Punishing autonomous machines? The example of autonomous cars

Introduction 3

1. Definitions
1.1 Machine
1.2 Autonomy
1.2.1 Proper autonomy
1.2.2 Improper autonomy
1.3 Punishment

2. Autonomous cars in the automobile industry
2.1 State of the art technology
2.2 Expected trends and innovations

3. Punishment of autonomous cars
3.1 Possibilities and reasons for a punishment
3.2 Requirements for a punishment
3.3 Complications regarding to the punishment of autonomous cars

Conclusion

Bibliography

Introduction

According to the statistics of the Federal Office, 86 percent of road-traffic accidents are caused by mistakes made by the car driver (cf. Lemmer 2014, 8). The creation of autonomous cars could solve this problem! Exploring the potential of autonomous cars is of particular interest to the military, in order to make it possible to drive in critical situations without human passengers. Military vehicles do not play a role in this essay due to the ethical implications which they evoke. Nevertheless, military exploration has provided a lot of knowledge to the current position of autonomous machines (cf. Christaller and Knoll 2003, 7f.). Should it be possible to create autonomous machines and to integrate them into society, rather than reconsidering the relationship between man and machine? Furthermore, the moral aspect of autonomous machines should not be forgotten (cf. Bendel 2013 c, cf. Bendel 2014).

This essay pays attention to one moral aspect of autonomous machines, namely punishment. In this context, we will talk about the punishment of autonomous machines and this will be shown using the example of autonomous cars. The aim of this essay is to demonstrate how autonomous machines could be punished, or if it is really possible for them to be punished. The basis for these ideas is taken from the philosopher and business information technologist Oliver Bendel, the dissertation Automaten als Träger von Rechten. Plädoyer für eine Gesetzesänderung by Andreas Matthias, the essays Killer robots from Robert Sparrow and Why you Can’t Make a Computer that Feels Pain by Daniel Denett as well as the anthology about autonomous machines by Thomas Christaller and Josef Wehner.

The central terms are defined in chapter 1. The important terms are: machine, autonomy and punishment, which will be introduced in chapter 1.1 to 1.3. Chapter 2 shows the latest knowledge in the automobile industry regarding autonomous cars and the expectations for the future of autonomous cars. Chapter 3 takes a more pragmatic approach. In section 3.1 the possibilities and reasons for punishment are shown. 3.2 summarizes the requirements for such punishment and Section 3.3 discusses the problems which arise from the application of punishment methods to autonomous cars.

In the conclusion the most important aspects are summarized. Finally, an outlook towards the problems arising from integrating autonomous cars into the traffic is given.

1. Definitions

The terms machine, autonomy and punishment are important to this essay. Consequently, this chapter provides the definitions that are used within this essay.

1.1 Machine

The term machine is a loan word of the latin machina and this word was adapted from greek mhxanh. The German language adopts this term from the French language in the 17th century. A machine could be (1) an apparatus, which is able to do exercises autonomously or under control, or (2) an apparatus, which makes life easier, for example an iron, or (3) all driving vehicles, which have a motor or a mechanism. A machine is often complexly constructed and is destined to convert energy and to work with it (cf. Florent 2013, 647; Robert 2013, 983). Technically, a machine is a gadget with movable and non-movable parts, which transforms force or carries out working stages independently (working machine) or which transforms energy to another form (force machine).

1.2 Autonomy

Evidence suggests that the term autonomy was already in use during the Antiquity. The Greek terms autos (self) and nomos (law) would mean to be able to give yourself laws / orders. This kind of thinking is also represented in the today’s use of the noun autonomy. In the past, the semantic was only a political category, where political bodies are able to give themselves laws / orders. Sturma describes that Sophocles for example, used this term for the intern independence from individual. From this point of view Antigone acted after her own law / order and the other women did not do so. Thanks to Rousseau and Kant, this term earned a higher meaning in the moral philosophy and should be separated from heteronomy and anomie (cf. Wahrig-Burfeind 2011, 216; Sturma 2003, 39; Kolmer and Wildfeuer 2011, 254f.).

Who or what is able to be autonomous? A piece of wood could not be described as autonomous. If we threw it into running water it would get washed away. What the piece of wood “does” is dependent on its environment. It does not decide its “own actions”. On the other hand, a refrigerator is semi-autonomous, because it responds to environmental influences and it incorporates its “inner state” onto its “decisions” by weighing up whether it should keep its engine working, depending on prevailing inside and outside temperatures. The semi-autonomous refrigerator also weighs up whether its reactions are helpful or not. The more autonomous the system, the more it may involve conditions of use and the avoidance of damage in its actions (cf. Dorner 2003, 112; Pfeifer 2003, 143). That is why autonomy is a gradual characteristic. Autonomy is separated into proper autonomy and improper autonomy in this essay. The grades of autonomy are defined in the following two sections.

1.2.1 Proper autonomy

The semantic content of proper autonomy is that individuals, not necessarily people, are able to give themselves orders and rules and that they work with their positions and actions.

“Ein typischer Fall personaler Autonomie sind Lebenspläne, in denen Werthierarchien, Bewertungsperspektiven sowie Wünsche und Interessen zweiter Stufe zum Ausdruck kommen.” (Quote Sturma 2003, 51).^[1]

An individual is able to act self-determined and free manner. However, it is not possible to recognize from an outside view if this person used his autonomy.

The most important fact is that self-determined acts must have a cause and that they do not occur by pure chance. This implies that the actor is responsible for his actions. Outside pressure, influences and coincidence are not allowed to influence these actions. Autonomy also demands to use its resistance. If the final point is not valid then the actor is not autonomous (cf. Sturma 2003, 49; Kolmer and Wildfeuer 2011, 257ff.).

Kant also added a moral reasonability to this term because personal legislation is defeated by the categorical imperative: Act only according to that maxim whereby you can, at the same time, will that it should become a universal law (cf. Sturma 2003, 39). This means that the first person adopts the third person’s moral point of view. It is no longer just about what should be done, but what should be done as a person. Only a person who has the aforementioned characteristics is able to do this. Influences of third persons can also be a danger to autonomy. This can be shown in the experiments of Asch (1951). He presented on the basis of simple fixed length, that group opinions can be influenced by manipulating members (cf. Roughley 2002, 167; Sturma 2003, 39f.; Kolmer and Wilfeuer 2011, 256; Asch 1951, 177 – 190).

1.2.2 Improper autonomy

We can see that there are many variations of the term autonomy. Self-determination belongs more to the background and in the foreground are softer formulations to automatic processes from self-movement or metaphorical interpretations of the self­concept ( cf. Sturma 2003, 49). This semantic transformation of the term ‘autonomy’ is visible in mathematics, artificial intelligence and robotics. In this case, we can talk about improper autonomy. Improper autonomy exists also when we have induced grades of liberty:

The term of autonomy applies to the characteristic of a machine, which is defined in domains of regulations and movements where it acts without a control for the whole time. (Translation of Sturma 2003, 50).

This implies that actions are predictable, because they are actions with programmed aims. But it is not predictable if the autonomous machine tackles with the aim itself and how others would think about it or if it considers to do this action or not. However the system works with changing environment areas and uses behaviours which are not constructed directly. The machine carries out actions and acts within its environment, that’s why it makes sense to talk about autonomous machines (cf. Sturma 2003, 50f.; Christaller and Wehner 2003, 25; Christaller and Knoll 2003, 52).

1.3 Punishment

The term punishment originates from the latin punitio as well as sanctio. The first term is an act of punishment which is used for mistakes, disregard of rules, an order, a sanction. The second term is about a restrictive measure (cf. Florent 2013, 900 + 985).

In ethical dimensions, punishment and reward are two of the most important factors of (human) action, which can have a significant effect on the future. At the same time it may exist as self-punishment. A punishment can be carried out in two ways: On the one side, adding an aversive appeal, a slap on the face or to complete hours of community service, may be a direct punishment. Withdrawing a positive appeal, for example, for a horse lover the prohibition to ride a horse, provides means of indirect punishment. Results of researches have shown that contingent punishment can cause an accelerated reduction of punished behaviour and in some cases, its disappearance as well. It should not be forgotten that punishment should take place in a defined and closed timeframe. Punishment causes sorrow / suffering for the punished person (cf. Becker and Hinz 2012, 596 + 599; Tenorth and Tippelt 2007, 696; Böhm 2005, 615). If the socialisation process does not work, a “post-education” with action-related punishments, like the production of an objective balancing or expiation, could help to get the socialisation process back (cf. Becker and Hinz 2012, 598).

Rewards could negatively influence autonomy because the individual is influenced in a manner that provokes dependence as well as affecting self-esteem. When punishments are too high, self-esteem can decrease and it could cause undesired generalisations which would destroy the relational level or the force of the punishment to be underestimated (cf. Becker and Hinz 2012, 597).

2. Autonomous cars in the automobile industry

Is the term autonomous machine a contradiction or not? As it is above-mentioned, autonomy means to be free of pressure and laws. To be autonomous means to have the capability to follow your own will and your own laws, which you have imposed on your own. But machines are fully determined systems:

Autonomy and heteronomy, independence and dependence, liberty and pressure and free man and slave or rather human and machine are dichotomic accompanied. (Trad.: Rammert 2003, 293).

In the following sections, an overview about state of the art technology and the expectations about self-depended and possibly autonomous cars, will be given. Regarding to this development, it will be proved, if the term autonomous car can be accepted.

2.1 State of the art technology

We can find the first ideas of autonomous cars in 1918, which was a time before the necessary computers and sensors were invented. An autonomous car is a very difficult concept. There are many projects concerning this domain; some of them are Google and Sartre, in which companies like BMW and Audi are involved (cf. Rojas 2013, 88; Bendel 2014). In this domain, cars not only are objects, which are used by humans; they also appear as subjects:

“Die uns bedienen und benutzen, die uns Vorschläge [...] unterbreiten und für uns weitreichende Entscheidungen treffen, und die teilautonom oder autonom sind.” (Bendel 2014)^[2]

An example could be a navigation system, which proposes different routes, thereby showing the status of the route and also pays attention to weather conditions.

Nowadays, cars with assistant services are being created, which are able to drive independently in a steered column and can park with the help of a computer-controlled motor. They are able to stop automatically in cases of danger or traffic jams and they automatically turn on the headlights at night time or in a tunnel. The German named “Lidarsystem” is able to read the distance to a range of hundred meters. The measurements, however, are not known exactly and are only accurate up to a few centimetres. In 2004/2005 the DARPA (Defense Advanced Research Projects Agency) participated at the Grand Challenge in the desert. They have shown in 2007 at the DARPA challenge that autonomous driving is possible. In this competition, cars had to show that they are able to drive in real driving conditions, such as motorways and country roads. The Urban Challenge in 2007, has also demonstrated some complications: One of it was that the sensors of different cars had disturbed the other cars. That’s why no car has started at the beginning (cf. Bendel 2013 a, Rojas 2013, 90f., Berger and Rumpe 2008, 260f.).

[...]

^[1] In English it means a typical case for personal autonomy are life plans in which hierarchies of values, perspectives of evaluation, as well as wishes and interests of the second degree are expressed.

^[2] Subjects which serve us and use us, offer suggestions and make critical decisions, and are partially or fully autonomous.