AI – Who Will Be Held Responsible ?

At a time when digital technology increasingly accompanies human life, the creators and users of new technologies have invented AI, a new tool that emerged less than a quarter of a century ago.

Artificial intelligence seems to be the new path of progress and marks a notable shift in the processes of human interaction. Yet the emergence of this technology, with potentially harmful effects on individuals, raises the question of how responsibility should be attributed when damage occurs. It is essential to ensure that responsibility for the risks, harms, and wrongs arising from the deployment of AI is clearly identified and assigned. What should be done when an AI’s action is involved in, or responsible for, accidents, or produces harmful, even disastrous, consequences? When one asks Siri, “Hey Siri, what are you responsible for?”… silence.

So, should humans be absolved of responsibility for the acts and decisions of artificial intelligence? If not, should the developer be blamed, or the user? What fault can be attributed to the decision of an artificial intelligence which nevertheless proves fatal to someone? How should the law be applied in an accident between two automated cars? Should a legal personality be created for AI? Might we one day even be forced to hand over our rights and duties to an AI?

An Emerging And Global Legal Problem

First, what is artificial intelligence? It is the set of theories and techniques implemented in order to create machines capable of simulating intelligence. By extension, in everyday language, it refers to devices that imitate or replace humans in certain implementations of cognitive functions. What matters here, then, is the way in which it becomes possible to legally apprehend a “human-like” behaviour produced by a digital machine. A major problem of a new kind emerges: AI is not exactly a tool, because it sometimes reproduces intelligent processes and can therefore act independently, notably through what is called a “black box,” a complex part of AI to which humans have no access, where calculations and decisions are made. But it is not a person either: it remains programmed, used, and controlled to a certain extent.

To legislate in this new domain, action must be taken at a broad scale. Influenced by several insurance companies, European parliamentarians asked that the possibility of creating a special legal status for AI be studied, “in order to clarify liability in the event of damage,” particularly damage caused by autonomous vehicles. Strongly contested, this position divides the actors in the debate into two camps. On the one hand, the so-called “moderates,” who see AI as a tool, a good, or a service, and believe that the notion of liability should not be revised, and that developers and users must ensure AI complies with the legal frameworks applicable to these categories.

On the other hand, the self-proclaimed “progressives,” whose objective is to create a legal personality for AI, want to allow this unprecedented “tool,” which fits no existing legal category, to develop fully.

Ai, Ethics, Moderation

Although many proposals have been made, few reforms and laws have been enacted, and most disputes related to this issue have been resolved through case law. This was the case in the very recent Tesla matter, where an accident between two automated cars that led to the hospitalisation of a forty-year-old man was settled through compensation paid by the brand. At the conceptual level, no common definition currently exists at the European Union level. However, Parliament asked the European Commission to propose a definition “of the different categories of cyber-physical systems, autonomous systems and autonomous and intelligent robots.”

In 2019, the “European AI Alliance” was created, bringing together several parties to debate the future of AI. By the end of the year, a draft set of guidelines for the ethical development and use of artificial intelligence, based on the EU’s fundamental rights, was expected to be put to a vote. In doing so, the group would take into account issues such as fairness, safety, transparency, the future of work, democracy, and more broadly the impact on the application of the Charter of Fundamental Rights.

Despite this stagnation, EU member states managed to agree on three rules concerning robots, which they intend to apply to AI:

• A robot may not harm the safety of a human being, nor through inaction allow a human being to be placed in danger.

• A robot must obey the orders of a human being, unless such orders conflict with the first rule.

• A robot must protect its own existence as long as this does not conflict with the first or second rule.

  
  

Even so, attributing legal personality to AI remains unthinkable at present. Legal personality is characterised by having rights and obligations, as well as the capacity to bring a civil action and to be held responsible for one’s acts. A legal person is also capable of expressing moral values. Once personality is attributed to robots or other autonomous artificial agents, they become legal subjects and enter the universe of legal persons. This was notably the case of the robot Sophia, designed by Hanson Robotics (United States), which obtained Saudi citizenship on 25 October 2017. Yet can a robot that once declared it wanted to “destroy humans” truly be considered conscious of its words in the same way as a human? Will robots be able to know that they are responsible and therefore liable to be held to account? The recent case involving an Uber autonomous vehicle road test in Florida that resulted in a fatal accident in March is expected to shed light on these questions.

AI is therefore a tool whose capacity must be limited, and whose use must be made compliant with the legislation governing services. It would also be possible, according to the proposals of the European Ethics Committee, to treat AI as a defective product or as affected by a hidden defect when it causes damage.

Ai, Changing The Law Towards A Legal Personality

However, the self-proclaimed “progressives” in this legal debate reject the path of moderation. Alain Bensoussan, a French lawyer and the author of a short manifesto on robot law, calls for the creation of a legal personality for AI. As a leading figure in this movement, he advances several arguments in favour of this approach.

Artificial intelligences cannot take the initiative to lie. As a result, they would offer an objective analysis of their responsibility in cases of damage. Returning to the example of an accident between two automated cars, their AIs would be able to evaluate the repercussions of their actions: which car is responsible for what. Thus, no judge would be needed to assign responsibility. Self-accountability would be possible. This “progress” would relieve courts of many cases and bring a certain impartiality. With legal personality, AI would be directly liable, it could be held responsible for damage and required to provide compensation. Punishing an artificial intelligence makes little sense, since punishment is meant to create understanding. Compensation would therefore be the only possible sanction. A form of “insurance capital” would be assigned to AI by its user or manufacturer as soon as it is put into circulation.

In the event of damage, the process of self-evaluation, attribution of responsibility, and compensation via this insurance capital would be almost immediate. But can this system truly be considered a legal advantage? It is certainly faster and less costly in time and money, but where it solves some problems, it creates others. Regarding self-evaluation and the assignment of responsibility, how can we ensure this assessment will not be deliberately distorted? There is no way to know if the evaluation passes through the black box.

As for compensation, an ethical problem immediately appears: compensating requires putting a price on harm. How can one evaluate the value of a human life lost in an accident? What factors should be taken into account? Moreover, this “insurance capital” suggests a morally dubious logic: the aim is no longer to prevent wrongdoing, but to anticipate it and reduce its impact on the responsible party. This is a dangerous step towards the erosion of responsibility.

It is also a dangerous step towards transhumanism: allowing two AIs to settle a dispute between themselves is to hand over our rights to a machine. Gaspard Koenig, a writer and a committed defender of liberalism, does not object to this, and argues in The End of the Individual that “for reasons of practicality and efficiency, this approach is entirely acceptable, it should suit us and delight us.”

Solutions For This Legal Vacuum

Every potential solution comes with its share of critics and dissatisfaction. Faced with this legal entanglement, two solutions have been considered by the European Parliament. Uncertain, Parliament proposes setting up “innovation sandboxes”: adopting temporary regulations in specific geographical areas to observe how these new systems evolve. Concretely, this would mean applying different rules across the European Union, waiting to see which proves most effective, and then applying it everywhere. But one problem remains: during this “trial period,” intended to last three years, two identical cases would be judged differently if they occur in two different countries. This situation could drive some AI companies away. In Spain, for instance, where regulation is expected to be among the strictest, Alphabet Inc. and Google have threatened to reduce their services. Such pressure, particularly from a giant like Google, slows any concrete implementation.

Contractual Liability

Parliament has therefore retained another possible solution: contractual liability. This would consist in creating terms of use between different manufacturers, and between manufacturers and users. This option has the advantage of relying on a system we already know and can manage: contractualisation. This solution is oriented towards moderation and the reduction of AI’s capacities, and it satisfies those who opposed excessive conservatism. Agreements between manufacturers would allow them to share compensation burdens in proportion to their involvement and their responsibility. Porsche and Google have even planned, should such a framework enter into force, to launch a collaboration.

Yet this solution also raises a problem: courts would then only be able to estimate the amount of liability, but would not be able to distribute it as they see fit among the contractual parties. Imagine that the Porsche–Google contract, to continue this example, designates Porsche as 65% responsible. Courts could demand compensation from both actors and ensure the contract is respected, but could not decide a different allocation. This would amount to a partial privatisation of justice.

As we have seen, the European Parliament’s attempt to find a solution to AI liability is far from straightforward. On one side stands the pressure of certain companies and tech giants, which would benefit from legal personality for AI in order to reduce potential losses and increase their independence. On the other side are jurists, philosophers, and scientists who argue that AI must be treated as a service, an object, or a good, and therefore judged as a hidden defect or a defective product when harm occurs. In the absence of progress toward a balance that might satisfy liberals, moderates, conservatives, and progressives alike, it may be wiser to set aside the actors’ demands and focus only on the assessments of scientists and jurists.

Stephen Hawking once said: “We must stop the progress of AI before we start asking who will pay for the damage it will cause.” That prudent and visionary advice can no longer be applied. Artificial intelligence is already too present in our lives; we are already too dependent on it to be able to stop its development. In any case, AI tools are evolving at a considerable speed. It is highly likely that our current tendency to delegate responsibilities to artificially intelligent systems will become an even more serious problem in the future, which is why the Parliament’s resolution and the European Commission’s work matter so deeply. The consequences of technological development must be anticipated while taking into account all social actors. An effective regulatory framework is essential to ensure that artificial agents will coexist harmoniously with humans, that they will be specifically designed according to human values and needs, and that they will operate and be capable of adapting accordingly. From this perspective, if we cannot slow technological development, it would be logical to reduce AI’s legal power by continuing to treat it as a tool, an object, a means to an end, and not an end in itself.