |
Week 8
| Name | Total [h] | Break-down |
|---|---|---|
| Laura Smulders | Meetings [], | |
| Sam Blauwhof | Meetings [], | |
| Joris van Aalst | Meetings [], | |
| Roel van Gool | Meetings [], | |
| Roxane Wijnen | Meetings [], |