Microsoft Word - 6-V9N1(2024)-AITI#12038(65-84).docx


Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 

English language proofreader: Chih-Wei Chang 

Challenges and Solutions to Criminal Liability for the Actions of 

Robots and AI 

Vladimír Smejkal1,*, Jindřich Kodl2 

1Department of Informatics, Faculty of Business and Management, Brno University of Technology, Brno, Czech Republic 

2Authorized expert of cryptology and information systems security, Prague, Czech Republic 

Received 20 April 2023; received in revised form 17 September 2023; accepted 23 September 2023 

DOI: https://doi.org/10.46604/aiti.2023.12038 
 

Abstract 

Civil liability legislation is currently being developed, but little attention has been paid to the issue of criminal 

liability for the actions of robots. The study describes the generations of robots and points out the concerns about 

robots’ autonomy. The more autonomy robots obtain, the greater capacity they have for self-learning, yet the more 

difficulty in proving the failure foreseeability when designing and whether culpability or the elements of a specific 

crime can be considered. In this study, the tort liability depending on the category of robots is described, and the 

possible solutions are analyzed. It is shown that there is no need to introduce new criminal law constructions, but to 

focus on the process of proof. Instead of changing the legal system, it is necessary to create the most detailed audit 

trail telling about the robot’s actions and surroundings or to have a digital twin of the robot. 

 
Keywords: robot, criminal liability, audit trail, criminal evidence 

 

1. Introduction 

Concerning robotization and the growing capabilities of robots, a novel phenomenon is emerging. It could be described 

as self-will, self-initiative, or the independent, autonomous decision-making of robotic systems, which deviates from the 

conservative position of non-thinking. Modern robotic systems are more sophisticated, and mere mechanisms are fully subject 

to the will and instructions of humans. Some robots currently find themselves in the new role of what is perhaps best described 

as a self-acting, i.e., autonomous, machine. This new aspect will undoubtedly be reflected in law. The authors, therefore, focus 

on the potential criminal liability for the unlawful consequences of robots’ actions and how to prove it. 

Legislation on civil liability is currently in the works. However, scarce attention has been paid to issues of criminal 

liability for the actions of robots. In the European Union (EU), for example, two pieces of EU legislation have been published 

which may impress the officials with the issue of liability for the actions of artificial intelligence (AI) in their entirety, i.e., the 

Artificial Intelligence Act (AI Act) and the AI Liability Directive. In reality, they only concerned with the strict liability of 

manufacturers for defective products, the rules for access to information, and the burden mitigation of proof concerning damage 

caused by AI. 

Specifically, the AI Act [1] provides for (a) harmonized rules for the placement on the market and AI systems service for 

their use in the EU; (b) a prohibition on certain AI practices; (c) scrupulous requirements for high-risk AI systems and 

obligations for operators. Meanwhile, the AI Liability Directive [2] builds on the AI Act, simplifying the legal process for 

victims to prove someone else’s fault caused the harm. 

 
* Corresponding author. E-mail address: smejkal@znalci.cz  



 Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 66 

In the area of liability for damage caused by a product defect, i.e., in the area of civil liability, liability is conceived as an 

objective liability. If proven that the damage was caused by the injured party, or if it can be reasonably assumed considering 

all circumstances that the defect did not exist when placed on the market or occurred later, the obliged person has the possibility 

of liberation. In the area of criminal liability, it is unlikely to set out some form of strict liability in law because the basic 

principles of criminal liability include the criminality of the conduct, the consequence, and the causation between the conduct 

and the consequence, i.e., proving the culpability of a perpetrator or group of persons. 

This paper does not construct AI and robots as entities equipped with criminal liability but describes the possibilities of 

addressing it within traditional legal systems. The authors disagree with claims that AI could prove to be an exceptional 

opportunity to change law and legal theory [3-4]. The most common tort cases are likely to emerge from poor programming. 

This will certainly be true until robots reprogram themselves. Attempts to hold manufacturers, programmers, and users on 

legal theories such as negligence or product liability under the criminal code predominate over the current level of “weak” AI, 

while the view of recognizing direct criminal liability for AI itself predominantly supports the “strong” AI in the future [5]. 

The difference between “strong” and “weak” AI is that “weak” AI aspires only to model the mind when solving partial 

and specialized problems, while “strong” AI is recognized as a universal model of human thinking in a software way, i.e., a 

replication of the human mind in a computer [6-7]. 

Authors need to prepare for this future. However, in the authors’ opinion, it is neither by creating robots as fictitious 

persons endowed with legal responsibility, nor does it make sense given the purpose of punishment involving retribution, 

incapacitation, general deterrence, and rehabilitation of the offender [8], or even the death penalty [9]. Once humans have 

decision-making authority over robots, they can simply deactivate or dismantle the robot. Nevertheless, it does make sense for 

a person - natural or legal - who was responsible for the robot or found to be at fault. They see the introduction of corporate 

criminal liability, as it is conceived in many countries and has long historical roots, as an appropriate way forward [10-11]. 

These prosecuted legal persons may primarily be the manufacturer or operator of the robot with shared culpability, which is 

also an option [12]. 

In recent years, the use of robots has been rapidly expanding in various sectors. Particularly, industrial production 

(welding, assembly, material handling, etc.), construction and maintenance of infrastructure, healthcare (robotic surgery and 

diagnostics), logistics and warehousing, transport (driving vehicles, autonomous vehicles, trains, or planes), agriculture, 

domestic and accommodation services, and the military (military robots). The increasing numbers and use raise the likelihood 

of incidents related to robots or their malfunctioning. The purpose of this study is to explore whether existing criminal law 

tools are sufficient to successfully address issues related to criminal liability for the actions of robots and AI without the need 

to create new constructs. 

2. The Term “Robot” and Categories of Robots 

A broader perception of the term “robot” has emerged than before. In the 20th century, robots were perceived as devices 

performing physical work (the word “robot” was derived in 1921 by the Czech writers Čapek brothers from the word “robota” 

which means “work”). It was only much later that “software robots” implementing robotic process automation (RPA) emerged, 

which can be used for any process that is repetitive according to certain rules - from sending emails based on certain criteria 

to high-frequency trading (HFT). Both types of robots can then be controlled using AI. The following text categorizes only 

robots in the classical sense, i.e., physically intervening in the environment, with special attention to autonomous vehicles. 

2.1.   Definition of robot 

The technical nature of the robot is constantly evolving from simple manipulators to autonomous systems, where the 

robot can perform the intended tasks based on the current state and sensor data without human intervention, moving in its 



Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 67

environment, and performing the intended tasks while comprising a control system and control system interface. The robot’s 

actions must have either direct or indirect responses in the physical world (not necessarily by direct mechanical manipulation, 

but also by passing a command to another system having an interface to the physical world, e.g., water supply network control 

or road traffic control). 

Software-only robots will be understood and classified as computer programs, which are not addressed in this paper. A 

robot, therefore, differs from information systems in its ability to directly interact with its environment physically. Furthermore, 

for a long time, the term robot (except household robots) was largely understood as an “industrial robot”, even though the 

word “industrial,” i.e., a system designed to perform simple, typically repetitive mechanical operations, may now seem too 

finite as robots are making inroads into significantly “non-industrial” areas, such as medicine.   

Robots can be fully or partially autonomous, or they can be remotely controlled (by a program in the cloud and/or by a 

human), often probably both in some predictable synergy. The degree of a robot’s autonomy is still being determined by the 

human, either a priori based on legislation or in a specific situation by the person in control, where the human operator is 

always part of the control loop of such a robot’s program. 

An example is the autopilot in an aircraft, in which, after setting certain conditions, the pilot hands over control of the 

aircraft, but oversees its operation and can intervene in this robot’s work at any time – correct its control of the aircraft, or 

switch it off completely. However, the conditions under which the pilot may hand over control of the aircraft to this system 

are defined in the aviation regulations and the aircraft manufacturer’s regulations by specific flight conditions. 

2.2.   Generations of robots 

Robots can be divided into five generations according to the level of their ability to make autonomous decisions: 

(1) The zero generation includes manipulators and robots usually without feedback, where any malfunctions or changes in 

the monitored areas (signaled by sensors) result in the next step being disallowed and the system being stopped (the so-

called “central stop”) while the maintenance staff is called. 

(2) The first generation includes robots with simple feedback capable of switching between several deterministically operating 

subprograms (developed in advance by a human) and working.  

(3) The second generation includes robots with optimization capability, which is the ability to select the optimal program 

from predefined programs based on a specified criterion, i.e., the precise rule governing the decision about the next known 

action. 

(4) The third generation is characterized by robots capable of independently modifying the original program (action plan) 

with a posteriori knowledge. Here, only the goal of the activity (task) is predefined, while the method of achieving the 

goal is left to the intelligence of the control system, which itself creates an action plan consisting of successive steps and 

activities to achieve the given goal. An action plan is a sequence of robots starting in an environment, described either 

numerically or symbolically – by logical statements that the robot interprets per se to achieve the defined global goal [13]. 

The formulation of the plan then consists of finding a way in the state space from the current state to the target state.  

(5) The fourth generation is represented by autonomous robots with social, human-like behavior, which means they choose 

the goals of individual tasks independently based on an appropriate global criterion, e.g., the principle of long-term 

existence/autonomy of such a system (survival, energy saving, etc.). 

Since the third generation, determining the cause of undesirable behavior is likely to be a problem and any search for legal 

liability will have an unparalleled difficulty. 



 Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 68 

More or less intelligent robots are becoming useful autonomous elements in complex industrial production sections. Their 

ability to make independent decisions, and select the most appropriate tasks will increase, as their will upon autonomous 

reasoning [14]. There are alarming predictions that AI will reach human capabilities by 2029, and humans and machines will 

gradually converge, reaching the point called the singularity in 2045. It is often debated whether robots will eventually 

dominate humans to take full control, i.e., it’s estimated that Kurzweil’s Singularity will occur in 2045 [15]. Similar 

catastrophic predictions have been published by renowned figures such as Steven Hawking, Bill Gates, and Elon Musk. In 

January 2015, they jointly signed an open letter on AI with other AI experts, and the letter was to call for research into the 

social impact of AI to prevent some potential “pitfalls” of AI, which is said to have the potential to eradicate disease and 

poverty, but scientists must not create something that cannot be controlled. The open letter entitled “Research Priorities for 

Robust and Beneficial Artificial Intelligence” details the research priorities in an accompanying 12-page document [16]. 

Some authors insist that evolution will always be under human control. However, robots do not yet have the consciousness 

and self-awareness, which is a necessary precondition for taking over control and the estimated point of reaching the singularity 

because we are approaching 2045, continuing to be postponed in scientific works [17]. There is also the question as to whether 

robots will ever become self-aware because the necessary prerequisites are the natural emotions of machines, the ability to be 

aware of their position in the world models, legal and moral rules, power/priorities, as well as communication-based on these 

models with other robots in the community, especially to plan actions towards a certain goal that the robots themselves have 

set [18]. Unless humans enable robots to set their own goals or activate themselves, the singularity is expected not to happen 

at all [19]. Nonetheless, the existing third-generation robots may already pose a problem. 

2.3.   Self-driving cars 

Some evident examples of semi-autonomous robots except autopilots are so-called autonomous (self-driving) vehicles. 

As the decisions are left to the robot’s control system, we will be more interested in: 

(a) Setting the decision algorithms so that their operation or failure does not violate life, health, or property. According to the 

authors, the tentatively correct solution is to set priorities corresponding to a human driver. Nevertheless, it varies 

especially with a cooperative strategy. Moreover, in this context, it is indispensable to mention the principle of necessity 

- an act that would otherwise be a criminal offense is not a criminal offense if a person commits such an act to avert an 

imminent threat to an interest protected by criminal law. 

(b) The ability of the driver to take control of the vehicle whenever needed. In other words, a pilot can switch off or “override” 

the autopilot. 

(c) The degree of protection of the control system against negative influences from the environment, either intentional 

(hacking) or unintentional (electromagnetic interference or operational failure), external mechanical influences (coming 

from the environment in which the vehicle is moving), and mechanical interference (by third parties), starting with 

maintenance work or an attack on the vehicle as a movable asset, which the control system should also be able to approach 

or warn at least. 

(d) Self-documentation capabilities, the ability to create an audit trail of the vehicle’s operation (similar to the existence of 

“black boxes” in aircraft), are required. 

Further development is expected to lead to a cooperative strategy that will further optimize the behavior of such systems 

working in a group (group intelligence), which will be applied by individual autonomous vehicles among themselves – the so-

called “cooperative driving.” Real-time vehicle-to-vehicle (V2V) and vehicle-to-traffic infrastructure communication will 

improve the information available to each road user. This will make almost any active involvement of the driver unnecessary 

and undesirable in many cases. The point is that cooperation between vehicles, or between vehicles and their environment, can 



Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 69

incur the perplexity of information for the drivers to process in real-time. Automated driving will therefore be effective where 

the driver would no longer be able to process the massive information provided in the required time, and even following the 

recommendations would be insufficient. The driver will thus be de facto (not de jure) useless except the situations where the 

driver must be able to return the vehicle to its original, less complex mode and to take over the control if, for example, the 

cooperating network collapses (e.g., due to a lightning strike). 

3. Liability for the Robot’s Actions 

To search for who is responsible for a crime committed concerning the use of a robot, we must apply the peculiarities of 

the robotic world to classical schemes of criminal responsibility, especially when resulting in the involvement of AI. Initially 

explicit, deterministic processes can change into unpredictable and non-deterministic processes. It is therefore necessary to 

discuss whether the traditional principles of criminal law will hold up under the new conditions. 

3.1.   A new dimension of liability 

The new dimension of liability will be based on the presence of a relatively simple line of action and consequence 

concerning potential liability for perhaps all machines and equipment falling under the zero, first, and second generation that 

has been used hitherto. Whether it is the design or manufacture of these machines, which starts with a project where parameters 

are always verifiable. Therefore, the relatively simple way of subsequent checks is whether any important fact has been omitted 

in the design. The same applies to the manufacturing process assuming that the design is flawless on the material, the question 

is whether this process was followed or whether, for some reason, a deviation emerged, which may be related to the ensuing 

unlawful consequence. The next stage is the actual use of the final product. the utilization is regulated by law if it has 

complexity and the potential to harm. Typically, this applies to all means of transport where the legislation explicitly provides 

the means of transport can be operated on public roads.  

In other cases, only liability for any unlawful consequences is regulated, such as liability for damage caused by an 

operational activity. In addition, many technical regulations and technical standards defining product characteristics are 

rendered. However, the question lies in the obligation to apply, whether the presence or absence of regulations and standards. 

Despite the propriety because of the duty to take preventive measures, it will be up for discussion here.  

Nonetheless, it is no longer the case with third- and fourth-generation robots. Meanwhile, the regulation of the use, 

possible liability for damage, and harm resulting from robot failure will be substantial. In the case of a consequence contrary 

to the purpose and intent of the machine or device, the behavior and its compliance with the law, possible fault as a subjective 

element of a possible criminal offense, and the causal relationship between this behavior and the consequence could be 

analyzed, however, several questions will arise as: to what extent this failure could have been foreseen in the design of the 

robot, whether any act occurred at this stage or perhaps during the manufacture, which could be causally linked to the 

consequence and thus whether it is possible to establish fault or the elements of a particular criminal offense were accomplished.  

Nevertheless, it will always be necessary to consider the absence of determining beyond doubt the resulting behavior 

under all possible circumstances in the simple case of a deterministic behavior given the random (unexpected) input signals 

and data (observations of the surrounding world). It creates a considerable space for technical and legal analyses focused on 

the foreseeability of errors in the design and programming of the robot, the prevention with appropriate modifications, and the 

category of force majeure. 

3.2.   Tort liability in robotics 

Presumably, the existence and mass use of robots after the third generation will not change the basic paradigm. If the 

main or sole cause of injury is the failure of a machine, it is important from a criminal law perspective whether or not the 



 Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 70 

failure is based on fault. It is well known that civil law also recognizes a strict liability for a consequence, but that is not what 

is at the focus of the analysis. It is necessary to analyze what can be regarded as “fault” within the meaning of the Criminal 

Code, i.e., the involved fault and is the moment an event that has occurred through no fault of any individual or a legal person, 

for instance, a failure or an accident happened through nobody’s fault but an unfortunate coincidence.  

This bipolar scheme of fault vs. mischance could certainly apply to robots, insofar as we simplify the problem, disregard 

the actions of persons not criminally liable, and introduce actions in circumstances that preclude criminal liability (for instance, 

tolerable risk). But the problem lies in the distinction, which is onerous to determine a robot’s autonomy and the ability to 

learn or self-program. Specifically, learning is simply setting the parameters of a fixed system, situation classifier, neural 

network, production system, etc., whereas self-programming can significantly modify the decision-making principles 

influencing a robot’s actions. 

(a) Undoubtedly, robots or robotic systems performing identically repetitive operations under controlled external conditions 

(typically industrial, assembly robots/manipulators) have been employed for many years. Manipulators on a production 

line, unless someone directly steps into their operation field, are virtually harmless concerning the arguments below. 

(b) Some robots operate in environments with variable external circumstances, i.e., environments with uncertainty. These 

circumstances include the possibility of a collision with a human, another robot, or some other movable or immovable 

asset. Typically, these are robotic cars, various autonomous mobile logistics robots, etc. Nevertheless, these are predefined 

and predictable robots despite the broad operating range. In other words, everything operable has been programmed by 

humans. 

(c) Autonomous systems, self-learning robots, intelligent robots, hybrid robots, etc., are complex devices. Meanwhile, they 

are sometimes regarded as a combination of a “living” (biological structure) and “non-living” component, whose reactions 

and procedures are unpredictable due to self-programming within this self-learning framework or the adoption of behavior 

patterns on a person or a group of persons. However, as the degree of autonomy increases, we know less about the 

“intracerebral” situation of such a device or the state before a certain incident.     

The limit of criminal liability for all three categories is the definition of negligence in Section 16 Criminal Code of the 

Czech Republic, Law No. 40/2009 Coll:  

(1) A crime is committed by negligence if the perpetrator (a) knew they could violate or endanger an interest protected by the 

Criminal Code in the manner specified in this Code but relied without reasonable grounds on not causing such violation 

or endangerment, or (b) was unaware their conduct may cause such violation or endangerment, although they should and 

could have been aware of this because of both overall and personal circumstances. 

(2) A crime is committed by gross negligence if the perpetrator’s attitude towards the requirement of due caution shows a 

manifest disregard for the interests protected by the Criminal Code. 

If intent, gross negligence, and wilful negligence are excluded. It is relatively easy to determine for category (a) machines 

what the person operating the robot should have been known despite the alleged ignorance due to the feasibility of predicting 

what can be expected from the robot. 

The situation is more complicated for category (b). Here, it may be difficult to establish what the person operating the 

robot should have known for their actions to be considered as a fault in the form of conscious negligence. For example, to 

what extent can the driver rely on the system controlled by active radar to stop his car in front of an obstacle when it normally 

does so, and this is clearly stated in the instructions? Moreover, postulating the car is equipped with a fault indicator for the 

system, is reliance on the system a valid reason not to have your foot on the brake when the car approaches an obstacle? Or, 

what is even more dangerous, when someone suddenly steps into the road?   



Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 71

The use of virtually autonomous robots as discussed in category (c) will pose even greater problems. For this group, it 

will presumably be difficult to prove intent. For example, if a “thinking” robot builds another robot instead of lifting loads and 

loading the person operating the robot and loads into a crusher. It will be the technical cause of the robot’s failure primarily 

investigated. On the other hand, however, similarly to any other industrial accident, it will also be necessary to investigate 

whether the operator or the manufacturer (of the original robot) implemented possible prevention of such failure. Preventing 

a “thinking” robot from producing a defective product instead of a functional robot is a typical example, which in turn means 

finding out what the “thinking” robot produced and, more importantly, why.  

Generally, there are two types of liability: liability under the Civil Code and liability under the Criminal Code. Meanwhile, 

liability for misdemeanors and administrative offenses related to robots can also be added. For criminal liability, there is the 

requirement of intentional fault unless the Criminal Code expressly provides sufficient negligent fault. Concerning intent, a 

distinction is made between direct and oblique intent, where the perpetrator is cognizant of unlawful consequences or 

understanding (the understanding also means the perpetrator accepts or reconciles himself to the fact that he may violate or 

endanger an interest protected by law as described in the criminal law). 

It will be relevant whether the potential perpetrator has been obliged to behave in a certain way by law, specific regulations, 

or even internal rules when operating the robot. In the context of robots, these may generally be such regulations as the Labour 

Code in the area of occupational health and safety, the protection of public utilities, or the EU regulations on medical devices 

as indicated in Regulation (EU) 2017/745. Moreover, whether the person has acted intentionally or negligently in breach of 

their duties, which may be qualified as intention, negligence, or omission. In this context, the ruling of the Supreme Court of 

the Czech Republic on 12 October 2017, file ref. 6 Tdo 1062/2017-28 seems to be relevant: 

“A breach of an important duty within the meaning of Section 143(2) of the Criminal Code cannot be understood as a 

breach of any regulation, but only of such duty laid down therein, where a breach of such duty usually leads to a threat to life 

and health of humans, if such a consequence can easily occur and often does occur due to such breach. In assessing whether 

a duty is an important duty, the key criterion is to consider the consequences of breaching a particular duty and the likelihood 

of those consequences occurring. For an omission to amount to an act, it must be the omission of a specific duty arising from 

the specific position of the wrongdoer, that is a situation in which society expects and relies on the actions of the specific 

person. First of all, the person who has a specific duty to act must be in a specific relationship to the interest protected by law. 

If this specific duty to act cannot be inferred from the position of the wrongdoer, the act as a condition for criminal liability is 

absent.”  

It is in contrast to the consideration of reasonable or tolerable risk. Even in everyday life, we cannot avoid situations in 

which the occurrence of harm cannot be entirely excluded. Therefore, it is necessary to consider in each case to what extent 

the occurrence of a harmful consequence could have been foreseen in the first place and what measures have been or could 

have been taken to minimize any harmful consequences. According to Section 31(2) of the Criminal Code of the Czech 

Republic, there is no tolerable risk if an activity endangers the life or health of a person without their consent with the activity 

being given following other legal regulations, or if the result seeks to achieve is manifestly disproportionate to the level of risk, 

or if the performance of the activity is contrary to the requirements of other legislation, the public interest, principles of 

humanity or good morals. 

The assumptions will thus be based on the statement above to ensure a robot is understood as a system. In addition to the 

purely mechanical components enabling the robot to do something, the robot mainly consists of computer hardware and 

software. Despite not being biologically alive, it is capable of self-learning through experience and interaction and autonomous 

thanks to sensors or data exchange with the environment. Beyond that, it can adapt its actions and activities to the environment. 

In other words, the robot’s external behavior is determined by the primary internal configuration (the initial program developed 



 Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 72 

by humans). On the other hand, with the subsequent modification of this program by the robot, various steps were presented 

as parameterization, creation of a database of patterns (models, heuristics), and at a higher level by the modification of the 

initial neural network based on the aforementioned interactions with the environment, i.e., what is closest to the human concept 

of “learning.”  

In such a case, it may be extremely difficult to establish a causal link between the failure of the robot and the actions of 

manufacturers and programmers. It is only after this causal link has been established (proven) beyond any doubt that the 

potential fault of an individual or a legal person can be examined regarding potential criminal liability. 

3.3.   Negligent fault in robotics 

It can be assumed that negligent crimes will predominate in robotics, although the “reprogramming” of a robot into a 

“killer” cannot be ruled out. Concerning criminal negligence, two basic categories of perpetrators must be distinguished. The 

first group includes perpetrators knowing their conduct (omission) could cause harmful consequences but relying without 

reasonable grounds on not causing such consequences. The second group includes perpetrators not knowing their conduct 

(omission) could cause harmful consequences, in which case the condition for criminal liability is the fact that they should or 

could have known this by considering the circumstances and personal situation (in particular their education, profession, 

position held, etc.).  

Given the desire of perpetrators to make quick money is endless, it is possible to envisage offenses committed with gross 

negligence concerning robots when the perpetrator’s attitude towards the requirement of due caution shows a manifest 

disregard for the interests protected by criminal law. Without a volition element (for criminal negligence), the law defines 

negligent fault through an intellectual component. Regarding conscious negligence, the perpetrator is cognizant of the 

possibility of the relevant criminal consequence. On the other hand, as for unconscious negligence, the perpetrator is not 

cognizant of this possibility. The criterion of negligence is the exercise of the required degree of caution, which could be either 

general (required of everyone) or specific, or higher (for instance, in the performance of certain activities or professions) – see 

e.g., Award of the Constitutional Court of the Czech Republic of 31 May 2016, file ref. III. ÚS 2065/15-2 or ruling of the 

Supreme Court of the Czech Republic of 27 June 2012, file ref. 5 Tdo 540/2012.  

The term “the required degree of caution” emerges here. This is a vague legal concept with a wide range of possible 

interpretations. The determination of the required (minimum) degree of caution could be sought in specific regulations such 

as technical regulations, technical standards, or the generally accepted rules resulting from the level of knowledge in a 

particular field as shown in the ruling of the Supreme Court of the Czech Republic of 30 January 2014, file ref. 6 Tdo 1450/2013 

concerning reasonable caution.  

Based on the antecedent statements, something similar is defined as the state of the art in the Inventions and Improvements 

Act, which comprises everything made available to the public using a written or oral description before the date from which 

the applicant is entitled to claim priority. If a robot commits an unlawful act due to the outdated design, the situation is similar 

to a non-lege artis medical practice.   

3.4.   Causality in robotics 

As mentioned above, it will undoubtedly be difficult to establish and appropriately prove the causal link between the 

behavior and consequence. In medico-legal disputes, where the initial input and the outcomes are known, the actual course is 

entirely unclear, and the presence of the so-called black box phenomenon has emerged [20]. Predominantly, it is about fault 

(usually unconscious negligence) and the predictability of the outcome. The causal link is usually undisputed unless there are 

multiple factors at work.  



Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 73

If a robot of the third and especially the fourth category (as described above) fails significantly, finding a causal chain 

will often be a major problem with an uncertain outcome. Here we may not always know all the factors having influenced the 

functioning of an autonomous mechanism. We may know its initial settings (or what they should have been if there had been 

no design or manufacturing error), but the question is whether we can find out all the information that influenced the robot’s 

operation. Besides, although currently, it is still easier to do than to find out all the processes taking place in the human body, 

the examination of the “black box” may not be successful or even feasible under certain circumstances (in the case of a biorobot) 

if we consider it to be the robot’s hardware and software. 

Another problem may be potentially multicausal, i.e., there may be concurrent causes or accumulation of several possible 

causes from adversity or unlawful event, from which it will be necessary to select one cause as the decisive cause in the given 

case. This may be the simultaneous combination of input data (i.e., the robot’s environment), the program that evaluates the 

data, and hardware (e.g., a mechanical hand) that performs the movement that gives rise to the unlawful act. One possibility 

might be to express the percentage rates of these influences if possible. 

In this regard, it seems appropriate to quote the ruling of the Czech Supreme Court of 27 February 2002, file ref. 3 Tz 

317/2001, according to the statement “the causal link between the act of the perpetrator and the consequence is not broken if, 

in addition to the act of the perpetrator, there is another fact which contributes to the occurrence of the consequence, but the 

act of the perpetrator constitutes the fact without which the consequence would not have occurred.”  

An example would be a robot’s motion controller not anticipating that the robot’s arm might go into a certain position 

simply because it is programmed to exclude that position. If someone forces the robot’s arm into such a position, thereby 

affecting the functionality and safety of the robot in such a way that it subsequently moves the arm into a different and 

unpredictable position causing injury, the programmers cannot be held responsible because this effect would not have occurred 

in the first place without the intervention in question. Given that, the authors can conclude one of the main requirements for 

robots: the maximum creation of an audit trail telling about every step the robot has taken and its internal state. 

 
Fig. 1 Factors influencing the behavior of self-learning robots [own picture of the authors] 



 Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 74 

Fig. 1 shows all the factors that can affect the current version of the robot control program. The initial design of the robot 

and its setup originates with the developer and manufacturer. An empty programming tool (e.g., a neural network) learns based 

on the knowledge and patterns chosen by the manufacturer, but the user or the AI itself may also be involved. The behavior of 

the AI system is not only dependent on the outcome of the learning process but also the robot’s environment (sensor 

information) including the behavior of other robots (e.g., in autonomous cars). The specific state of the system can therefore 

change expeditiously. 

3.5.   Robot software and tort liability 

One of the main problems with robotics is the current unavailability of software with a guarantee of reliability and 

invulnerability. The well-known Murphy’s Laws of programming state attest that “There is no program that is completely free 

of bugs, and that each time you remove a bug, you introduce another, hidden and more insidious bug into the program. This 

means that it is not possible to create a program that is completely free of bugs. Any software product will always contain 

some bugs.” 

An example leading to criminal liability could be a programming error with a certain combination of parameters (e.g., the 

position of the robot, its arm, load, etc.), causing an unpredictable state or behavior that can impact the environment, e.g., a 

mechanical attack on surrounding objects. Similar cases are known from the past in the field of space exploration, where 

rockets or satellites had to be destroyed due to software errors (sometimes directly in the program logic, sometimes due to a 

“simple” typing error). 

Primarily, it seems possible to consider the liability of the author(s) of the program, who in the least “did not know that 

their conduct was likely to cause such an infringement or endangerment, although they should and could have known it, having 

regard to the circumstances and their situation.” This brings us back to the question of fault. 

In this respect, the Constitutional Court’s ruling is relevant according to the statement “the limits of the circumstances 

that the perpetrator can or cannot foresee cannot be defined only hypothetically (for then everyone would have to foresee 

essentially everything), but must always be based on the existing objective circumstances resulting from a particular life 

situation, which can be characterized by a variety of factors that the perpetrator perceives with his senses and can then evaluate 

them according to his knowledge and other subjective dispositions. In terms of negligent fault (Section 5 of the Criminal Code), 

this means that, in addition to the degree of caution required by the general rules of safe conduct, there is also a subjective 

component, which consists of the degree of caution that the perpetrator can exercise in a particular case. There can only be 

fault by negligence when, at the same time, duty and the possibility of foreseeing an injury or threat to an interest protected by 

criminal law are present [21].” However, the question is whether it was within human power to foresee the occurrence of such 

coincidence (e.g., in the input data) to which the software should have been able to respond. 

There may be various causes for robots’ unpredictable behavior, ranging from hardware defects to software bugs. 

Common causes include:  

(1) Hardware defects: Robots are made up of sensors, actuators, and control systems, which can fail and cause unpredictable 

behavior.  

(2) Software bugs: Robots rely on complex software systems, and bugs in the code can cause unexpected behavior. 

(3) Incorrectly calibrated sensors: If the robot’s sensors are not correctly calibrated, it can lead to incorrect readings and 

unpredictable behavior.  

(4) Environmental interference: Robots can be affected by external factors such as electromagnetic interference, which can 

cause unpredictable behavior.  



Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 75

(5) Insufficient testing: If a robot has not been thoroughly tested, it may behave unpredictably in real-world scenarios. The 

quality of testing is the key to minimizing the risk. It is appropriate to proceed by decomposing the system components 

into individual functionalities that can be described by simple finite automata whose behavior can be comprehensively 

analyzed to a large extent.  

To address unpredictable robot behavior, it is important to diagnose the root cause through thorough testing and debugging. 

In some cases, hardware components may need to be replaced, and commonly the software may need to be updated or 

completely rewritten as well. 

In this respect, it is probably appropriate to start from the principle of “required (necessary, reasonable) degree of caution,” 

which should be embedded in the principle of “secure by design” by the robot manufacturer, which means that the designed 

product (which could be understood as the whole robot or just its software) is designed to be secure from the very beginning 

of its development and in all its stages [22]. The problem is that neither the hardware nor the software can be completely free 

of all errors with 100% probability. Additionally, by the requirement for reliability of communication with a system reliability 

of 99.9999%, the downtime of a robot can be 31.5 seconds, which is incredibly long!  

Most texts dealing with safety and information systems or robots emphasize that a necessary step to increase safety and 

thus reduce the probability of an operational accident and possible subsequent liability is the existence of a risk management 

system as shown in ISO/IEC 20000 IT service management. When assessing specific cases, it will likely be a question of fact 

and subsequently of law, as to whether the risk was normal or extraordinary, predictable or unpredictable, and what measures 

were taken to minimize it. 

Whether the risk was tolerable is primarily a matter for the court. However, when assessing risk in robotics, it is very 

likely that not only the current state of knowledge will have to be considered, but in the event of a leapfrog change in technology, 

which is beyond the scope of that state of knowledge, leading an entirely new situation will have to be assessed. If a risk 

analysis has been carried out covering all foreseeable harmful situations, if the risk of an activity causing harm has been ruled 

out, and if it is clear that the intended (and legal) goal cannot be achieved in any other way, then it is admissible to move on 

to actions involving elements of risk and consequently compliance with these conditions which should be considered when 

assessing the circumstances precluding the unlawfulness. 

The specific examples that possibly happen will vary depending on the degree of autonomy of the robot and other 

circumstances. There may be an error in the program that makes the robot’s actions not only unauthorized but also a threat to 

an interest protected by the Criminal Code. Such an error is unlikely impossible, but the error and functionality of a safety 

mechanism stopping the robot before the incident will be crucial. One of the criteria for assessing potential liability is whether 

functional testing has been performed before starting the manufacture of a robot. 

3.6.   Other factors affecting robots’ operation 

It is necessary to point out other factors affecting robots’ actions. Every program works with data, while some can be 

entered as constants by the manufacturer, variables by the user, and collected by the robot itself through sensors. Therefore, 

even if the program is working perfectly, the robot may act incorrectly in an unauthorized way, and it will probably be easy to 

find out what has happened only in the first two cases. If a given case involves the effects of external factors whether 

unintentional or intentional, we will depend on the possibility of analyzing the processes inside the robot. 

This is typically true of self-learning robots, which are autonomous systems that can improve their performance over time 

based on experience, trial, and error. Their behavior can be explained using basic algorithms and models that enable them to 

process data, making decisions, and learn experientially. They are sophisticated systems that can adapt and improve over time, 

which enables them to perform increasingly complex tasks and respond to changing circumstances in real-world environments. 



 Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 76 

One common approach to self-learning in robots is reinforcement learning, which is a type of machine learning involving 

an agent (robot) performing actions in the environment to maximize a reward signal. The robot learns through trial and error 

and takes actions possibly leading to a higher reward and improving its overall performance over time. 

Another approach to self-learning in robots is unsupervised learning, which means that the robot analyses large amounts 

of data and discovers patterns or relationships on its own without explicit instructions or labeled data. It facilitates the robot 

learning about its environment and predicting future events. In both cases, the behavior of self-learning robots is determined 

by their algorithms and the data (datasets) on which they have been trained. The quality of their decisions and the speed at 

which they learn depend on factors such as the complexity of their models, the amount of data to which they have been exposed, 

and the quality of the algorithms used to process that data. 

The behavior of these robots is determined by a combination of several factors, including: 

(1) Sensors: Self-learning robots typically use various sensors to sense their environment such as cameras, microphones, and 

touch sensors. These sensors feed data to the robot’s learning algorithms, which use this information to make decisions 

and perform tasks. 

(2) Algorithms: The behavior of self-learning robots is determined by algorithms controlling their decision-making process. 

These algorithms may include artificial neural networks, reinforcement learning, evolutionary algorithms, etc. 

(3) Goals and tasks: Self-learning robots are designed to meet specific goals and perform tasks. For example, a self-driving 

car may be designed to navigate roads safely and avoid collisions, while a service robot may be designed to help people 

with various tasks. The robot’s behavior is customized by these goals and tasks, as well as its algorithms and sensory 

inputs. 

(4) Experience: As the self-learning robot interacts with its environment and performs tasks, it collects data and gains 

experience to improve its future performance. For example, a self-driving car can learn to better navigate roads after 

encountering challenging driving scenarios. 

In summary, the behavior of self-learning robots is a complex interplay of their sensors, algorithms, goals, tasks, and 

experience. A robot may also behave contrary to its intended purpose or programming due to an internal defect (hardware 

error, but more likely software error) or an external factor, while both may or may not be relevant concerning a third party’s 

fault. 

Unintentional influence can occur through a combination of external signals picked up by the robot’s misevaluated sensors 

as a result of an incident. A question of assessment lies in whether the situation was both unforeseen and unforeseeable to such 

an extent of not having been predicted by the manufacturer. If the analysis excludes that it is not an unpredicted and 

unforeseeable situation and the failure of the robot emerged due to reasons per se (defect in design, program, etc.), the situation 

should be assumed predictable to cause partially or completely by an external influence including the intervention of another 

person. If a human factor is found to be such an influence, it will need to be addressed (in addition to finding the particular 

person) the issue of intent versus negligence. One can imagine the negligent action of someone who inadvertently or entails 

another activity changes the physical configuration of a robot that begins to move along a different path.  

In the case of intent, this may be an external attack where the attacker attempts to influence the processes taking place 

within the robot enough to change its functions or functioning and subsequently cause an incident. The person who can 

influence the robot to act to endanger an interest protected by the Criminal Code can be anyone: the owner (user, operator, 

etc.) who (wilfully or unwilfully) enters incorrect data into the robot – then it will probably be about assessing possible 

negligent conduct in an unlawful act, but perhaps even about assessing whether the elements of the criminal offense of 

“Damage to a record in a computer system and on an information carrier and negligent interference with computer equipment” 



Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 77

have been accomplished. In the case of another third party (a hacker) who can attack the program that controls the robot (but 

of course also the data) to make the robot do something not under its intended purpose, the situation is clearer, i.e., this will 

constitute a misuse of a thing (using the robot as an object or instrument of a criminal offense). 

As part of the assessment of a particular case, it will be necessary to determine which of the robot’s actions can be 

considered the result of the negligent (or intentional) conduct of an individual. In the case of zero to second-generation robots, 

the robot can be perceived as a deterministic automaton (DA), where all its states are known detectable or inferable, i.e., it is 

clear at any point of robots’ actions and rationales chronologically. For the third generation of robots, where we already admit 

the robot’s ability to independently create a program experientially, the situation is already more complex because this gives 

the robot the character of a non-deterministic automaton (NDA), where it can reach a certain state through multiple paths states 

simultaneously.  

Possibly by the theory of computer science, we will be able to convert an NDA into a DA and establish the reasons for 

the robot’s behavior. Nonetheless, it is not universally applicable. At the outset, such a robot will always have the same initial 

state, but once it has been trained, it will depend on many independent factors as the internal state diverges after an incident. 

The original instruction set and dataset used would leave the robot as a deterministic mechanism without further learning 

(training) of the model. Nevertheless, the application of different behavioral patterns with different outcomes will translate 

into the robot’s architecture differently each time, depending on the patterns and models used. 

In the fourth generation, when we talk about autonomous mechanisms having their intelligence, the algorithmic 

complexity (variability) inside the robot will be incredibly high that it cannot be ruled out that it will be impossible to determine 

the exact events and rationales inside from a certain point in time. AI tools such as neural networks or fuzzy logic, evolutionary 

programming, and genetic programming, are moving us from a world where we can tell the causation to a probabilistic one. 

This is a probabilistic prediction, where no one can guarantee that a given system will evolve and behave as predicted. Then 

we get into the area referred to as NP-hard problems, where it will be necessary to look for approximate or partial solutions 

giving a satisfactory answer at least in some cases. It is a question, however, whether this approach is applicable in finding 

criminal evidence. 

Owners (operators) of robots whose actions may endanger the safety of the public in a malfunction (due to their weight, 

nature of the operation, etc.) or whose operation may be labeled as “particularly dangerous,” which may also be held criminally 

liable for any accident caused by the robot. Their tort liability for negligent conduct may be at least twofold: 

(1) They have acted contrary to the manufacturer’s instructions (manual), e.g., overloading the robot instead of performing 

proper maintenance. 

(2) They failed to exercise the required degree of caution given the characteristics of the robot. 

In this context, it is also possible to mention the potential liability of the robot owner in the failure to perform the mandatory 

software update released by the manufacturer.  

Švestka and Smejkal dealt with a similar issue much earlier concerning the updates of antivirus software. “In the case of 

computers, there is no legal obligation for their operators to use antivirus software, let alone to update it. However, even if we 

were to admit this, which of course could only be done by way of law (we note, though, that this is a vision that is difficult to 

imagine), then the determination of the liability for damage must rest on proof beyond reasonable doubt that the owner 

(operator, user), or which one of them, has committed the specific misconduct. If a hacker gains access to a computer and 

subsequently causes damage to another party – whether directly or by using someone else’s computer – it will be very difficult 

to prove in practice whether the fault can be attributed to the manufacturer of the operating system, to an application program 

purchased by the user, or whether the misuse of the computer was caused by neglecting to update a “protection” program, etc 

[23].”  



 Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 78 

The authors believe this is still true for a robot. However, it could happen that the obligation to update the software would 

be stipulated in the contract or the technical conditions for the use of the robot, i.e., in documents binding on the owner. The 

owner would probably be liable for the damage caused under civil liability and such conduct could also be considered 

negligence, perhaps even gross negligence from a criminal point of view. 

3.7.   Legal persons as potential perpetrators of robotic crimes 

For completeness, it is also necessary to mention the potential criminal liability of legal persons. This liability arises from 

the fact that not only an unlawful act committed in the interests of a legal person or the course of its business activities by 

persons such as managers but also employees or persons in a similar position, may be attributed to the legal person. However, 

even if an unlawful act committed by the above persons can be attributed to a legal person, the legal person may still be released 

from criminal liability if he has made all the efforts that could reasonably prevent the commission of the unlawful act by its 

managers or employees. 

This construction is also decisive in the assessment of whether the legal person is liable for the unlawful consequence of 

the robot’s actions. Regarding the accidents caused by robots, the foremost question will be whether the liability of an 

individual for the accident can be established, or whether criminal liability is involved as the case may be. However, it should 

be noted that it may not always be a question of finding a specific individual or specific individual identified as having 

committed an unlawful act attributable to a legal person. Consequently, the criminal liability of the legal person is not affected. 

Thus, the mechanism for establishing criminal liability of a legal person for an accident caused by a robot is more complex 

because it does not end with the identification of the liable individual but continues with the assessment of whether the 

individual acted in circumstances in which their unlawful act is attributable to the legal person. If the answer is affirmative, it 

only remains to consider whether the legal person has exempted himself from criminal liability by considering the efforts to 

prevent the commission of the unlawful act by the persons. The problem of robots was brought back to the question of the 

extent of potential foreseeability of a robot’s reaction in a certain situation and the extent of the liability to be inferred for lack 

of relevant foresight. It is imaginable that the criminal liability of a legal person, a manufacturer of a third or fourth-generation 

robot, and the employees did not foresee the robot’s harmful behavior due to the insufficient understanding of the robots’ 

design. And the legal person, their employer, failed to address this professional deficiency. More loftily, the legal person as a 

manufacturer may have been technically capable of creating a monster but was unable to oversee the capability of this monster. 

4. Prevention and Evidence 

Actions of both robots that will act directly, i.e., causing mechanical movements, and robots that will act only based on 

information such as robots trading on the stock market to consequently be capable of manifesting as negative or directly 

contrary to the law. Two problems emerge as follows: 

(1) How to prevent or at least minimize the likelihood that such actions will occur? – prevention. 

(2) How to establish the reasons leading to the negative actions including the liability of a particular individual or legal person? 

– evidence.   

The situation will be quite different for deterministic robots, where it is possible to justify any time why the robot acted 

as it did. If the contents of its memory are available, the computer program and data will be located despite the absent possibility 

to predict in advance in many cases what values the data will take. In other than deterministic processes, i.e., where the robot 

itself decides how to react to the stimulus, the situation will be more complex. In these cases, there will be a wide range of 

possibilities, from the simplest systems to the most complex self-learning ones, which will be able to modify their code based 

on processes that cannot be predicted to a greater or lesser extent. 



Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 79

4.1.   Prevention 

The problem referred to in point 1 above will need to be addressed by a risk analysis exploring all the possible states that 

the robot can get into and try to set up defense mechanisms to prevent negative actions from occurring. The higher level of 

autonomy of the system cooperating with other systems is a typical example for autonomous vehicles. Undoubtedly, we have 

reached a point where the level of protective measures may prevent the full use of the system. 

Minimizing the risks associated with the robot’s operation involves a multi-step process, which includes the following:  

(1) Risk assessment: The first step is to assess the potential risks associated with the robot’s operation. This involves 

identifying the types of hazards the robot may encounter with the potential consequences of these hazards. This may 

include simulations, testing, and other methods to evaluate the safety of the robot in different scenarios. 

(2) Safety design: The design of the robot should include safety features minimizing the risk of harm to humans, animals, 

and the environment. For example, if the robot is designed to work near humans, it should be trained to respond 

appropriately to sudden movements or other unexpected events. This may include physical barriers to prevent the robot 

from coming into contact with people or other objects, sensors to detect potential hazards, and emergency stop buttons 

to quickly shut down the robot during an emergency. Robots should be operated in a controlled environment and be 

subject to strict operating procedures to minimize the risk of accidents. 

(3) Secure communication: Robots should use secure protocols for communication, such as encrypted communication 

channels to prevent unauthorized access or tampering with control and data transmission. 

(4) Authentication and authorization: Robots should have mechanisms to securely authenticate the identity of users and 

devices to control access to sensitive data and functions based on roles and permissions. 

(5) Data protection: Robots should have built-in mechanisms to protect sensitive data, such as encryption and secure storage 

to prevent unauthorized access or theft of data. 

(6) Software security: Robots should be developed using secure software development practices such as code reviews and 

security testing to avoid vulnerabilities and security flaws in the software. 

(7) Physical security: Robots should be designed to be physically secure, e.g., by tamper-proof covers. 

(8) Testing and verification: Robots should be thoroughly tested and verified to ensure the intended function and proper 

safety features. This may include testing the robot in different scenarios and environments to identify potential risks. 

(9) Training and learning: Users should be trained and instructed on how to operate the robot safely and effectively. This 

may include instructions on how to use safety features and what to do in an emergency. 

(10) Monitoring and maintenance: Regular monitoring and maintenance of the robot is important to ensure its continued safe 

and efficient operation. This may include checking the robot’s sensors and safety features, updating software and 

firmware, and replacing worn or damaged parts. 

(11) Incident reporting and response: Incidents involving robots should be reported and investigated to determine the cause 

and prevention of similar incidents in the future. A plan should be responded to potential incidents including procedures 

for emergency shutdown, evacuation, and medical assistance. 

Following these steps will help minimize the risks associated with operating the robot and ensure the safety of animals, and 

the environment. 



 Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 80 

4.2.   Evidence 

Regarding point 2, this is a question of obtaining enough information to determine the behavioral rationale. It means 

collecting a huge amount of data is not dissimilar to collecting data in the so-called black boxes found in airplanes at least in 

the form of monitoring the robot’s last moments of operation. However, even this may not provide a clear answer to the 

question of who is to blame for the robot’s negative actions, especially in cases where the robot will be interacting with the 

environment. It is important to note that sometimes the cause of the actions may be due to a combination of factors and require 

a combination of the above approaches to fully understand and resolve the problem. 

Typically, an autonomous vehicle is surrounded by a very heterogeneous environment consisting of other vehicles, other 

entities, and objects located in the environment, but it will also receive information from a variety of sources, ranging from 

cooperating vehicles and traffic control-related signals. However, meanwhile, these can be false signals, ranging from exterior 

interfering signals of unintentional origin to intentional attacks. 

 
Fig. 2 Information about the incident applicable to criminal proceedings [own picture of the authors] 

Fig. 2 illustrates a description of the pre-trial criminal procedure, which consists of securing clues about the robot’s 

activities. The activities include the analysis of technical documentation, the interview with witnesses, documenting the crime 

scene, the examination of evidence to resolve any inconsistencies, the evaluation of both individual and collective evidence, 

and finally the assessment of someone’s possible culpability. 

4.3.   Digital twins 

A new phenomenon that has emerged in connection with Industry 4.0 is called “digital twins,” where each physical 

element has its virtual representation, where its behavior and interactions with the environment are simulated by a software 

module. “A digital twin is a virtual version of a physical entity, whether product, factory, or some other type of asset or system. 

The digital twin unites business, contextual, and sensor data to represent the physical object [24].” Software modules, 

representing physical elements in a virtual space, collaboratively solve tasks, coordinate their activities, and make decisions 

using services they provide to each other or call up through the Internet of Services (IoS). The concept originated in products 

and then in machines and entire production lines [25]. 

According to the research report named Digital Twins in Smart Cities and Urban Modelling by analyst firm ABI Research, 

digital twins will become a substantial tool in city modeling and optimizing the operation of smart cities [26]. Conceivably, 

digital twins could be used to analyze the causes of a delinquent robot’s actions, whereby individual variants could be modeled 

to determine the most likely course of action. Furthermore, virtualization could also be used to digitally recreate the event. 



Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 81

 
Fig. 3 Digital twins [27] 

Fig. 3 illustrates the properties of a digital twin as a software representation of a physical asset. It enables us to model all 

possible states of the robot in the framework of prediction and analyze the behavior of the robot based on sensed variables in 

real operation. 

In a previous article, it was written that “the first corpse doesn’t count.” In other words, it cannot be blamed on the 

manufacturer who is likely to be the main entity liable for the robot’s actions due to the impossibility to foresee all states, 

situations, and influences. Black swans will always exist. On the other hand, if that manufacturer does not learn from all 

previous experience, i.e., if it fails to take into account the current state of knowledge of science and technology, then it could 

be liable at least for negligence. 

5. Results and Discussion 

From this perspective, the bottom line is that the robot either  

(a) Operates in a predicted and desirable way, i.e., it operates correctly. 

(b) Operates in a predicted but undesirable way, i.e., it operates incorrectly. 

(c) Operates unpredictably. 

In the cases referred to in (b) and (c), it is critical whether damage or non-material harm is caused in this way. If so, it is 

important from a criminal law perspective whether an interest protected by criminal law has been violated in the antecedent 

manner and, if so, who is liable for this.  

In the case of the predicted undesirable operation of a robot, it is critical to know how such undesirable behavior occurred, 

whether it was predictable, and why measures were not taken to eliminate such operation. In the case of unpredicted robot 

operation, the same determinants will need to be examined, i.e., whether measures were taken to prevent or minimize the 

undesirable and unpredicted operation. 

Generally, liability for illegal or criminal acts by robots or AI systems is a complex issue that has not yet been fully 

resolved by the legal system. In some cases, the manufacturer of the robot may be liable if it is proven that the robot was 

defectively designed or manufactured. In other cases, the owner of the robot may be liable if the robot has been used illegally 

such as having been programmed to commit criminal offenses. However, there may also be the factor that the programming 



 Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 82 

or data used to train the UI system, i.e., the programmer or data provider could also be held liable. Meanwhile, the liability of 

regulatory authorities cannot be ruled out either due to the absence of clear regulations or technical standards for the 

manufacture and/or safe use of the robot, while regulatory authorities may be held liable for failing to properly supervise the 

technology. 

Ultimately, considering both general and specific circumstances, the design and capabilities, and applicable laws and 

regulations, liability for the actions of robots is likely to be determined on a case-by-case basis. However, as robots become 

more advanced and autonomous, it becomes increasingly difficult to determine who should be primarily liable for their actions. 

In Article 12, the European Parliament resolution on 16 February 2017 [28] states that: “it should always be possible to 

supply the rationale behind any decision taken with the aid of AI that can have a substantive impact on one or more persons’ 

lives.” The European Parliament “considers that it must always be possible to reduce the AI system’s computations to a form 

comprehensible by humans with considering that advanced robots should be equipped with a ‘black box’ which records data 

on every transaction carried out by the machine including the logic that contributed to its decisions.” The problem will be 

whether the neural brains are artificially or biologically equipped in truly advanced robots. 

As a result, we will be able to translate all the robot’s decisions into a comprehensible form and the fully identifiable logic 

of its decisions. The black box originating from the aviation and containing all the necessary information can serve as an audit 

trail representing significant if not being the main or only evidence in determining the place, robot’s state, and the causation 

of the incident. Given that, one can conclude the question posed at the beginning of this article: was it an intentional fault, 

negligence, accident, or a unforeseeable event – a Black Swan [29]? 

There is a need for preparations in both the area of legal theory and practice reflected in the legislation and decision-

making practice, the area of technology reflected in the elaboration of standards and the creation of “best practices,” and in the 

area of criminology reflected in the emergence of new practices or sub-areas within the field of criminology. The application 

of methodologies and tools such as big data, AI, and digital twins could increase recall and thus help reduce entropy in specific 

robot-related crime investigations. 

6. Conclusion 

This article highlights that in the area of liability for damage caused by a defective product, i.e., civil liability, liability is 

conceived as a strict liability with the possibility of releasing the obliged party from liability if the damage caused by the 

injured party is proven or if it’s reasonably assumable. Considering all the circumstances, the defect did not exist while being 

released or later. In the area of criminal liability, it would probably be impossible to set some form of strict liability in law 

since the basic principles of criminal liability include the criminality of an act, while the consequence caused and the causal 

link are set between the act and its consequence, i.e., proving the fault of a particular perpetrator or group of persons.  

As long as AI and robots do not possess human characteristics (consciousness, subjective experience, emotions, 

motivation, will, creativity, social interaction, morality, and ethics), it is premature and unnecessary to create artificial legal 

constructs to sanction the wrongful actions of robots and AI. As the analysis shows, the appropriate solution is to apply the 

existing principles of criminal liability for the actions of legal persons - manufacturers, owners, and/or users of robots. To 

solve the problem of imputability of liability, it is necessary to focus on providing evidence in the form of the most detailed 

audit trail telling about the actions of the robot and its surroundings, or having digital biplanes of higher generation robots. 

The current state of robotics and AI does not require any other legal solution. 

In the authors’ opinion, the current possibilities of criminal law are sufficient to punish the unlawful actions of robots 

where the principles of civil law are insufficient. Inventing new social constructs consisting of punishing robots as such is 

impossible due to the dependence of being created by an owner, operator, or other responsible person. In the future, the 



Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 83

paradigm shift that other authors talk about would only make sense if robots were revolutionized as a separate entity with more 

power than humanity. Subsequently, the question is whether robots would create their legislation, hostile to humans. However, 

these considerations, in the authors’ view, are not on the agenda at all. The purpose of this paper is to stress that as much as 

possible, and the focus should be on providing leads, collecting evidence, and implementing the principle of imputability 

within existing criminal law. 

Future research in combating robot-related crime should combine technical, ethical, legal, and social aspects of this issue. 

It should also contribute to the development of effective measures to combat robot-related crime and ensure the safety of 

citizens. Meanwhile, four insights have emerge from the authors’ perspectives. First, the behavioral identification of new 

threats and patterns is associated with robots within different sectors. Second, the analysis of risks and the proposal of measures 

are implemented to increase the technological safety of robots and autonomous systems. Third, the search for tools to analyze 

the behavior of self-learning robots is deployed in specific situations (incidents). Last, the investigation concerning the current 

possibilities of legal systems allow permitting of the implementation of criminal prosecution is deployed to propose new 

procedures and facts in this area. Consideration should be given to whether existing legal instruments are sufficient or whether 

changes are needed. 

Acknowledgment 

The authors would like to thank Mr. Libor Preucil, Ph. D., Head of Intelligent and Mobile Robotics & Center for Advanced 

Field Robotics, Czech Institute of Informatics, Robotics and Cybernetics of the Czech Technical University in Prague, for his 

valuable consultation. 

Conflicts of Interest 

The authors declare no conflict of interest. 

References 

[1] Proposal for a Regulation of the European Parliament and of the Council Laying Down Harmonised Rules on Artificial 

Intelligence (Artificial Intelligence Act) and Amending Certain Union Legislative Acts, European Commission, 

COM/2021/206 final, 2021. 

[2] Proposal for a Directive of the European Parliament and of the Council on Adapting Non-Contractual Civil Liability 

Rules to Artificial Intelligence (AI Liability Directive), European Commission, COM/2022/496 final, 2022. 

[3] R. Calo, “Robotics and the Lessons of Cyberlaw,” California Law Review, vol. 103, no. 4, pp. 513-563, June 2015. 

[4] M. Simmler and N. Markwalder, “Guilty Robots? – Rethinking the Nature of Culpability and Legal Personhood in an 

Age of Artificial Intelligence,” Criminal Law Forum, vol. 30, no. 1, pp. 1-31, March 2019. 

[5] S. Ahn, “Artificial Intelligence and Criminal Liability,” Korean Journal of Legal Philosophy, vol. 20, no. 2, pp. 77-122, 

2017. 

[6] J. R. Searle, Minds, Brains and Science, 13th ed., Cambridge, Massachusetts: Harvard University Press, 2003. 

[7] M. Minsky, Computation: Finite and Infinite Machines, Englewood Cliffs, N.J.: Prentice-Hall, 1967. 

[8] F. Focquaert, E. Shaw, and B. N. Waller, The Routledge Handbook of the Philosophy and Science of Punishment, 

London: Routledge, 2020. 

[9] E. Watamura, T. Ioku, and T. Wakebe, “Justification of Sentencing Decisions: Development of a Ratio-Based Measure 

Tested on Child Neglect Cases,” Frontiers in Psychology, vol. 12, article no. 761536 January 2022. 

[10] F. Pollock, “Has the Common Law Received the Fiction Theory of Corporations?” Law Quarterly Review, vol. 27, no. 

2, pp. 219-235, 1911. 

[11] G. Hallevy, “I, Robot-I, Criminal-When Science Fiction Becomes Reality: Legal Liability of AI Robots Committing 

Criminal Offenses,” Syracuse Science & Technology Law Reporter, vol. 22, pp. 1-37, 2010. 

[12] L. Wang, “New Challenges Posed by Robots to China’s Civil Code in the Age of Artificial Intelligence,” Frontiers of 

Law in China, vol. 17, no. 1, pp. 33-41, March 2022. 



 Advances in Technology Innovation, vol. 9, no. 1, 2024, pp. 65-84 84 

[13] R. E. Fikes and N. J. Nilsson, “STRIPS: A New Approach to the Application of Theorem Proving to Problem Solving,” 

Artificial Intelligence, vol. 2, no. 3-4, pp. 189-208, 1971. 

[14] N. Bostrom, Super-intelligence: Paths, Dangers, Strategies, Oxford: Oxford University Press, 2014. 

[15] R. Kurzweil, The Singularity Is Near: When Humans Transcend Biology, New York: Penguin Books, 2006. 

[16] S. Russell, T. Dietterivh, E. Horvitz, B. Selman, F. Rossi, D. Hassabis, et al., “Letter to the Editor: Research Priorities 

for Robust and Beneficial Artificial Intelligence: An Open Letter,” AI Magazine, vol. 36, no. 4, pp. 3-4, December 

2015. 

[17] M. R. Ford, Rise of the Robots: Technology and the Threat of a Jobless Future, New York: Basic Books, 2015. 

[18] E. Brynjolfsson and A. McAfee, The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant 

Technologies, New York: W. W. Norton & Company, 2014. 

[19] R. Logan, “Can Computers Become Conscious, an Essential Condition for the Singularity?” Information, vol. 8, no. 4, 

pp. 161, December 2017.  

[20] A. Doležal and T. Doležal, “Proof of Causation in Medical Malpractice Cases in the Czech Republic,” The Lawyer 

Quarterly, vol. 5, no. 3, pp. 195-205, 2015. (In Czech) 

[21] Constitutional Court of the Czech Republic, The Ruling of the Constitutional Court of the Czech Republic, May 20, 

2004; similarly, for instance, Supreme Court of the Czech Republic, The Ruling of the Supreme Court of the Czech 

Republic, September 06, 2001.   

[22] M. de la Cámara, F. J. Sáenz, J. A. Calvo-Manzano and M. Arcilla, “Security by Design Factors for Developing and 

Evaluating Secure Software,” 10th Iberian Conference on Information Systems and Technologies, pp. 1-6, June 2015. 

[23] A. Maurushat and K. Nguyen, “Correction to: The Legal Obligation to Provide Timely Security Patching and Automatic 

Updates,” International Cybersecurity Law Review, vol. 3, no. 2, article no. 495, December 2022. 

[24] L. Gould, “What Are Digital Twins and Digital Threads?” Automotive Design & Production, vol. 130, no. 2, pp. 30-32, 

2018. 

[25] Siemens, “Digital Twin,” https://www.plm.automation.siemens.com/global/en/our-story/glossary/digital-twin/24465, 

November 05, 2018. 

[26] ABI Research, “Digital Twins, Smart Cities, and Urban Modeling,” Research Report AN-5239, September 15, 2019.  

[27] Bonn, “Implementation of Digital Twins to Significantly Improve Logistics Operations,” Trend Report DHL, June 27, 

2019. 

[28] European Union, European Parliament Resolution of 16 February 2017 with Recommendations to the Commission on 

Civil Law Rules on Robotics (2015/2103(INL)), Official Journal, C 252/239, 2018. 

[29] N. N. Taleb, The Black Swan: The Impact of the Highly Improbable, New York: Random House, 2008. 

 

Copyright© by the authors. Licensee TAETI, Taiwan. This article is an open-access article distributed 

under the terms and conditions of the Creative Commons Attribution (CC BY-NC) license 

(https://creativecommons.org/licenses/by-nc/4.0/).