The Role of Vehicle Diagnostics in Supporting the Law-Abiding “Behavior” of Self-Driving Vehicles ^†

Abstract

The aim of this paper is to look at the pillars for developing the law-abiding “behavior” of self-driving cars. The paper will analyze the potential of artificial intelligence, machine learning mechanisms, and the transformation of rules into algorithms for self-driving cars, and highlight the vehicle diagnostics context. The analysis is expected to demonstrate that the transformation of rules into algorithms alone is not sufficient for unhindered transport, as there is a strong need for the application of vehicle diagnostics to support and enhance obstacle and accident-free transport. The technological revolution in the field of self-driving vehicles calls for the development of common ground between technical sciences and law, which are trying to make the idea of safe and sustainable transport part of everyday life.

1. Introduction

A significant technological and cultural change is that with the proliferation of self-driving vehicles, the aspects of freedom, safety, and “immersion” in transport are being given different emphases. This is not surprising, since driving has long been a way of expressing personal freedom; it is still a valid statement that the car can be understood as a relevant extension of the “ego” [1]. Drivers establish and maintain a real personal relationship with their cars, a special relationship that has increasingly influenced transport trends. In recent years, however, the great sense of freedom associated with driving has been supplanted by the buzzwords safety and ‘immersion’. What matters nowadays is the extent to which a vehicle can provide its passengers with a sense of introspection for the duration of the journey, i.e., the ability to completely shut out the outside world and immerse themselves in their inner, personal world of thoughts. These changes send out a very important message about the evolutionary trends in automobile culture, especially when seen in the context of the increasingly dominant need for automation [2,3]. This demand has been virtually parallel to the mention of sustainable transport since the 2000s. Sustainable transport is an attempt to reconcile the protection of our planet (and everything that exists on it) with fair economic considerations in order to ensure a viable future for all. It is best to think of sustainable transport as a process rather than an outcome—so it can be seen to be incomplete and often needs to focus on problems [4]. The aim of this paper is to highlight a specific set of challenges related to self-driving vehicles and to attempt to outline a legal and technological explanation for the phenomenon. The novelty of the analysis lies in the depth of the mechanisms of machine learning that it reveals, stretching the pillars of ensuring the law-abiding “behavior” of self-driving cars; and, in addition to the legal and technological prerequisites for promoting legal compliance, it shows that there are legal and moral dilemmas and seemingly insoluble situations which, unlike certain micromobility tools also aimed at sustainable transport, prevent self-driving vehicles from becoming active and predictable players in the world of transport.

2. Discussion

2.1. The Relationship Between Self-Driving Cars, AI, and Machine Learning

Self-driving vehicles have been at the core of the concept of sustainable transport for several years now, and their presence would bring undeniable benefits to society. However, despite their benefits, they are perhaps the technological advances in the transport industry that raise the most concerns [5]. The world of transport is characterized by the fact that other social norms take a back seat to legislation. The explanation is the following: various means of transport put people’s lives, physical safety, health, and property at risk, so a legislative environment is certainly needed to ensure predictability and safety in transport. As long as the means of transport are entirely controlled by humans, this safety is more tangible; however, the dilemma is the rise of self-driving vehicles and the teachability of the algorithms that control them and the implementation of these algorithms in the transport system.

Since 2014, the SAE J3016TM taxonomy has been the most cited source on driver automation. SAE J3016 distinguishes between six levels of driving automation [6]. It is essential that a given self-driving car is able of performing several driving automation functions at different levels, but the specific level is determined by the function that is activated in a given situation. The document also points out that there are three actors in driving: the human user, the driving automation system itself, and other vehicle systems and components. The levels of driving automation are determined by the specific role of each of the three primary actors in the execution of the DDT and/or DDT fallback. As a result, up to levels 0–2, the driver is effectively in control of the traffic and her environment, with the automated system taking over for levels 3–5 [7]. It is through these latter levels that traffic situations are created in which the self-driving car’s legal compliance is left to the correct programming by artificial intelligence.

There is no doubt that the “law-abiding training” of these vehicles will be carried out by artificial intelligence. In their monograph, Russel and Norvig identified four philosophical trends in the development of AI and its possible connotations. On the one hand, AI has long been referred to as a system that thinks in a human way (i.e., reflects the workings of the human mind), while on the other hand, they emphasize its role as a system that acts in a human way (following Turing). It often implies a rationally thinking system, expressing that in AI, it is as if it were a more perfect, rational entity than human thinking, not least that it can be a rationally acting system [8].

It is worth drawing the boundaries between artificial intelligence—machine learning—deep learning, as these terms are often used synonymously. AI is an umbrella term and is, therefore, the broadest of the three terms mentioned above; it covers the processes by which a software is able to make decisions automatically [9]. Machine learning is referred to in a narrower sense and includes supervised learning, unsupervised learning, and reinforcement learning mechanisms. Its essential element is the ability to generate autonomous knowledge from acquired experience. This is performed by feeding into the system a set of examples, data, and patterns that enable it to recognize rules, either autonomously or with human intervention, and to make decisions based on them. Deep learning is a variant of machine learning that uses neural networks. These interleave multiple layers of so-called learned databases, which increase the system’s abstraction capacity for decision-making [10]. These concepts shed light on why the automotive industry has such a problem in creating self-driving vehicles. Deep learning and reinforcement learning are of particular relevance to the automotive industry. Deep learning for vehicles is useful in the context of image processing and is also used for object classification in radar signal processing. It is also conceivable to use deep learning to help the system to implement a lane tracking mechanism from the camera image and the steering angle. In the case of reinforcement learning, the agent performs specified reactions in its environment, and consequently the state of the environment to change and return a new state vector to the agent [11].

Machine learning, therefore, enables the self-driving car to perform its tasks automatically, and to demonstrate machine autonomy. The very nature of AI means that it performs tasks not just once, but repeatedly, and, as we have already pointed out, it can adapt to a changing environment—and the diversity of changes is a specific feature of transport. In addition to constantly sensing the environment, the self-driving car evaluates, processes, and analyzes the data from the AI and draws conclusions based on the input of the rules. Of course, the challenges of traffic conditions (e.g., weather conditions, sometimes unpredictable behavior of other road users, etc.) cannot be left out of this process, which will force the AI to re-evaluate its own decisions and consequences, and machine learning can thus ensure the evolution of the system, while preparing it for future challenges.

2.2. Legal Conditions for Ensuring Law-Abiding “Behavior” of Self-Driving Cars

At present, the most intense normative regulatory challenges in the course of technological development are related to changes in vehicles, as the dynamics of legal development are partly generalizing and expanding, and partly status-based; they do not focus primarily on change but make the status quo the defining point of the area. In the current revolutionary state of the automotive industry, this development of standards is only minimally capable of meeting overall regulatory needs and is certainly not a preventive approach that could be followed by machine development strategies but can at best maintain the necessary standardization of processes by means of parallel solutions. In addition to all this, the legislator has formulated the rules of the road in such a way that the human being is the focal point, the alpha and omega—the operation of vehicles depends on her knowledge of the law, experience, and perception, and not least, human life is the primary value to be protected in the creation and application of the relevant rules. In comparison, the following reflections linger on the vision of a future in which self-driving vehicles, driven by algorithms, will be active players in the world of transport, also seeking to demonstrate compliance with the law. What problems might arise?

In the case of law-abiding behavior, it is a question of whether the conduct of a person, whether accidentally (spontaneously) or consciously, corresponds to the act ordered by the legislator. The almost innumerable events of everyday life are precisely such voluntary compliance [12]. From a substantive legal point of view, legal information is not a prerequisite for law-abiding behavior, but a good quality of detailed legal knowledge is a functional necessity for responsible vehicle AI management. The autonomous vehicle comes into contact with this information at two points: during design and production, and during operation, maintenance and repair. The precise knowledge and application of the rules, and the assessment of specific (extreme) traffic situations, is aided and controlled by deep learning; in the case of humans, the brain undoubtedly stores and interprets the law, and combines it with the experience and routine of the driver. In the case of self-driving cars, however, the primary ‘raw material’ is also in question, since the dilemma is: how does the machine learn the legislative environment, which is known to be expressed only in the language used by human beings?

The father of modern legal thought, Leibniz, also realized that law cannot be made perfectly predictable and mathematical because natural language contains uncertainties. In today’s world, laws are written in legal jargon (which derives from natural language), and there is often a need for language to be more precise—but for lawyers (and machines), this is difficult to achieve, and so there is a constant challenge to reconcile ‘paper law’ with ‘living law’. Of course, there are areas of law where efforts have been made to algorithmize language, but these areas remain resistant. The situation is even more complicated if we include in this formula the translation of rules into simple (mathematical) codes—in fact, this is what is needed when we talk about AI-based decision-making [13,14]. In fact, it is a multiple translation: the natural language has to be translated into the language of law, and then the resulting rules have to be transformed into algorithms. At this level of linguistic translation, legislation is frozen for the time being [15]. It should be remembered that social relations, including the application of the law, take place in a discursive space. At present, it is not possible to establish a precise and reliable basis for preparing machines for learning and problem solving, since the very possibility of translating questions of fact and law into machine language is already in doubt [16]. The challenges of the linguistic aspect should also not ignore the fact that law is always built on the specific foundations of a specific legal culture—that is why it is inconceivable with our current tools that it would be possible to create so-called universal algorithms that would express exactly the same content in every self-driving car in every legal system in the world (except for rules that formulate general commands). For the time being, the transport subsystem of our world is tailored to thinking humans, not thinking machines.

2.3. Technological Conditions for Ensuring Law-Abiding “Behavior” of Self-Driving Cars

From a technological point of view, the “behavior” of an autonomous vehicle is a technical system controlled by remote access and requiring intervention capabilities. From this point of view, the “law-abiding behavior” of the vehicle is nothing other than the alignment of intervention and control capabilities with the legal regulatory environment and its continuous changes. The standardization of the self-driving vehicle is implemented in the technique of the operation process, since legal information must be interpreted within the control operation; therefore, the controllability and accountability of this information maintenance (the technical-executive capability for legal application) must also be regulated. It is clear from all this that in order to ensure compliance with the law in the case of self-driving cars, the technological aspect is just as important as the legal aspect. The legal aspect alone, i.e., the compliance of the vehicle’s ‘behavior’ with legal requirements, is not sufficient; it is necessary to ensure that the mechanisms described are interwoven with a network of vehicle diagnostics, so that all road users can be assured of safe and efficient transport.

Automotive diagnostics is primarily a concrete way to access information on the general technical condition of a vehicle. In the context of machine learning, a self-driving vehicle uses cameras, sensors, radars, and LIDAR to measure and gather information about its environment, while also assessing the state of its own internal systems. This allows any faults or inconsistencies to be detected almost immediately—by processing data and information in real-time, the self-driving vehicle can modify its operation without delay, or, if necessary, signal to the occupant that intervention is required. In this process, it should be seen that when designing self-driving technology, developers are mindful to be proactive in drawing attention to faults that need to be avoided, and even the vehicle may show sensitivity to very early signals and warnings. The protection of human life is a priority in this respect, but the cost savings for the owner, user, or operator of the vehicle through continuous monitoring are not negligible [17].

These solutions will help to ensure that the self-driving vehicle does not go astray, so that possible inconsistencies and errors do not compromise the ability to make legally compliant decisions. Self-driving vehicle diagnostics can also enable systems to be integrated, so that the flow of information can be direct and fast, eliminating the risk of unexpected situations and accidents. The diagnostic data obtained during operation are not only controlled by the vehicle itself, but are also transmitted to the manufacturers, who can extract a lot of useful information from the data obtained. Such benefits can include the ability to identify avenues for further improvements or, where appropriate, to detect when a software update is needed. The downside of diagnostic data is not negligible: the amount of information present may be sensitive data, so data protection is intertwined with the compliance aspect. Last but not least, the legal and technological conditions for law-abiding conduct should be mentioned in connection with a problem that neither the law nor the automotive industry can be said to be sufficient to solve, either alone or in combination—and this sheds light on why self-driving vehicles are somewhat overshadowed in the concept of sustainable transport compared to micromobility devices. The vision of the legal and technological community is for self-driving vehicles to make the safest possible decisions, even in extreme or crisis situations. Unfortunately, life has proven time and time again that even in the world of human-powered and human-operated vehicles, accidents involving human lives do occur. This experience has led researchers in recent years to engage in philosophical thought experiments that have called for the formulation of so-called moral algorithms. There is undoubtedly a moral dilemma about the choice to be made by the vehicle in the event of an unavoidable accident [18]; no system of standards gives you the right to kill people, but what should programmers do, what pre-generated priority should they teach the vehicle? In this context, moral philosophy treatises talk about the so-called problem of value choice, and illustrate this with the so-called trolley dilemma (not originally invented for self-driving vehicle technology) [19]. So, until the advent of self-driving vehicle technology, no system has yet faced the problem of how to make the right decisions when neither law enforcement nor complex vehicle diagnostic systems provide reassuring guidance—and this is the grey area of the introduction of self-driving cars [20]. The question is whether the future will be able to solve this serious dilemma, and whether we have sufficient legal and technological tools to transform future transport to this extent.

3. Conclusions

The world of transport, with all its norms, operates in such a way that the final decisions are always made by people—transport is, therefore, a human-dominated domain. Self-driving vehicle technology is turning this situation quite upside down, with the idea that no human intervention is needed to operate, ‘behave’, or ‘make decisions’ in self-driving cars. At the same time, transport requires that the actions of the actors must be carried out in accordance with the requirements of the law, and in the world of transport, the protection of human life is always a primary value. For drivers, the pillars of law-abiding driving can be identified: willingness to follow the norm (not compulsory, as law-abiding behavior can occur spontaneously), driving routine, and perception. The “behavior” of self-driving vehicles is provided by AI through machine learning, but the question is: can traffic rules be translated into the language of “codes”? In the view of this study, no, because law cannot be mathematised (although the need to do so has long been raised among legal scholars). Traffic safety is ensured not only by behavior that complies with the law, but also by a number of technological conditions; here, vehicle diagnostics nodes in particular are an important support for designers and programmers. In technological terms, there are indeed solutions available for diagnostic achievements that help to avoid or eliminate accident situations. However, neither the legal nor the technological conditions take account of the unfortunate situations that occur in transport and are also experienced by human operators: unpredictable, tragic accidents can happen, and the solution to these situations can be found by programmers ‘in advance’ using algorithms. The big question is: is anyone entitled to encode value choices into algorithms? This paper argues that this liminal situation is currently a kind of dead-end for self-driving vehicle technology, and a harbinger of the unknown world that is commonly referred to as the black box of AI. It cannot be stressed enough, the dogmatic sciences, including jurisprudence, include so-called non-computable cases, for which no algorithm can provide the answer. The world of transport cannot be imagined without man, and even if there are further developments in AI, they will only be in support of him, i.e., a cooperation between man and machines in this field, where man remains at the center of everything; the control (decision-maker) of transport means itself requires human intervention, since all its decisions are decisions about human destinies.