Constitutional Resilience in the Algorithmic Administrative State – From the rule of law to the rule of code and back

I. Introduction

Society, economy, and public administration are facing tremendous digital transformations, such as increasing automated decision-making, the rise of data-driven products and services, and the emergence of disruptive technologies such as distributed ledger technologies. At the same time, these transformations bring about digital challenges and threats, such as exclusion in case of digital illiteracy, discriminatory profiling based on AI, automation bias, problems of transparency and accountability, and problematic quality of (big) data input. To address these evolutions, this blogpost urges to further explore the concept of ‘constitutional resilience’ and more specifically the interaction between ‘the rule of law’ and ‘the rule of code’.

II. Constitutional resilience

To understand how constitutional resilience works, it is important to first grasp that constitutional law has two core functions. Firstly, constitutional law enables the functioning of the legal order by constituting institutions and by conferring power upon these institutions to fulfil their public tasks. Secondly, constitutional law functions as a defense mechanism encompassing both substantive and procedural safeguards to preserve the core functions and values of democracy and the rule of law. Hence, constitutional law enables us to cope with transformations and threats in our society, such as populism, anti-democratic forces, emerging technologies, climate change or multi-level governance.

Constitutional resilience is the ability of the constitutional order to cope with (among others socio-technical) transformations and threats while safeguarding the core functions and values of democracy and the rule of law.

This blog post departs from the assumption that there is a dynamic, reflexive relation between constitutional resilience of the rule of law framework and digital transformations. On the one hand, constitutional resilience defines the ability of the constitutional order to deal with socio-technical transformations and threats while safeguarding the core functions and values of the democratic rule of law.

On the other hand, digital transformations and threats have an impact on the resilience of the constitutional order itself and could thus transform its constitutional rules and values, usually in an incremental way, sometimes in a revolutionary way. Think for instance about the traditional distinction between on the one hand general rules, traditionally adopted by a democratically elected legislator, and on the other hand the application and execution thereof via concrete decisions of administrative authorities. This distinction plays an important role in relation to the principles of legality, democracy, fundamental rights, the separation of powers and judicial review. Nonetheless, the use of AI and algorithmic decision-rules with the capability to further develop themselves without human intervention challenges this fundamental distinction and these core principles of the democratic rule of law.

III. The rise of the algorithmic administrative state

In the past decades, the enormous expansion of government tasks and complexity in the social welfare state have led to the rise of the ‘administrative state’ in many democratic rule of law countries. Until the 1960s, one could usually safely assume that an administrative body composed of people exercised powers conferred by the legislature. Such an assumption does not longer hold in many areas of government action. Digital technologies have thus introduced the ‘rule of code’ or automatization in order to deal with the increase and complexity of government tasks. Moreover, the pace of digital evolutions and automatization is tremendous. As a result, one could say that the ‘digital/algorithmic administrative state’ saw the light (see figure 1 and 2, source: book chapter of J. Goossens and J. de Poorter).

In the administrative state, the legislator attributes and delegates extensive rule-making and administrative powers to administrative authorities. At the same, Parliament is facing tremendous challenges to adequately control the enormous amount of government action, a problem which is exacerbated by the rising level of complexity and opacity when algorithmic systems are used to support or fully automate administrative action. Moreover, complexity and opacity increase due to the interaction between these two in the algorithmic process. Finally, effective and meaningful judicial review of the algorithmic black-box proves to be a difficult challenge.

The use of AI in this algorithmic state enables to calculate the application of a rule based on a set of data derived from previous cases. Casey and Niblett coined this ‘micro-directives’, i.e. an extensive catalog of commands for every conceivable scenario. These micro-directives are not created on the basis of general rules adopted in advance by democratically elected bodies, but are generated by a self-learning algorithm that is guided by predetermined general principles and policy objectives and is based on data derived from a dataset built up from previous cases. In the most extreme case of the use of machine learning algorithmic models, the frequent application of case-based algorithms can even lead to what Casey and Niblett have called “the death of rules and standards”. In this regard, Shiffrin also rightly pointed out that the marginalization of the importance of general rules laid down in legislation and policy carries with it the danger of losing the moral aspect in public decision-making.

IV. From the rule of code to the rule of law and back

As abovementioned, the increasing use and possibilities of AI and big data give rise to further automatization and to new horizons for the rule of code. Self-learning algorithmic systems, for instance, ‘find’ and further develop algorithmic decision rules based on supervised or unsupervised learning. A second example are blockchain-based smart contract applications such as decentralized autonomous organizations (DAO’s). They embody a new form of rule of code or regulation and governance by code: autonomous code-based rules.

Automated or even autonomous decision-making processes must nevertheless comply with the existing legal rules. Therefore, this blog post urges to further explore how one must regulate and organize to go back and forth between the rule of law and the rule of code, a fortiori in case of automated administrative decision-making in the algorithmic administrative state which is characterized by broad discretionary powers. Such research is important, as automatization and the rule of code pressurize core principles of the democratic rule of law such as the principle of legality, safeguarding fundamental rights like the right to privacy, and the proper functioning of checks and balances, including effective judicial review.

V. By design approach

It is important to study how underlying general constitutional values such as transparency and accountability as well as general principles and concrete norms of good administration such as reason-giving, proportionality and due diligence can be operationalized in practice and embedded in the socio-technological design. It is often impossible to reverse the consequences of the application of an algorithmic system from the moment technology is developed and used. At that time, addressing the question whether the use of algorithms is (legally) acceptable is too late. Therefore, it is vital to pay attention to legal compliance in the phase of the development of algorithmic systems instead of analyzing whether an algorithmic system is compliant with the law after being implemented. Moreover, in order to successfully implement a by design approach of legal compliance interdisciplinary dialogue and cooperation is necessary.