Recently (March 2021) I was invited to speak at the Citizen Debate on AI & Law to present a European perspective on the future of AI regulation. The debate was organized by the new FARI Artificial Intelligence for the Common Good Institute, a collaboration between the Flemish and French Free Universities of Brussels. This gave me the opportunity to introduce the four Rs that should enable and constrain ‘4R AI’: resilient, robust, reliable and responsible AI and to relate them to the idea of robust law.
In the previous blog (part I) I have discussed what is meant with robust AI (a term of the trade), in this blog (part II) I develop a notion of robust law and in the final blog (part III) I will inquire how data-driven legal search may affect the robustness of law and the rule of law.
What is law?
Law is a complex system of legal norms (rules, principles and policies) and of legal relationships (between legal subjects, with regard to legal objects). It is also a practice that interacts with the world it aims to enable and constrain. Though some may think that law is only there to prohibit (constrain), the main role of law in modern society is to enable legal subjects to do things they could otherwise not achieve. This is the case because law creates a number of ‘institutional facts’ such as marriage, property, murder, contract and even money. These institutional facts are the ‘performative effect’ of legal speech acts, such as legislation and case law. Though oral societies also create institutional facts, their performative effect will differ from those of law-as-we-know-it. For instance, in a nomadic society of hunters and gatherers property may not exist at all, because in nomadic societies the relationship between people and things is one of use, respect, worship or magic, rather than exchange or possession. In a sedentary society with agriculture and cattle, what we call property may refer to land or livestock that belongs to an extended family or a kinship group and the rules for who decides about such property may be very different from our notion of individual property rights (alienation was not allowed and commodification of both land and labour is a recent invention).
The point here is twofold:
- the performative effect of speech acts that qualify something as a marriage, murder or contract demonstrates that effects are not always causal but often constitutive (speech acts do what they say),
- legal effect is the performative effect generated by a specific subset of written speech acts, such as enacting legislation or issuing a judgement (which also highlights the relationship with the attribution of power).
Law can be validly described in different ways (see above), but the most telling description is that law attributes legal effect if specified legal conditions have been fulfilled. As indicated, the notion of legal effect must not be confused with that of causal effect. The civil servant who registers a marriage does not ‘cause’ the marriage, even though by way of the registration the civil servant brings the legal effects of being lawfully married into being. Though the law stipulates and brings about that such registration results in a whole range of consequences (obligations, rights) that may also affect third parties, the relationship between law and legal effect is neither one of causation nor one of correlation. Instead, it is a specified performative effect attributed by what lawyers call ‘positive law’ (the law that is in force in a specific jurisdiction).
So, law enables (providing us with positive freedom to e.g. get married, own property) and it constrains (thus providing us with negative freedom from others doing things to us like murdering us or stealing our property). In doing so the law provides us with reasonable expectations about others’ actions, in turn enabling us to plan our life. If anything, it seems, law provides for legal certainty, thus enabling not just individuals but also society as a whole to make plans for the future. Yet another way of saying the same thing is that law is constitutive of our shared institutional world and regulative as to how we should act within that world.
Interestingly, a shared world assumes at least a tacit agreement about some basic tenets, generated and supported by a more explicit ‘social contract’. In modern law this becomes evident in the concept of jurisdiction, which highlights that legal norms are not universal rules; their application is restricted to a specific jurisdiction, often the territorial jurisdiction of a state. This territorial jurisdiction, in turn, constitutes the population living on that territory as a ‘demos’, and within the context of a constitutional democracy this ‘demos’ will decide its own fate by way of representation, deliberation and participation.
What is robust law?
As discussed in Part I of this blog series, in computer science robust AI stands for AI systems whose behaviour or functionality does not change unintentionally due to unanticipated perturbations within their environment. The concept of robustness in AI is connected with the concepts of resilience, reliability and responsibility, where resilience and robustness may refer to the more technical levels of guaranteeing a stable functionality and safety, while reliability and responsibility may refer to societal expectations in terms of conformity, fairness and to ‘robustness’ in terms of fundamental rights. We must admit, however, that the concepts are used in different ways within different disciplines while their usage overlaps even within computer science, engineering, science technology and society studies (STS) and law.
In this blog I use the notion of robustness to integrate both the more technical requirements (does a system do what it claims to do) and proper anticipation of societal impact (has such impact been considered and mitigated if necessary). Applying this to the law means that I take ‘robust law’ to stand for a legal system that incorporates checks and balances to resist turbulence due to unforeseen changes in the environment it constitutes and regulates. Since law ‘makes’ and configures our shared institutional world, thus providing trust and foreseeability, it is crucial that law itself remains sufficiently stable. Not in the sense of being rigid and thus brittle, but in the sense of enabling people to navigate their everyday life world, including the institutional facts that ground everyday life.
Such navigation will often require that law is adaptive while offering continuity, making sure that legal protection is not diminished due to changes in the environment. It may be that privacy was previously protected behind the walls of the home, making the tort action of trespass the right way to protect the right to privacy. In the US some justices of the Supreme Court still consider that the constitutional right to privacy depends on physical trespass. If, however, mobile phones contain far more details about our private life than entering a home would uncover, the right to privacy may have to shift towards the violation of reasonable expectations of privacy, for instance when a service provider or justice authorities obtain remote access to the content of the phone. To afford ‘practical and effective’ fundamental rights, law must be sufficiently robust, meaning it will have to reconfigure legal norms to offer equivalent legal protection. As we have argued elsewhere, this may imply a need for technology specific law – to compensate for a loss of protection due to transformations in the technological landscape.
For robust law it seems that legal certainty is pivotal (see the xkcd comic above, at some point we need closure, i.e. a decision which standard is in force). Just like with robust AI, we need to be able to trust that law will honour our legitimate, reasonable expectations. Moreover, it is the law itself that will decide whether the AI systems that increasingly ‘run’ and configure our shared world are trustworthy, i.e. robust. Where law enables those who suffer the consequences of unreliable AI systems to obtain compensation for damage caused, it institutes an incentive structure that is conducive to robust AI. For such liability (and e.g. compulsory certification) to achieve its purpose the law itself must be robust in the sense of being resilient, robust in the ‘technical-juridical sense’, reliable and responsible. In the words of the European Court of Human Rights, to count as law the output of a legal system must be foreseeable, accessible and incorporate the safeguards of the rule of law.
Rule of law as a precondition for robust law
The idea of the rule of law brings together three goals that define law while distinguishing it from both politics and morality. These goals are:
- justice, i.e. treating equal cases equally to the extent that they are equivalent while demanding proportionality between harm and compensation or wrongfulness and punishment,
- purposiveness, i.e. being instrumental in achieving political objectives and
- legal certainty, i.e. making sure that legal rights and obligations are interpreted in a coherent way and enforced if needed.
If law would only cater to justice it would be reduced to morality (natural law), if law would only cater to purposiveness it would be reduced to politics (an instrument to achieve policy goals) and if law would only cater to legal certainty it would be reduced to administration and enforcement.
Nevertheless, these overlapping and intertwined goals firmly ground the law in a moral stance (distributive and corrective justice), they clarify that law serves political objectives (it is instrumental without succumbing to instrumentalism) and ensure that legitimate expectations are enforced if necessary (neither depending on agreement of those who violated such expectations, nor on the agreement of the executive that must enforce the law).
The need to work towards all three goals means that whereas law serves political objectives it does so in a way that honours equal treatment. This asserts law’s instrumentality without succumbing to instrumentalism:
- Instrumentalism refers to an understanding of means as seen as neutral tools to achieve an end. Such neutrality enables a comparison of different means in terms of efficiency, precisely because instrumentalism makes different means interchangeable.
- Instrumentality highlights that the means must have a certain quality in light of the ends they serve. Having a legitimate aim does not suffice for the state as a justification to infringe human rights, the infringing measure (the instrument to achieve a policy goal) must be ‘in accordance with the law’ and ‘necessary in a democratic society’, as art. 8.1 of the European Convention of Human Rights tells us.
A similar difference must be made between legality and legalism:
- Legalism would merely require legislation that regulates the infringing measures, whatever the content of such legislation; it is based on a thin or formal conception of the rule of law.
- Legality, however, would assess the quality of the relevant legislation and check whether it guarantees the accessibility of the law, the foreseeability of its application and specified safeguards to limit infringements in volume, measure and time. Legality is based on a substantive and procedural conception of the rule of law.
Instrumentalism and legalism make law brittle:
- Instrumentalism does so because it may easily exchange law for whatever instruments are more efficient in serving a policy goal, e.g. nudging people behind their back or simply applying brute force against them. Instrumentalism thus also invites lawyers to hand over the study of law to economists, sociologists, political scientists whenever they claim to detect better instruments to achieve policy goals; resulting in the colonisation of law by approaches that put all their balls in one basket, for instance in the basket of ‘efficiency’, as if that were the only goal of the law.
- Legalism makes law brittle because it ignores that procedural legal norms have a normative kernel that should not be violated by a blind adherence to a static interpretation of the rule. Legalism and formal conceptions of the rule of law ignore the open texture of the foundational concepts of law, such as legal subjectivity, legal obligations, ownership, contract, due process, individual rights and so on. Legalism views the law as a static language that remains the same in changing circumstances, whereas legality and both substantive and procedural conceptions of the rule of law honour the integrity of law that grounds a legal tradition in the moral principles shared by those under its jurisdiction.
A substantive and procedural conception of the rule of law thus help to explain why the rule of law – as opposed to a rule by law – is a precondition of a robust law.