1 Introduction

Currently, the rapid technological advancement of AI in the civilian sector is accompanied by accelerating attempts to apply this technology in the military sector. However, new technologies always raise ethical questions. In particular, the use of AI-equipped lethal autonomous weapons systems (LAWS), also known as “killer robots,” raises even more acute ethical questions because unlike other areas of robotics, the use of lethal force can sometimes involve serious infringements of human rights, such as killing or maiming innocent people. Since the 2010s, opposition to the development and use of LAWS has emerged at the civilian level and in intergovernmental discussions.

This study focuses on the argument that LAWS violate human dignity, as this argument has been a major driving force underlying opposition movements in recent years.^{Footnote 1} Human Rights Watch has warned that “Fully autonomous weapons are also morally problematic because they threaten the principle of human dignity” (HRW 2016: 23); the International Red Cross has also warned that “if the connection between the human decision to use force and the eventual consequences is too diffuse, then human agency in that decision is weakened and human dignity eroded” (ICRC 2018: 11).

There are two main types of human–machine relationships that are the subject of debate regarding the ethical validity of LAWS. The first is the human–machine relationship on the side of the Agent, which is associated with the concern that they would effectively remove “meaningful human control.” The second is the relationship between LAWS and those on the side of the Target, which raises the recurrent question about “death by algorithm.” As we will see in this paper, dignity-based objections to LAWS raise the question of how the emergence and proliferation of autonomously acting machines will affect the traditional value of human dignity with respect to both these aspects.

Indeed, if there is a widespread opinion that LAWS are contrary to human dignity, this could be a good reason to look at this issue with great concern, as is the case with other international regimes governing controversial weapons. For instance, international treaties ban the production, possession, and use of biological and chemical weapons, anti-personnel mines, and cluster munitions as inherently inhumane regardless of their military effectiveness. National and international rules that absolutely prohibit torture and other cruel, inhuman, or degrading treatments are also rooted in people’s negative perceptions about these practices.

Nevertheless, the precise meaning of why and how LAWS violate human dignity is not always clear. One obvious reason is that the concept of human dignity itself remains ambiguous. Recent philosophical literature on this concept commonly distinguishes between conceptions of value and status (Beitz 2013; Lucy 2017: ch. 4; Riley 2018: ch. 2). However, relying on this distinction alone is insufficient to capture the multifaceted nature of the concept. Therefore, in this study, the concept of human dignity is categorized into five components when examining the ethical validity of LAWS.

This study also uses this classification to conduct a social survey of what people mean when they think that LAWS violate human dignity. It is important to be specific about people’s attitudes because they affect the potential impact of the issue, such as whether ethical concerns about LAWS are conditional or principled. Since people’s perceptions of weapons in general have some bearing on their ethical status, the first step is to explore more closely what people actually find problematic regarding the use of LAWS against human beings.

The remainder of this paper proceeds as follows. First, drawing on philosophical research on human dignity, this study distinguishes between value and status conceptions and further subdivides the multiple layers of meaning within them into five components. Next, this study uses these components to organize existing dignity-based objections to LAWS. Finally, this study presents the results of a social survey conducted by the authors with the above classification in mind and discusses the implications. Through the above discussion, this study highlights the importance of detailing and empirically testing the conceptual meanings of human dignity in advancing the ethical debate on LAWS.

We begin by making a preliminary distinction regarding the form of LAWS we have in mind in this study. In general, LAWS can be divided into two types: semi-autonomous and fully autonomous. In the former case, the machine performs tasks autonomously, from target selection to attack execution, whereas humans monitor the machine’s actions as a failsafe and intervene to abort the attack if necessary (i.e., human on the loop). In the latter case, there is no room for human intervention in the machine’s actions, from target selection to attack execution (i.e., human out of the loop). It is the latter form, which has not yet been fully deployed and is the focus of today’s international disputes.

Weapons that operate automatically without human intervention predate LAWS. Some weapons operate spatially beyond human control, such as land mines, sea mines, and guided missiles, whereas others operate temporally outside human control, such as automatic close-in weapon systems. The difference between these weapons and LAWS is that the latter have the ability to adapt to their environment and conditions through machine and deep learning using AI, without the intervention of other entities. An AI can change its internal state from the one at the development stage, and it can also change it from the one at the stage when the user decides to use it.

This does not mean that LAWS can acquire the same or similar levels of autonomy as humans. Indeed, the following argument emphasizes that this is not the case.^{Footnote 2} The critical problem is that machines with learning capabilities beyond human control can select the Target and carry out attacks on the battlefield. These decision-making capabilities, like other applications of AI, such as self-driving cars, have the technical advantage of reducing the cognitive, emotional, and physical burden on humans, but also carry the disadvantage of decreasing meaningful human control over machines. The advantage and disadvantage are two sides of the same coin and are inherent characteristics of the AI technology itself. In this sense, LAWS are considered to evoke new concerns that are distinct from the use of other weapons.

The concern here is related to the so-called “responsibility gap,” in which responsibility for mistakenly shooting the wrong Target becomes unclear (Sparrow 2007). For instance, imagine LAWS attacking a human target without a data link to humans, resulting in an accidental shot. The actions of self-learning and self-guided machines are more or less uncontrollable by both designers and users, and there may be limits to holding them criminally responsible for murders or injuries.^{Footnote 3} So, who should be held responsible? This kind of problem arises because machines are not autonomous in the same sense as humans—after all, the “autonomous” in the term LAWS is somewhat misleading. The question being asked is whether machines that do not have the same capacity for self-judgment as humans, should be allowed to act independently of human control.^{Footnote 4}

2 What is human dignity

Human dignity is one of the key concepts in postwar society. It has been enshrined in national and international laws, such as the Universal Declaration of Human Rights and the Basic Law of the Federal Republic of Germany, and subsequently incorporated into several national constitutions and international human rights treaties, as well as individual human rights conventions on the elimination of discrimination against women and racial discrimination, the prohibition of torture, and the rights of the child. This concept has been discussed in the field of biomedical ethics in terms of the validity of euthanasia, genetic engineering, organ trafficking, cloning, and other forms of medical treatment. The concept of human dignity has been interpreted in several ways. This section begins by presenting two basic dimensions that have become mainstream in the contemporary philosophical literature.^{Footnote 5}

2.1 Value conception

The first understanding of dignity involves viewing it as something inherent to human beings. This means that the individual has value and is unique and irreplaceable. If a person with dignity is endangered, it represents potential loss of the world’s only inherent value, and even citing the interests of the wider society cannot compensate for this loss. This conception is compatible with postwar society’s view of fundamental human rights, which upholds the “inherent dignity” of human beings in the Preamble to the Universal Declaration of Human Rights as well as the Preamble to the International Bill of Human Rights.

Immanuel Kant supported this conception in his various works. The German word for dignity is “Würde,” a close relative of the English word “worth.” Kant contrasts price and dignity as aspects of value. Relative value is represented by price, and can be replaced by its equivalent. In contrast, something that has intrinsic value is called dignity: “what constitutes the condition under which alone something can be an end in itself does not merely have a relative worth, i.e., a price, but an inner worth, i.e., dignity” (Kant 2011: 99).^{Footnote 6}

Kant associated dignity with the third formula of the categorical imperative or the formula of autonomy. Autonomy refers to the ability to act and make decisions without being controlled by someone. Kant states that this is the ability to make moral laws and obey them at the same time. Unlike natural beings, rational beings can acquire moral laws against their natural inclinations. The inner humanity of each person who is not only human but can be moral deserves dignity: “the dignity of humanity consists in just this capability, to be universally legislating” (Kant 2011: 109).

Elsewhere, Kant relates dignity to the second formula of the categorical imperative, or the formula of humanity: “Humanity itself is a dignity; for a man cannot be used merely as a means by any man […] but must always be used at the same time as an end” (Kant 1991b: 255, cf. 230). Natural existence is merely a means, i.e., a commodity with an exchangeable relative value. Conversely, human beings who acquire moral laws in defiance of natural inclinations become independent of external control. In this kingdom of ends formed by human beings, dignity must be equally and mutually respected.

The value conception of dignity was invoked in the case of the controversial German Air Safety Law (Rosen 2012: ch. 2 sec. 6). Since the postwar Basic Law, Germany has held that “Human dignity shall be inviolable” as a constitutional foundation (Article 1). The Air Safety Act, enacted in response to the 9/11 terrorist attacks, authorized the shooting down of hijacked planes, raising the question of whether this violated the Constitution. The Court ruled that the sacrifice of innocent passengers, even to prevent greater casualties, was a violation of human dignity, because it caused the loss of their intrinsic value that could not be replaced.

2.2 Status conception

The second interpretation of dignity is that it is something given to a person. In this comparative notion, the subject enjoys the status of something, indicating that a person with dignity is ranked higher than others in certain respects. This is often the basis for granting exclusive privileges to certain subjects. Dignity is derived from the Latin word dignitas, which means “to be respectable or dignified.” In this sense, the status conception predates the value conception and is more faithful to its etymology.

From one perspective, dignity as status has been used as a notion of superiority or inferiority between humans and nonhumans. For instance, the ancient Roman philosopher Cicero considered dignity a criterion that distinguished humans from animals (Cicero 1991: 41–42 [1.105–106]). He believed that the ability to obey one’s duty is inherent in humans and that, in this respect, humans are superior to other creatures. The Renaissance works of Giannozzo Manetti’s De dignitate et excellentia hominis and Pico della Mirandola’s Oratio de hominis dignitate also represent ideas that emphasize the specialness and excellence of human beings compared with other creatures.

Dignity as status has also been used to indicate superiority or inferiority among humans. In another context, Cicero views dignity as something that can only be granted to a few people within a social hierarchy. In particular, in his defense of mixed government, he criticizes democratic government ruled by a single class of plebeians as undignified, because it excludes elements of monarchy and aristocracy: “when everything is done by the people itself, […] that very equality is itself inequitable, in that it recognizes no degrees of status [gradus dignitatis]” (Cicero 1999: 19 [1.43]).

The revolutionary era of the eighteenth century added the notion of social equality to this understanding. Under the old regime in France, “dignité” referred to the privileges of the nobility, until the French Revolution, by abolishing them in favor of universal human rights, ensuring that the status enjoyed by the few was now open to all. The French Declaration of Human Rights states that “All citizens, being equal in its eyes, shall be equally eligible to all high offices [dignités], public positions and employment” (Article 6).

The legal philosopher Jeremy Waldron represents the status conception in the era of universal human rights. In his view, universality and equality accord with the current concept of human dignity, which arose historically as a result of guaranteeing all the dignity traditionally accorded to limited positions and statuses. Now “every man a duke, every woman a queen, everyone entitled to the sort of deference and consideration, everyone’s person and body sacrosanct” (Waldron 2012: 34). Human dignity enshrined in postwar international human rights treaties can thus be seen as the result of the upward and equal elevation of the status of all.

2.3 Distinctions between humans and machines

As explained above, today’s concept of human dignity is a combination of a more modern value conception and a more traditional status conception. As a key concept in postwar society, dignity is based on the understanding that all human beings have intrinsic value and are of equal status in this respect. One can say that a subject has a deserved status because it has a deserved value (Beitz 2013), and in contrast, one can also say that a subject has a deserved value because it has a deserved status (Riley, 2018: ch. 2). Some even argue that both reflect different aspects of a single concept of dignity (Lucy 2017: ch. 4).

From the perspective of the human–machine relationship, both value and status conceptions imply that humans and machines are essentially different entities. That is, the equality of the value and status of all humans is combined with the understanding that machines do not have the same status as humans because they have no intrinsic value. For instance, according to Kant, only humans are essentially different from objects and animals in that humans can behave autonomously and morally without inclinations. By focusing on the intrinsic value of human beings, he simultaneously expresses the difference in ontological status between humans and nonhumans as follows^{Footnote 7}:

Thus once the germ on which nature has lavished most care—man’s inclination and vocation to think freely—has developed within this hard shell, […] it even influences the principles of governments, which finds that they can themselves profit by treating man, who is more than a machine, in a manner appropriate to his dignity. (Kant 1991a: 59–60)

In questioning the relationship between AI, including LAWS, and human dignity, it is necessary to clarify the differences between humans and machines that are recognized in the concept of human dignity. This study classifies these into five components. Broadly speaking, the first three of these find their elements in the Kantian value conception, and the final two represent the comparative status of humans in relation to other humans and nonhumans. Value and status conceptions are inextricably linked in all components in the sense that a person’s value determines their status, while a person’s status is derived from their value. The five components are as follows:

Agency: human beings can act and make decisions autonomously.

End-not-means: human beings are ends in themselves in that they have autonomous capacities.

Uniqueness: human beings are ends in themselves and therefore are unique.

Reciprocal respect: human beings are ends in themselves and therefore deserve equal and mutual respect.

Specialness: human beings are more than nonhumans.

3 Classifying dignity-based objections to LAWS

As noted at the beginning of this paper, the argument that LAWS pose a threat to human dignity is a strong one among the objections. However, reflecting the polysemous nature of the concept of human dignity itself, it is not always easy to organize these objections. In the previous section, this study analyzed five components based on the distinction between value and status conceptions as well as the relationship between them. In this section, this study organizes the five variations of dignity-based objections to LAWS according to these components.

The authors do not believe that all these variations are equally valid; some of them may miss the mark. In any case, to examine the validity of dignity-based objections to LAWS as one of the focuses of today’s international disputes, it is first and foremost necessary to distinguish in what sense the concept of dignity is being called for. This study does not aim to provide a full examination of all these variations but will mention some specific evaluations as far as space allows.

3.1 Agency

First, according to Kant’s formula of autonomy, humans are capable of exercising their autonomous faculties and establishing moral laws. As the Agent, human combatants can decide whether to pull a trigger, sometimes independent of their military command. However, machines operate within their programming and do not deviate from it, either intentionally or negligently. “Unless re-programmed, the machine will engage the targeted person upon detection. Killing a person, however, is a truly existential choice that each soldier needs to justify before his own conscience” (Leveringhaus 2016: 92).

This is related to the problem of mindless killing, regarding which LAWS lack emotional factors such as pity and hesitation (Leveringhaus 2018). Even at the cost of deviating from military orders, humans can often show compassion for their Target because they understand the inherent value of human life. However, algorithms do not exhibit such pity or hesitation; nor are they moved to beg for life. This lack of agency is the key difference between humans and machines as the targeting Agent.

AI may succeed in imitating human moral judgments. Nevertheless, such AI still does not behave morally (Purves et al. 2015). The internal logic of a deep learning network is incomprehensible to humans, and even if it were somehow explainable, it would not be considered intentional or reasoned. In this sense, an AI cannot be a moral agent; it cannot have intentions or reasons that guide a particular decision in the first place. Therefore, the question of whether the AI is acting for the right reasons is misguided.

However, whether these subjective factors deserve serious moral consideration remains an open question. Even human combatants cannot be blamed simply for their internal state of mind—whether they are attacking the Target for the noble cause of defending their country, out of hatred for another race, or because they are psychopathic individuals who enjoy killing for its own sake. The condition of “right intention” in just war theory is taken into account when debating the ad bellum permissibility of going to war, but is not required for each act of engagement in the battlefield. Therefore, it would not be possible to categorically state that the existence of LAWS lacking this sense of agency is deemed immoral.

3.2 End-not-means

Second, as in Kant’s formula of humanity, humans capable of autonomously establishing ends must be treated as ends in themselves, unlike objects. Human beings are not to be treated as mere tools, even when they become necessary and legitimate military targets. However, when an autonomous weapon recognizes a human being as an object to be eliminated, it equates the human being with an object that cannot be an end in itself: “for an autonomous robot/drone to be programmed to kill a human is to treat a rational being as if it were merely an object” (Johnson and Axinn 2013: 134).

Of course, combatants are also instruments of war in a sense. However, this does not prevent us from recognizing that they are still flesh-and-blood humans and not weapons. Napoleon once said that “Soldiers are made to be killed,” as if individual combatants were as dispensable as chess pieces that he could discard and forget (quoted in Walzer 1977: 136). The case today is quite different; attacks that cause unnecessary suffering are illegal under international law, and once captured, combatants are entitled to humane treatment. Military necessity must always be balanced against the principle of humanity.

The inherent wrongness of the act of killing itself—the doctrine of “thou shalt not kill”—is an ethical principle deeply rooted in human history and common to almost all cultures and religions across time and place. Even when justified on lesser-evil reasoning, harmful acts always remain “evil.” Due to this multiplicity of moral judgments, people may experience psychological distress, such as PTSD, after committing a harmful act against others, even though they are convinced that it was the right decision. In short, humans can and do face moral dilemmas in which commitments to incommensurable values cause significant emotional distress.

This component requires certain qualifications on the part of the Agent (Enemark 2014: ch. 7). People may have a moral reminder that an action is both right and wrong. As Anselmo, a character in For Whom the Bell Tolls, says, “The killing is necessary, I know, but still the doing of it is very bad for a man” (quoted in Norman 1995: 199). It would be unlikely to program such psychological characteristics into a machine that ultimately processes information based on the binary logic of “0 and 1;” in any case, there does not seem to be any military value in doing so.

3.3 Uniqueness

Third, humans are unique and irreplaceable because of their ability to define their ends. This is fundamentally at odds with the algorithm that activates LAWS. One of the remarkable capabilities of AI is pattern recognition, in which machines can identify patterns, such as collections of things that are similar. In machine learning, AI autonomously learns patterns from input data. In deep learning, this type of learning is performed using multilayered neural networks. LAWS use image recognition and behavioral analysis to determine whether the Target is legitimate.

It is quite possible that the decisions made by LAWS will closely mimic human decisions regarding who is the legitimate Target and who is not. However, in pattern recognition, the original situation of different people with different backgrounds living in different circumstances is stripped of its uniqueness and replaced by specific statistical characteristics, regardless of how detailed they may be. Whether a particular Target is attacked depends not on who they are but on what statistical attributes they happen to have. In this sense, “Death by algorithm means that people are treated simply as targets and not as complete and unique human beings” (Heyns 2016: 370).

The methods used as criteria for targeted killing are now moving away from identifying individual attributes to signature attacks that follow a specific pattern of the Target’s hostile behavior such as communication and interaction with others. There are several problems with this shift. First, pattern recognition cannot probabilistically eliminate the possibility of false negatives. Second, systematic human biases may have influenced target selection. Consequently, there is a concern that unrelated individuals may be targeted based on inaccurate or misperceived information.

Of course, the ability of LAWS to discriminate is outstanding compared to indiscriminate attacks, such as area bombing.^{Footnote 8} The principle of distinction between combatants and civilians is the cornerstone of international humanitarian law of our times, and considering this, the use of LAWS programmed to make such a distinction is undoubtedly a moral advance. However, this distinction itself is based on general patterns rather than on the individuality of the Target. The technical characteristics of AI, which are good at pattern recognition that extracts commonalities from large amounts of data, do not necessarily sit well with the principles of postwar society, which has sought to prohibit collective punishment and instead emphasizes individualized liability.

3.4 Reciprocal respect

Fourth, human beings are inhabitants of a “kingdom of ends” who deserve equal and mutual respect. With the introduction of LAWS, the physical and psychological distance between warring parties has decisively increased, and the Target is unilaterally killed without knowing the intent or purpose of the Agent. Reducing the risk of human casualties on the Agent’s side to zero makes it clear that the Agent’s and the Target’s lives are weighed on different scales. “Autonomous weapons create a hierarchy of human dignity […] Their lives are deemed less worthy than human combatants replaced by autonomous weapons, and they are treated as mere relative ends that can be subjected to efficient disposal” (Ulgen 2018: 195).

This is concerning in terms of risk asymmetry (Kahn 2002). Killing, an act not permitted during peacetime, is permitted during wartime because the battlefield is a special place where killing and being killed do not occur occasionally, but regularly. In this sense, all combatants involved are and should be of equal moral status. The use of LAWS undermines the military ethos, which holds that belligerents can kill their opponents only given mutual risk. From this perspective, a war in which one of the belligerents takes no risk is not morally justifiable.

This can also be problematic in terms of lack of personal interaction (Sparrow 2016). What a person does to another is not just an expression of their attitude toward the situation but an expression of their attitude toward the person. Even the act of war can be seen as a kind of interpersonal relationship, and the ethics of war, such as the principles of distinction, necessity, and proportionality, are part of this attitude toward the other. When we entrust killing to machines, this interrelationship is definitively severed. The use of LAWS against enemy combatants should be considered “mala in se,” because it shows a profound disrespect for fellow human beings on the Target’s side.

Note, however, that the asymmetry of risk and lack of personal interaction are not necessarily problems unique to LAWS.^{Footnote 9} Asymmetry between the Agent and the Target is not limited to modern warfare, but is a phenomenon that has existed throughout the history of war and has become more prevalent with the advent of precision-guided munitions and unmanned aerial vehicles. There is nothing inherently immoral about this kind of asymmetry and some even argue that the asymmetry is morally preferable for minimizing damage to both one’s own troops and enemy civilians (Schulzke 2016; Strawser 2010). The Reciprocal respect objection to LAWS does not seem particularly convincing, either in terms of its own logical validity or in light of the reality of war.

3.5 Specialness

Finally, human beings are autonomous and, therefore, ends in themselves, and thus differ from nonhumans, surpassing them in status. The issue of dignity here is rooted in the difference in ontological status between humans and machines, and the sense of superiority of the former. As revealed in the classic Three Laws of Robotics, we have an understanding that machines are human tools and should be at the service of humans. Programming a machine to kill humans beyond the possibility of accidental malfunction seems to contradict this basic intuition.

The tenor of the Frankenstein Complex, the fear that autonomous AI monsters will seek to displace and destroy humanity, has been the subject of many works of science fiction. The fear that a machine that must be under human control might plot an uprising seems deeply rooted in our intuition. The reluctance to be killed by someone with whom communication has been severed is matched by our strong aversion to being reduced to nonhuman status: “people should not die at the hands of a robot. It would be a kind of species disloyalty” (Simpson and Müller 2016: 318).

Indeed, death by algorithm evokes the merciless cruelty of a high-tech mousetrap (Johnson and Axinn 2013: 134; Reichberg and Syse 2021: 153). The manner in which a killing is performed has a significant impact on whether it is psychologically and morally acceptable. Even if one is killed, dying in a fair fight, being murdered, or being preyed upon by animals are not the same experiences for the victim. Similarly, even if one is killed, one should not be killed by an information-processing algorithm.

This argument focuses on the inequality in the ontological status between humans and machines. Indeed, as we saw in the previous section, when the concept of dignity was discussed in the literature historically, it was implicitly associated with human dignity. Currently, this concept is often used in biomedical ethics to draw a line between humans and nonhumans. This is consistent with the absolute prohibition of certain weapons and treatments that are inherently inhumane. The authors find this line of reasoning most interesting as a dignity-based objection to LAWS. We will return to this point in the evaluation of the social survey results in the next section.

4 Survey and analysis

In the previous sections, this study delineated dignity-based objections to LAWS via five components of human dignity. While these components are interrelated, they focus on separate aspects of the human–machine relationship. The differences in these objections affect the long-term prospects of the issue. When people say that LAWS violate human dignity, we must specify what they actually mean. This study also addresses this issue and extends the literature by providing empirical evidence regarding whether people value human dignity with respect to the ethical validity of LAWS and, if so, how specifically they value it. In this section, this study presents the results of a social survey conducted by the authors in conjunction with the conceptual framework so far discussed, and explore the implications.

4.1 Survey method

Several studies have empirically examined normative attitudes toward international conflict (Benbaji et al. 2015; Press et al. 2013; Sagan and Valentino 2017; 2019). However, social surveys of the ethical validity of LAWS are relatively rare, especially those that further delineate the complex concept of human dignity. To determine people’s attitudes based on the LAWS scenario, the authors developed a questionnaire following the methods of previous studies, but added specific issues of LAWS and conducted an experimental Internet survey among residents of Japan.

The survey was distributed through QIQUMO’s online survey platform, using a randomization technique that probabilistically varied the order of the scenarios and questions presented to the respondents. Power calculations were performed for the participants in the study. With a population size of 123.4 million (the population of Japan), a confidence level of 95%, and an acceptable margin of error of 0.05, a target of 385 respondents was set. Regarding the possibility that the quality of responses to online surveys is not high, this number of responses was the result of collecting only those who appropriately responded to the questions about their understanding of LAWS and their efforts to complete the online survey.

The survey details are as follows: a first pilot survey was conducted in December 2021 with a sample of 100 people; after improving the survey methodology, a second pilot survey was conducted in June 2022 with a sample of 200 people, which was incorporated into the main survey that was conducted in June 2022 with a sample of 1200 people. We see no reason to consider any particular bias because the population of those who participated without a particular assumption on LAWS was consequently distributed similarly to the standard population distribution. This point of statistical validity is an important factor that makes the randomized controlled experiments adapted in this study possible. The main survey included two questions to check for appropriate responses, and the responses of 445 respondents were used after a data quality check.

4.2 Survey results

The survey was designed as a series of questions, with respondents who selected a particular option being asked to answer the next question. First, the definition of LAWS was provided, followed by a fictitious case of international conflict. Respondents who answered “no” to the question of whether LAWS should be used were then asked to answer why they felt that way, either by selecting one of the options or writing freely. The results of the survey are summarized in Table 1.

Figure 1 shows the statistical details of this result based on the main survey.

Which of the following is your reason for not using LAWS?

Full size image

In our surveys, many respondents consistently chose “Because it is against human dignity for machines to kill and injure humans” as their reason for opposing the use of LAWS. This finding is consistent with the emphasis on dignity issues in existing opposition movements and academic discussions, as noted at the beginning of this paper. However, it should be noted that more responses to this survey supported the use of LAWS than opposed it. In the main survey, 217 (48.8%) respondents supported the use of LAWS, more than double the 107 (24.0%) who did not support its use.

In addition, as a question directly related to the topic of this study, this survey asked respondents who selected “Because it is against human dignity for machines to kill and injure humans” to answer why they thought so, either by selecting one of the options or writing freely. Given the classification presented in the previous sections, option A was designed to correspond to the Agency component, option B to End-not-means, option C to Uniqueness, option D to Reciprocal respect, and option E to Specialness. The results are summarized in Table 2.

Figure 2 shows the statistical details of this result based on the main survey.

Which of the following is your reason for thinking that LAWS is against human dignity?

Full size image

The results were somewhat inconclusive depending on the specification. In particular, options A, B, and E, which focus on differences in capacity or ontological distinctions between humans and machines, attracted many responses. Generally, the lack of human agency in machines and the possibility of being killed or injured by those who lack reason or emotion were the primary reasons underlying the objection to LAWS. The so-called physiological “yuck factor” was also a concern such as aversion to the killing of humans by machines, as reflected in option E (Kelly 2011).

Several caveats should be noted when interpreting these results. First, the questions were structured such that respondents who said that they did not support the use of LAWS were asked to provide further reasons for their opposition and the absolute number of respondents was small for some questions. Therefore, we could not statistically analyze trends, such as which of the different conceptions of dignity are more or less likely to be supported based on the correspondents’ sociological attributes, such as age and gender. However, another study has shown that men are more likely to support the use of LAWS (Arai and Matsumoto 2024), and it will be possible to find similar trends in differences based on their demographics in future studies.

Second, this was not an international survey but a survey in one country, Japan. Therefore, the results should be complemented by a larger survey on the relationship between AI, including LAWS, and human dignity. For instance, attitudes toward the importance of dignity may vary greatly depending on whether a person holds certain religious values; whether the development and proliferation of technology, including AI, may affect the way we perceive the relationship between humans and machines. In today’s international disputes, there is a significant difference in opinions between countries, with proponents of LAWS and those who want to regulate them. This difference may be based purely on political or military reasons but may also be a matter of values.

4.3 Implications

The implications of these results are now explored. It is not possible to know directly from this survey which component(s) of dignity as classified in this study the respondents actually had in mind when they invoked the concept of human dignity. However, the trends in responses may be interpreted in specific ways. For instance, options A, B, and E, which received many responses, emphasized the essential difference between humans and machines, such as that they are different in terms of reason and emotion and that humans should not be killed by nonhumans.

Of particular note is the presence of relatively frequent “yuck factor” responses, as indicated by option E. The existence of negative intuitions that distinguish humans from nonhumans has been frequently mentioned in the fields of social psychology and biomedical ethics (Haidt 2012: ch. 5; Kass 2002: ch. 5; Nussbaum 2004: ch. 2). There is sharp disagreement on whether these intuitions have authority as the “wisdom of repugnance.” In any case, it should be kept in mind that in the field of the ethics of war, not only rational notions of who is autonomous or who is responsible, but also sensory factors play a role in the question of whether the use of LAWS is considered right or wrong.^{Footnote 10}

Dignity-based objections to LAWS may then be rooted in a more general perception of robotics and not just in the military domain. For instance, they may share the same problematic structure as dignity issues regarding social robots in everyday life (Sharkey 2014). Caregiving robots may be able to meet the material needs of one who needs care, but it is difficult to meet the person’s spiritual and relational needs for understanding and empathy. The issues of human dignity are relevant to people’s social lives in general, beyond matters of life and death, and they cannot be fully resolved simply by talking about the military merits or technical advantages of LAWS.

Moreover, it may be necessary to question the validity of the assumption that humans are more than machines. This may be nothing more than a form of speciesism—a prejudice that needs to be overcome. If this is merely a negative bias against novelty, it can be mitigated by habituation to their existence and operation. However, empirical findings, such as the “uncanny valley” hypothesis, have long been debated in psychology, and it still makes sense, at least for now, to assume some ontological distinction between humans and machines, which we call dignity.

5 Conclusion

The findings of this study are as follows. First, the distinction between value and status conceptions made in the existing philosophical literature is not sufficient to capture the relationship between humans and machines. This study classified human dignity into five components that simultaneously included both dimensions. Second, based on the above classification, this study distinguished several variations of dignity-based objections to LAWS, and conducted a social survey that revealed that people’s attitudes had relatively strong connections to the components of Agency, End-not-means, and Specialness, among others. Third, some trends in people’s attitudes found in the survey results can be interpreted as bearing on the question of dignity in relation to robotics in general, which has been addressed in the fields of social psychology and biomedical ethics.

What we have learned from this study is that, in advancing the case of dignity, it is important to pay attention to the ontological distinction between humans and machines, at least as part of their distinction. With the technological development of AI, LAWS may in the future become better—in some cases, better than humans—at distinguishing between combatants and civilians, and identifying signs of surrender. Even then, we may not be able to completely ignore the negative intuition that being killed or injured by machines is, in some sense, profoundly inhumane. To understand what underlies the appeal to human dignity, we must first carefully determine the meaning of dignity and then examine the validity of such objections on both conceptual and empirical grounds.

The technological development of AI is gaining momentum at an accelerating pace and is penetrating our daily lives in various forms. At the same time, recent years have witnessed several international armed conflicts that continue to cause suffering worldwide and it has been reported that LAWS not under human control may have killed or injured human beings in 2020, for the first time in history, in the Libyan civil war. These global situations may change people’s attitudes toward AI including LAWS, both now and in the future. The idea of dignity has historical origins dating back to classical antiquity, and it would be worthwhile to test the theory and application of this idea by exploring the perceptions of the human–machine relationship in our increasingly volatile times.