Our focus has been on “legislation as code”. This is because, along with the notion of “machine-consumable legislation”, is a phrase that has been used by advocates for better rules approaches, the Better Rules programme and among rules as code discussions.
As discussed at Part 2, we have observed that the important distinctions between the sources of any “rules” being coded is sometimes overlooked.
It is important to understand the difference between legislation and other kinds of rules. There are for two reasons for this:
First, better rules and rules as code advocates have referred to their work by the label of “legislation as code”. One intended outcome that frequently arises is the notion of machine consumable legislation. Advocates are therefore proposing significant changes to the development and implementation of legislation as specific kind of legal instrument.
Second, legislation has a particular legal status in relation to other sources of understanding what the law is. That status is superior to all other sources, generally speaking. If advocates for law-as-code approaches are setting their sights on legislation, this has a significant effect on the relative risks of producing coded models and how they will impact wider legal and regulatory systems.
Legislation is one of the two primary sources of law in New Zealand. In New Zealand, “legislation” is a statutorily defined term in the Legislation Act 2019. It “means the whole or a part of an Act or any secondary legislation”47 and includes Acts, Bills, Legislative Instruments, Other Instruments, and Supplementary Order Papers (although Bills and Supplementary Order Papers mainly relate to proposed legislation).48
Legislation is produced by the New Zealand Parliament (which is comprised of the Legislature and the Governor-General, as representative of the Sovereign), one of the three branches of government.
Because of the doctrine of Parliamentary Sovereignty, legislation is the highest law of the land. It is superior to the other primary source of law, case law.
Secondary legislation is sometimes referred to as “regulations”, but should be understood as being distinct from “regulation” more generally, which includes a range of non-legislative instruments. Secondary legislation is also defined and “means an instrument (whatever it is called) that—(a) is made under an Act if the Act (or any other legislation) states that the instrument is secondary legislation; or (b) is made under the Royal prerogative and has legislative effect.”49
Instrument is defined expansively, and “includes secondary legislation and any instrument that is not legislation (for example, an administrative document).”50
Typically, legislation is a communication written in a natural language, although it is common practice in New Zealand to express some rules as mathematical or algebraic formulae where the use of language would be unnecessarily confusing to a reader.51 For example, some statutes in New Zealand incorporate formulas, tables, examples and flow charts.
In New Zealand, the body responsible for preparing legislation is the Parliamentary Counsel’s Office (PCO). The PCO uses a range of technologies beyond simple word processors, including extensible mark-up languages (XML) and schematron validation software that identifies patterns in XML text. New Zealand’s official legislation is available online.
The PCO receives drafting instructions from government agencies when legislation is to be prepared. These instructions are subject to legal privilege, which makes studying them empirically extremely difficult if not impossible.
There is extensive academic, public sector and private sector work on the flaws in legislation and the consequences of these flaws. It is vital that any advocate for improved legislation, or critic of statutory flaws, is at least passingly acquainted with some of the material analysing why legislation is the way it is, what flaws have been identified in its preparation and maintenance, and why suggested reforms are difficult to implement. Starting points for investigation include reports by the New Zealand Law Commission,52 the legislation guidelines produced by the Legislation Design Advisory Committee,53 the Legislation Act 2012, the principles of clear drafting (and other resources on the Parliamentary Counsel Office’s website),54 the Cabinet Manual,55 and digests of Parliament’s regulation review committee – which is responsible for reviewing the lawfulness of secondary legislation.56 At one point, a treasury briefing paper shows the New Zealand Government considered a potential “Regulatory Responsibility Act” to create an Act to require better legislation.57
When speaking of coding “rules”, it is vital to consider the legal authority of different sources of rules. Legislation is fundamentally different from case law, policy, technical standards, incentive schemes, business rules, operational requirements, software systems, or any other kind of policy tool that might be described as “regulation” comprising a wider “regulatory system”. Legislation is not the same as regulation. There are fundamental differences between legislation and other legal or non-legal instruments that reflect constitutional considerations like Parliamentary Sovereignty and the rule of law.
The exact text of legislation is debated by democratically elected representatives in the House of Representatives. It is also published for public comment through Select Committee processes. There is a presumption in statutory interpretation that Parliament has deliberately chosen anything that is included in legislation as a matter of preference over any other means of expressing the rule. In short, it is assumed that Parliament meant to say what it said. Legislation represents an often artfully ambiguous statement of what the law requires people to do in order to facilitate democratic disagreement between Parliamentary factions that generally disagree on matters of policy.
We make these points as a foundation for our discussion later about how the notion of “translating” or paraphrasing legislation using different words to achieve what we presume is the same intended effect is fundamentally fraught at a constitutional level.
Hildebrandt has written a text called “Law for Computer Scientists and Other Folk”,58 which summarises important topics she has explained while teaching law to masters students of computer science.
Hildebrandt succinctly captures the way that interpretation of text is an inherent part of the anthropological shift from governing by oral expression to governing by written expression. Because legislation is written down it is much more permanent and accessible, but it must embrace its interpretability. Concomitantly, society needs methods of narrowing interpretive scope, because legislation must have meaning with some consistency across time and geographical space:59
The reach of handwritten manuscripts is far beyond that of orality, both in space (the same text can be copied and read across geographical distance) and in time (the text will survive its author and the very same text can be read by later generations). The distantiation this involves has curious implications for the interpretation of text; as a text emancipates from the tyranny of its author, its meaning will develop in response to subsequent readers that need to interpret the same text in new circumstances. The rigidity of written manuscripts, so much less ephemeral than spoken words, thus generates a need for iterative interpretation. This also results in the possibility to counter and contest specific interpretations. … In the end, the stability of text combined with the ambiguity of human language turns interpretation and contestation into a hallmark of the law, thus offering a very specific type of protection that is at the root of the legal protection offered by modern positive law.
This feature of written text is intensified further by the introduction of technologies that replicate written text, such as printing presses:60
The proliferation of text and the comparative speed of producing identical copies deepen the distantiations in both time and space between text and author, author and reader, and, finally, meaning and text. This intensifies the quest for stable meaning in the face of increased opportunities to contest established interpretation.
The need for interpretation that is core to text-driven law results in an increasingly independent position for the courts. Originally, judges are appointed by the sovereign to speak the law in his name: rex est lex animata (the king is the living law). Kings thus feel free to intervene if a court rules against their wishes. However, as the proliferation of legal text requires study as well as experience, courts increasingly distance themselves from the author of the law (the king), providing a buffer zone between the ruler and those ruled. Montesquieux’s famous iudex est lex loquens (the court is the mouth of the law) announces the end of ‘rule by law’ by the sovereign, thus revoking the old adage of rex est lex animata. This signifies the beginnings of what we now term ‘the rule of law’, based on an internal division of sovereignty into legislative, administrative, and adjudicative functions that provides for a system of checks and balances. Core to ‘the rule of law’ is an independent judiciary that is capable of sustaining legal certainty, justice, and the instrumentality of the law — if necessary, against the arbitrary will of either the legislature or the administration.
As we discuss later, setting aside any other practical effects it may engender, suggesting that interpretations of the law (or expressions of Executive will) embodied in code could replace natural-language based legislation fundamentally changes the relationship between Executive, Legislature and Judiciary.
It is important to acknowledge that ambiguity is not necessarily a sign of bad legislation. Affording latitude for interpretation is a deliberate strategy deployed by legislative drafters who use language techniques that increase the scope for interpretative flexibility when and where it is required. These techniques, such as ambiguity, vagueness, and generality, are all tools in a drafter’s toolkit for when the policy requires it.62 Those drafting techniques can be overused or underused relative to the legislative intention, but their utility should not be in question.
A trained drafter is generally aware of the complementary role between the Legislature and the Judiciary in generating law. However, drafting is frequently undertaken by persons without adequate training.
If a question of statutory interpretation comes before a member of the Judiciary, it is not necessarily a sign that the statute is inadequately drafted. Put bluntly, the advent of legal disputes around what legislation requires is not an inherently bad thing. Such disputes are a necessary part of the law-making cooperation between statute and case law, and New Zealand courts are often empowered by statute to receive applications and issue judgments “declaring” how the law should be understood.63 In fact, in such disputes the role of lawyers in New Zealand is to “uphold the rule of law and to facilitate the administration of justice”:64 i.e. provide the court with all the relevant information and interpretations necessary for it to determine what the law means and how it applies to a set of facts.
If this ambiguity or imprecision is unsuitable for “code as law” then it is the code that must move to the law, rather than the other way around. If it cannot, that is not a problem for the law. This holds true no matter how strong the desire to achieve the intended policy effect.
With that said, we fully agree that not all ambiguity or imprecision is desirable. Authors such as Kevin Ashley distinguish between semantic and syntactic ambiguity:65 the former is often desirable for various reasons, whereas the latter is seldom justified.
Equally, just because ambiguity is desirable from one perspective does not mean it is inconsequential for the people trying to ascertain what the law means. The resolution of ambiguity may require the application of judgement or assessment processes, or the need to defer to some kind of authority to generate confidence in how an ambiguous provision should be understood. This can entail practical consequences such as delay, additional procedural steps, or uncertainty. The service design perspective brought by a better rules approach can assist with understanding the way that ambiguity in drafting may shift burdens onto particular groups, and where the sticking points in a process are likely to occur.
Aside from ambiguity, statutory drafting regularly empowers a particular person with discretion to make a decision using judgement. This requires the use of interpretive techniques to understand the scope of the discretion available and how it should be exercised.
Hildebrandt, citing Ronald Dworkin, notes that discretion is a core part of the law, rather than being an area that law does not touch:66
As Dworkin argued: “[D]iscretion is not the absence of principles or rules; rather it is the space between them.” … [W]ithout rules or standards the concept of discretion makes no sense; the mere fact of being bound by a particular authority creates the need to judge whether a norm applies, what decision it calls for and how it should be performed. The rule-bound nature of discretion makes possible a discussion about the interpretation and employment of discretionary competences; it allows a learning curve by requiring those who intervene to give reasons for their actions if called to account. Those reasons are—in part—the norms that regulate their behaviour as public officials, but in the end, those reasons also include the situated interpretation of those norms. In that sense, discretion is not close to but the opposite of arbitrary rule.
We note this because:
the presence of discretion in a statute does not make modelling the law impossible, it simply requires the creation of a space within or input to that model where discretion can be lawfully exercised by a human decision-maker; and
in Part Four, we point to the way that a human decision-maker is called upon to assess whether the use of an AES to perform a legal task can be “satisfied” or the system’s “reasonable reliability”.
Legislation is just one component among many that comprise the wider system of laws and rules, some of which are not written down and cannot be easily modelled (if at all).
We can offer no better framing of legislation’s position among the other constituents of modern law than that encapsulated by Hildebrandt:67
Modern law is the set of rules and principles that determine positive law; they establish what ‘counts as’ or ‘qualifies as’ a violation of a legal norm or a legal right. The rules and principles that constitute modern positive law are generated by the binding sources of the law: legislation, case law and treaties, in combination with doctrine, fundamental principles and customary law. They are enacted by legislators and courts (that produce the binding sources of the law) and applied and thus interpreted by government authorities. In a constitutional democracy, that interpretation can be challenged and the final word on how the law must be interpreted is with independent, impartial courts.
The increasing frequency with which statutes cross reference other statutes, as well as the relationship of any given statute to a body of case law, means that interpreting the law is a process of synthesising multiple inputs into a contextually dependent output.
In short, a statute on its own does not state the law sufficiently to understand the law’s effect. Subsequently, law as code applications must account for a much wider range of legal devices and norms than just the legislation itself. It will not be uncommon to find that an attempt to code legislative effect must also incorporate the effects of a much wider range of law, some of which is not written down (e.g., constitution, norm, and custom).
On this basis, it is impossible to understand the notion of “legislation as code” as being an exercise in directly translating single pieces of legislation into a single machine executable version.
The rules for interpreting what a statutory provision means are themselves governed by statute. In New Zealand, most lawyers practicing today will have been trained to apply the Interpretation Act 1999.68 Section 5 is situated within Part 2 of the Act, entitled “Principles of interpretation”. Section 5 states:
5 Ascertaining meaning of legislation
(1) The meaning of an enactment must be ascertained from its text and in the light of its purpose.
(2) The matters that may be considered in ascertaining the meaning of an enactment include the indications provided in the enactment.
(3) Examples of those indications are preambles, the analysis, a table of contents, headings to Parts and sections, marginal notes, diagrams, graphics, examples and explanatory material, and the organisation and format of the enactment.
Subsections (2) and (3) clarify that it is not only the text of a provision that guides its interpretation, but also the broader structure of the Act, including any other “indications” provided in it. This means a statutory provision cannot always be considered in isolation from its relative structure in an Act.
We think it is notable that s 5, which itself attempts to illustrate the relationship between text and purpose, relies on metaphorical expression – “in light of”. In terms of coding the law, it is an open question how this should be interpreted and implemented.
Some advocates for “legislation as code” seem to work on the basis that legislation can be modelled without the need to consider case law. Some argue that the influence of case law can be incorporated into the way the law is modelled, but in doing so, must surely concede that the model they create reflects only one arguable interpretation of the law.
Even a person’s approach to interpreting (or, per s 5(2) “ascertaining the meaning” of) an Act must also consider case law. When it came to interpreting legislation pursuant to the Interpretation Act 1999, failing to account for leading case law on the Interpretation Act could lead to an error of law. This error would subsequently be reflected in computational modelling. In the Supreme Court of New Zealand in Commerce Commission v Fonterra Co-Operative Group Ltd  NZSC 36,  3 NZLR 767 at , the Court concluded that:69
 It is necessary to bear in mind that s 5 of the Interpretation Act 1999 makes text and purpose the key drivers of statutory interpretation. The meaning of an enactment must be ascertained from its text and in the light of its purpose. Even if the meaning of the text may appear plain in isolation of purpose, that meaning should always be cross checked against purpose in order to observe the dual requirements of s 5. In determining purpose the court must obviously have regard to both the immediate and the general legislative context. Of relevance too may be the social, commercial or other objective of the enactment.
Since the Supreme Court’s decision, section 5 of the Interpretation Act 1999 and associated sections have been updated and incorporated into ss 8-12 of the Legislation Act 2019 and now incorporates the Supreme Court’s reference to “context”.
Since section 5 of the Interpretation Act was amended by its incorporation into the Legislation Act, the equivalent provision to s 5 now includes explicit reference to the “context” of legislation, presumably to reflect the Supreme Court’s statements above.
The legislation itself prompts readers to compare s 10 to the old s 5 (as it was under the Interpretation Act), as an example of the way that even repealed law can influence the interpretation of current law. Section 10 now states:
10 How to ascertain meaning of legislation
- The meaning of legislation must be ascertained from its text and in the light of its purpose and its context.
- Subsection (1) applies whether or not the legislation’s purpose is stated in the legislation.
- The text of legislation includes the indications provided in the legislation.
- Examples of those indications are preambles, a table of contents, headings, diagrams, graphics, examples and explanatory material, and the organisation and format of the legislation.
In theory and in practice, this means that proper interpretation of any statutory provision must account for factors beyond the arrangement of the words within that provision. Not only must the provision be interpreted from its text, but also where that text sits within an Act, by reference to the wider construction of the Act, and also its context, including its social, commercial, or other objectives.
Moreover, approaches to statutory interpretation can change over time in different contexts. Academic scholarship, including judicial comment extra-judicially on that scholarship, notes that New Zealand has had three distinct “interpretive eras” in relation to taxation statutes.70
Another principle of statutory interpretation in the Interpretation Act (and now the Legislation Act 2019) is section 6:
6 Legislation applies to circumstances as they arise
Legislation applies to circumstances as they arise.
As such, Parliament anticipates that words expressed in legislation can reasonably shift in meaning over time in response to changes in circumstances, the use of language, and social context.
International and domestic human rights instruments can influence statutory interpretation, limiting the power of the Executive. Computational models of legislation would need to account for such matters.
It is well established that international human rights instruments can be used as a guide to statutory interpretation in New Zealand. Unless there is clear statutory indication, judges will assume that Parliament has not legislated contrary to its international obligations.71 This is a vital safeguard when it comes to the rights of vulnerable groups protected by international human rights instruments. For example, this approach has been applied in the context of the United Nations Convention on the Rights of Persons with Disabilities,72 and in the United Nations Convention on the Rights of the Child.73 Similar arguments have been made on the basis of international conventions protecting refugees and the impacts of climate change.74
The New Zealand Bill of Rights Act 1990 also represents Parliamentary acknowledgement that statutory interpretation can shape the balance of power in New Zealand society between different groups. Sections 4-6 of the Act deal with the relationship between the NZBORA and other enactments that appear on their plain text to limit civil and political rights. Section 6 states that:
6 Interpretation consistent with Bill of Rights to be preferred
Wherever an enactment can be given a meaning that is consistent with the rights and freedoms contained in this Bill of Rights, that meaning shall be preferred to any other meaning.
Section 4 of the NZBORA reflects the separation of powers between the legislative and judicial branches. It deprives the judiciary of the ability to “strike down” laws in the manner of other constitutional systems:
4 Other enactments not affected
No court shall, in relation to any enactment (whether passed or made before or after the commencement of this Bill of Rights),—
(a) hold any provision of the enactment to be impliedly repealed or revoked, or to be in any way invalid or ineffective; or
(b) decline to apply any provision of the enactment—
by reason only that the provision is inconsistent with any provision of this Bill of Rights.
It is uncontroversial to say that legal disputes frequently turn on how the ambiguity in a legal provision should be interpreted. The interpretive rules, principles, maxims, and norms discussed above help to ensure consistency in that process. They also impose a procedural finality on argument in the Courts, where judges choose the interpretation most persuasive to them and give reasons for that interpretation.
It is these interpretations – formulated and expressed by judges in written decisions – that hold weight. They may be incorrect, but this will only be determined by a process of appeal that brings the matter before other judges within a judicial hierarchy, who may prefer an alternative interpretation reached through a process of legal reasoning.
Ultimately, the power to say what statutory language really means in a particular context is to have the power to determine the law’s final effect, even in a situation where the meaning of the text might appear to be “plain and ordinary”.
An example which illustrates the power of statutory interpretation to dictate the law’s final effect is drawn from Australian legal history as described in a public address given by Hon Justice M J Beazley AO. Her Honour’s account describes the experience of Ada Evans, one of the first women to seek to practice law in Australia:75
Although she graduated with her cohort, practice was denied to her. The Legal Profession Act 1898 (NSW) had established a Board to approve “properly qualified persons” for admission to the bar – but the conventional thinking at that time was that a “woman” was not a “person” for the purposes of such legislation. This was despite the Acts Interpretation Act 1897 (NSW) providing that “[w]ords importing the masculine gender shall include females”.
The speech continues:
In a newspaper interview shortly after her graduation in 1902, [Evans] noted that when she had sought to be admitted, the Chief Justice had “pointed out that women were not admitted in London, and so could not be here”. In 1915, Gwyneth Bebb tested that theory in London – she brought an action against the Law Society seeking a declaration that she was a person within the meaning of the Solicitor’s Act 1843 (UK) – but without success.
In short, dictating the “correct” interpretation of legislation is a very powerful legal tool that is constitutionally reserved for the Judiciary. An interpretation of the law that is not produced by judges can never be described as “the law” – this includes those that are embedded in rules as code applications, and computational models.
Some advocates for rules as code seem to hold the view that prescriptive rules expressed in natural language do not require interpretation, and that an absence of any scope for interpretation is desirable. At a practical and a theoretical level, we do not think this can be sustained.
Any person who claims prescriptive rules do not require interpretation should be asked to nominate a specific provision that can be tested.
Highly prescriptive drafting may reduce the degree to which a rule’s interpretation is capable of reasonable argument. However, frequently, such drafting requires more words to achieve that effect, not less, thereby making statute law even less accessible to the general public and even more arduous for public servants to comprehend and model Parliament’s intent.
Some members of Executive government advocate for rules as code on the basis that others have wilfully or carelessly misapprehended the “policy intent” of a regulatory system as expressed in natural language rules, including legislation. These people appear to ignore how or why textual provisions may be interpreted differently, leading to different potential meanings.
It is true that competing interpretations are barriers to Executive government actors achieving the policy goals they are told to implement. As such, there is a will to remove these barriers – through the substitution of ambiguous natural language with unambiguous code, for example. We have documented some instances of this in an Appendix with selected excerpts from publicity materials on better rules and rules as code.
On this point, we highlight that the Judiciary has ultimate control over statutory interpretation precisely because of the risk of Executive government’s general and contextual inability to appreciate the following:
As such, measures taken by the Executive and Legislative branches which effectively remove the Judiciary, or the capacity for Judicial interpretation, raise constitutional issues.
Frequently, discussion in policy circles focuses on the way that courts or other public servants have failed to understand or implement the “policy intent” held by the original people developing the policy. By contrast, lawyers and judges focus on “Parliamentary intent”, which is primarily drawn from the words that have been used by Parliament (the Legislature) in legislation.
This distinction is essential from a rule of law perspective. “Policy intent” is drawn from the Executive branch and may or may not be reflected in legislative drafting, depending on the words that have been used and the realities of policy and political processes.
This separation of powers and the principle of the judicial branch’s independence keeps the power to decide on the interpretation of the law separate from the other branches.
Interpretation preserves the ability of individual citizens to come to an authority independent of Executive government and say “this is not what the law means, no matter what the Executive thinks”, or “this is not how the law should be understood or applied in my case for these reasons”.
We raise a useful case study from recent New Zealand legal history that illustrates the constitutional interplay between Executive, Judicial and Legislative branches, as well as the role of international human rights instruments in statutory interpretation and influence on the Executive at a political and diplomatic level. The saga demonstrates:
There has been significant legal activity around the New Zealand Public Health and Disability Amendment Act 2013, expressed fully in the case of Attorney-General v Spencer  NZCA 143. We find this example particularly relevant because it is plausible that the family carers assessment process is something that might have been semi-automated or modelled using computational models, or where it might have been seen to be valuable to remove the matters of interpretation that compromise “policy intent”.
The saga relates to the legislative and litigation steps taken by the Government and Ministry of Health to give effect to an Executive policy position: that relatives of people with disabilities should not be paid for the support they provide to a family member to the same extent as a non-family carer. Plaintiffs alleged this was discriminatory following a litigation process under the Human Rights Act 1993 and were successful before the Human Rights Review Tribunal. The Tribunal found the policy to be inconsistent with the New Zealand Bill of Rights Act 1990 and the Human Rights Act 1993.
Subsequently, the New Zealand Parliament passed a retroactive statutory legalisation of the approach that had been found to be discriminatory. Further, the amendment ousted the Court’s jurisdiction to hear any claim that the policy was discriminatory. Litigation relied frequently on the use of declaratory judgment proceedings, which seek a Court’s authoritative interpretation of what a legislative provision means.
In the Spencer case at - the Court of Appeal made this pithy statement about the distinction between the Executive’s intent, as it argued its intent was when the law was made, and what the statutory language actually says. The Court said:
 … We cannot be expected to strain Parliament’s language to incorporate by implication or corollary within the scope of that permissive definition a prohibitory [sic] policy which had the opposite purpose and effect, or to read into the plain words of the text what the Ministry now submits the words were meant to say.
 As this Court has previously observed, “the inquiry is not as to what the legislature meant to say but as to what it means by what it has in fact said”.76 To that we would add the words “Nor can the Court be expected to adopt an interpretation based on what Parliament has not said.” The political context cannot assist where the legislature elects to frame its formal response to judicial decisions in terms which plainly do not reflect the intention now ascribed to them.77 And there is nothing in the statutory purpose … which would lead us to a different conclusion.
Litigation on a similar topic continued, and later in Chamberlain v Minister of Health  NZCA 8;  2 NZLR 771, the Court relied on an international human rights instrument to influence its approach to statutory interpretation, contrary to the Executive’s intended position:78
 New Zealand is a party to the Convention on the Rights of Persons with Disabilities and its Optional Protocol. Our interpretation of all relevant legal and policy instruments must account for New Zealand’s international obligations.
Also relevant is the way that an international human rights body, the United Nations Committee on the Rights of Persons with Disabilities, exerted its influence on the Executive at a diplomatic and political level in its 2014 concluding observations:79
The Committee notes that, in 2012, the New Zealand Court of Appeal affirmed that the policy of not paying some family carers to provide disability support services to adult disabled family members constituted unjustifiable discrimination on the basis of family status. The Committee is concerned that the Public Health and Disability Amendment Act 2013 reversed this court decision by denying carers’ pay to some family members. The Committee is also concerned that these provisions also prevent some family members who are carers from making complaints of unlawful discrimination with respect to the Government’s family care policy. The Committee notes that the independent monitoring mechanism has recommended reconsideration of this matter.
The Committee recommends that the State party reconsider this matter to ensure that all family members who are carers are paid on the same basis as other carers are, and recommends that family members who are carers be entitled to make complaints of unlawful discrimination in respect of the State party’s family care policy.
Spencer also provides an opportunity to consider occasions where the legislation deliberately prevents access to the courts. This is achieved through ouster clauses (also known as privative provisions). The Legislation Design Advisory Committee have this to say on the existence of ouster clauses:80
Ouster clauses are objectionable because they interfere with the courts’ constitutional role as interpreters and expounders of the law. In general, legal obligations are enforceable by the courts. Where judicial review is ousted, it is often argued that the public body whose decisions cannot be reviewed is not subject to the law and therefore has legally unlimited power. … [T]he undoubted normative strength of the presumption against ouster clauses means that Parliament should only seek to oust the courts’ review jurisdiction in exceptional cases.
If legislation were to be published in code, the fact that it is unable to be interpreted in the same way as natural language – perceived by advocates as a feature of a rules as code approach, not a bug – effectively denies the Judiciary from performing its constitutional role in relation to legislation and ousts its jurisdiction in the manner of an ouster clause.
Legislation is produced according to a prescribed process. The details of this process do not need to be reproduced here.81 It is sufficient to say that when the prescribed process is followed, the output is the creation of a statute which is inarguably valid law – regardless of its content or effect. In New Zealand, the judiciary cannot invalidate legislation for inconsistency with a written constitution.
In our opinion, even though Parliament could enact code as legislation, it should not do so. Neither should code be given the status of legislation. Writing authoritative code within legislation itself, and giving that code the status of legislation, degrades the rule of law and the separation of powers.
This conclusion is probably unwelcome for some advocates for legislation-as-code. It is also our most important conclusion in the sense that we aim to guard against significant risks of substantial harm, even though it is difficult to predict exactly what shape that harm might take.
New Zealand has a constitutional commitment to the rule of law. This commitment is acknowledged in s 3(2) of the Supreme Court Act 2003. The Legislation Design Advisory Committee conceptualises the rule of law in the following ways.82
As part of her writing on using algorithms in legal contexts, Hildebrandt neatly captures the components of the rule of law.83
In the context of a constitutional democracy the Rule of Law has at least two requirements. First, the scope and the content of the law are determined by a democratic legislator. Second, the final decision on what constitutes the correct interpretation of the law is in the hand of the courts. … This means that individual citizens have a means to challenge the administrations interpretation of enacted law, thus preventing a mere rule by law that employs the law as a neutral instrument to achieve the goals of policy makers. Instead, constitutional democracy entails that enacted law is seen as an instrument to achieve the goals of the democratic legislator, whereby the instrument embodies the constitutional constraints that are inherent in the Rule of Law.
The Judiciary’s independence from those who make the law (the Legislature) and those who initiate and later enforce the law (the Executive) is an important component of our constitution, known as the separation of powers.
The separation of powers is about preventing the concentration of power in society and its key ingredient is that each of the three branches of government should act as checks and balances on the other branches, but otherwise not intrude into their responsibilities.
Per the Legislation Guidelines from the Legislation Design Advisory Committee, the separation of powers can be summarised as follows:
The Courts have a constitutional role that is separate to Parliament.
They are interpreters and expounders of the law.
In general, legal obligations are enforceable by the courts.
Where courts cannot or are incapable of reviewing the law, it can mean that a public body has unlimited power and is not subject to the rule of law.
Only in exceptional cases should Parliament create conditions by which the courts are incapable of reviewing the law.
As such, any reform that generates risk to any pillar of government within the rule of law or separation of powers should be treated with extreme caution.
From the academic writing we have reviewed, and based on our understanding of how law and code operate, we have concluded that enacting code as legislation creates significant risk to the rule of law and separation of powers.
When law is expressed in code, the question of how the law should be interpreted is subsumed into the code itself. This deprives the judiciary of the ability to interpret the written law to assess whether the interpretation reflected in the code is legally justified.
Hildebrandt refers to this as “algorithmic regulation” or “code-driven regulation”: 84
Algorithmic regulation refers to standard-setting, monitoring and behaviour modification by means of computational algorithms. Such algorithms may be self-executing, meaning that standard-setting integrates with behaviour modification. I call this code-driven regulation.
Hildebrandt describes the way that coded systems (by comparison with statistical or data-driven machine learning systems) function based on deterministic logic in response to the inputs provided to the system, so long as those inputs are recognised.85
Code-driven regulation depends on IFTTT … ‘if this then that’, providing the fundamental logic for all algorithmic decision systems. This type of decisional logic is deterministic, entirely predictable and basically consists of simple or complex decision trees. Whoever determines ‘this’ as a condition of the ‘that’ decides the output of the system, which has no discretion whatsoever. … The point here is that we are dealing with an entirely deterministic system that is self-executing.
This deterministic manner of operating is frequently suggested as a benefit of coded systems over natural language legal systems. There is some indication that better rules or rules as code advocates see the absolute removal of interpretation by publishing rules-as-code as a good thing.
Importantly, the deterministic nature of a coded system is not assisted at all by the suggestion that only “the logic” or syntax of a piece of legislation is being captured. Any decision made about what the logic of a piece of legislation is and how it will be captured remains subject to an exercise of statutory interpretation for the various reasons described above. It is impossible to exclude a situation where, for example, a logical connector must be read in a different manner than its plain and ordinary meaning requires, if, for example, the number of other textual, purpose-based and contextual indicators supporting that interpretation are overwhelming.
Equally, the fact that a coded system is deterministic is not mitigated at all by the way that coded models can incorporate “gaps” or inputs for human judgement or to account for ambiguity. The decision to code those inputs into the model, and to decide what those inputs are and how they relate to each other, are also matters of statutory interpretation. By encoding them, we take something that was capable of interpretation and argument – expressed in text – and create something without that possibility, and thereby oust the role of the Judiciary.
Hildebrandt clarifies that the deterministic way that code-driven regulation (or rules-as-code) operates does not remove issues of interpretation. Rather, it embeds those issues into the way the system is designed, and can thereby deprive others from understanding those interpretive decisions or challenging them through legal processes.86
Though it may seem that the overdetermination and the lack of discretion imply complete transparency and the absence of interpretability issues, this is not at all the case. Such issues are, instead, hidden in the formalization that precedes the operations of the system. … [Decisions made using code-driven regulation] can be explained by referring to the decision trees that have been implemented, but whether this explanation also justifies the decision depends on how the legal norms have been translated into computer code.
Legal rules written in human language are ambiguous, and while interpretation is inevitable, it is also essential for the rule of law and the separation of powers:87
Legal norms are expressed in human language, which is notably ambiguous, and any particular interpretation is, therefore — in principle — contestable. This is not merely cumbersome compared to code-driven regulation, but — on the contrary — also protects us from over-inclusive as well as under-inclusive legal norms.
Administrative decisions taken by code-driven regulation must thus always be contestable on the double basis of: ‘the decision is based on legal conditions that do not apply because the system got the facts wrong’, and ‘the decision is based on a wrong interpretation of the relevant legal norms’. The latter means that, even if the IFTTT standard applies, one can always appeal that this code-based standard is an incorrect application of the relevant legal norm in the case at hand.
This means code-driven legal models must be able to be contested in two ways:
This conclusion significantly influences our suggestion that coded models should, if at all, be delivered subject to the AES regime already present in some New Zealand legislation, and that the “reasonable reliability” safeguard included with that regime should be clarified.
Giving effect to the law in code changes the technological delivery of law, which will have incidental effects that are difficult to foresee.
Hildebrandt,89 as well as New Zealand Judge and scholar David Harvey,90 have relied on Elizabeth Eisenstein’s work91 to explore the link between the law as we understand it now and the affordances of printed text as a communicative medium.92
… [M]odern law has evolved from the information and communication infrastructure of the printing press, creating a body of written legal rules, written case law and doctrinal treatises that determines the substance of positive law. The systematic nature of modern legal systems builds on the need for systemisation, rationalisation and linear thinking that is inherent in the affordances of the printing press. This has triggered the growth of a class of legal professionals with the task of maintaining legal certainty in the face of proliferating legal texts (codes, cases, treaties and doctrine). The structure of modern law, with its emphasis on separate national jurisdictions, institutionalised appeal, constitutional review, litis finiri oportet, and the separation of legislator, administration and courts, has nourished the law as a relatively autonomous domain. This has eventually turned the rule by law that was typical for absolutism into the rule of law that is typical for constitutional democracy. Speaking of autonomous law does not, however, imply that law could ever function as a stand-alone device. Rather, under the Rule of Law the legal system acts as a buffer between ruler and ruled, creating the possibility to contest state-authority in an appeal to a court that is in fact supported by the authority of the state … [A]ll this cannot be taken for granted, because the Rule of Law is not only a historical artefact but also rooted in a specific ICT infrastructure that may soon be overruled by another.
In other writing, Hildebrandt captures the way that technological choices can influence patterns of behaviour. This is an essential starting point for considering how code-driven regulation, if adopted in New Zealand, might introduce unintended effects.93
[T]echnology is neither bad nor good, but never neutral. … The fact that technology is never neutral refers to the fact that any technology has normative force in the sense that it induces or enforces specific behaviour patterns and/or inhibits or rules out specific behaviour patterns.
Hildebrandt gives some examples of how the “normativity” of particular technologies have produced incidental behavioural effects.94
The term normativity denotes more than mere regularity but less than morality: speed bumps generate slow driving, books generate silent reading, the Internet triggers peer-to-peer file sharing, the Forum Romanum generated a hierarchical distinction between speaker and audience, the Greek Agora triggered peer-to-peer discussions on the marketplace, artificial light generated longer working hours, Western music notation generated harmony and counterpoint, letters of credit generated trade beyond the local environment and in the end - paper money.
Diver’s writing on “digisprudence” compares the normativity and affordances of code with the normativity and affordances of text.95 He has considered what effects code-as-law might have, based on the characteristics of code, which he summarises as being that code is “ruleish”, opaque, immediate, immutable, and pervasive. By ruleishness, we understand him to mean that code-as-law reflects the way legal rules are thought to work by some, but only in their most restrictive and anti-democratic incarnation, described as legalism. His writing includes, as he describes it:96
…an analysis of code’s regulative characteristics from a legal theory perspective, from which I develop the concept of computational legalism. This idea is borne of the parallel I observe between code’s ruleishness – its reliance on strict, binary logic instead of interpretable standards – and its conceptual equivalent in the jural realm, known as legalism. Legalism is a perspective that eschews a holistic interpretation of legal norms, instead requiring that citizens merely follow legal rules as they are presented to them, without enquiring as to their efficacy or their legitimacy beyond the question of where they come from.
Diver notes that:97
Even in the most tyrannical state there is space to interpret, and perhaps to disobey – the hermeneutic gap between the text of a norm on the page and its translation into behaviour in the world makes this at least a notional possibility. In the environments where code is designed, however, the elision of that gap is not only easy to do but is entirely standard, not necessarily through malice or intentional obfuscation (although they are certainly a problem), but simply by the underlying characteristics of code, which by nature presents norms to the end-user that ‘just are’.
Diver notes the way that the safeguards provided by interpretation can be withheld from end-users of code, and that coded systems will only reflect these legal safeguards if they are incorporated in advance in system design.98
The hard edges of code rules admit of no interpretation or latitude beyond what the designer has had the foresight (and incentive) to implement. These are strengthened by the immediacy of code: it executes without delay, imposing those potentially harmful rules without deliberation. While the end-user may in some cases have the opportunity to alter the default configuration of the rules, the literature shows that they tend not to do this, deferring instead to the perceived knowledge and expertise of the designer. In any event, the provision of an option is contingent on both the designer deciding to do so and the interface making it clear what the options are and what they mean.
Diver’s digisprudential framework is oriented towards a similar goal to the advocates for a better rules approach as we describe it. He directly investigates ways that better design processes for code (such as, we say, a better rules approach) might confer some of the legitimacy conferred on law onto code as well: in his words, “Can mechanisms for designing legitimate legal normativity be adopted to ensure the design of legitimate technological normativity?”99 This is important because:100
As code is increasingly the medium upon which other parts of social, political, and commercial life are built, it seems reasonable to assume that it will become the target of more and more law. However, laws that fail properly to be embodied in the code that they target tacitly undermine law-making as an expression of democratic will …
Policymakers and senior decision-makers should be cognisant that the notion of code-as-law is a topic subject to careful reflection over an extensive period. This scholarship should be incorporated into their thinking. But we also note that some of the suggested responses to dealing with these issues match closely the approach taken by better rules advocates. For example, by the adoption of open, transparent multidisciplinary design approaches. This is why we endorse the exploration and development of better rules approaches, subject to the caveat that any code-driven regulation that is produced must remain contestable and, ideally, always be subjugated to the authority of natural language legislation.
A core claim explored through a better rules approach is the way that co-creating law, policy and code simultaneously will result in three means of expressing the same intention. In Part 2, we explore the way that this parallel drafting process might be said to minimise a “translation gap”.
Considering our analysis above, some might say that while coded models of pre-existing law are flawed because they are only interpretations, using parallel drafting processes is a means of ensuring that the same intent is expressed identically through different methods of expression (whether natural language or code).
In response, we argue that there will always be a gap in meaning between natural language legal instruments and coded models of that instrument, even where they are produced through a process of parallel drafting.
When we are considering whether natural language law could be translated or reformulated into machine readable languages, we engage with complex questions around whether the words of a statute can be revised, paraphrased or reformulated without changing the law’s meaning or legal effect. One traditional view was that this was not possible:101
Unlike a case law rule, a statutory provision cannot be paraphrased and still be law. Its connotation may change, but its formulation is fixed. This characteristic both eases the task of the interpreter by providing him with a starting point for interpretation and complicates it by presenting, at least superficially, no leeway for interpretation through the manipulation of the form in which the rule is expressed.
Despite the orthodox position on whether statutory provisions can be translated or reformulated without changing their meaning and effect, the New Zealand Parliament has legislated for a “legislation revision programme”, which purports to do exactly that, subject to significant legislative and constitutional safeguards.
We describe this programme in detail for two reasons.
First, to demonstrate the level of constitutional assurance required when it comes to expressing the same piece of legislation in different words, even in the same language, and even without any intent to change the legislation’s effect.
Second, to note how the legislative revision programme has features that allow for the insights generated by a better rules approach to be incorporated into natural language legislation without changing its meaning, subject to constitutionally adequate oversight.
The legislation revision programme also supports our conclusion that legislation should not be written in code, and that there are two methods of incorporating the insights from a better rules approach.
The legislation revision programme is an existing method of using the better rules programme to improve legislation for implementation in digital systems, or to better express Parliament’s intent, or clarify ambiguity.
Later, we identify a separate statutory regime (around “automated electronic systems”) that might allow for “rules as code” models of the law to be operationalised within a wider legislative framework while respecting the primacy of legislation.
The purpose of the legislation revision programme is to enable progressive and systematic revision of legislation by re-enacting it: 102
in an up-to-date and accessible form, but (except as authorised …) without changing its effect.
The legislation revision programme illustrates that while Parliament may revise the way it expresses its intent through legislation, this requires significant constitutional and legislative machinery to stabilise the effect of the law despite changes in the way it is expressed.
Part 3 subpart 3 of the Legislation Act 2012 creates a scheme for statutes to be “revised”. Sections 59-62 of the Act also give legislative guidance as to how revision Acts should be interpreted by reference to “old law”, a statutorily defined term at s 59.
The Act requires the Attorney-General to seek input from the public and the House of Representatives on a programme of revisions to be initiated over a three-year period.
A revision Act’s provisions, per s 60:
are the provisions of the old law in rewritten form; and are intended to have the same effect as the corresponding provisions of the old law.
This general statutory direction can be “overridden” expressly, or “by necessary implication” – i.e., as a matter of statutory interpretation:
Section 60 is overridden to the extent that a provision of a revision Act—
(a) is expressly provided by the revision Act to be intended to change the effect of the old law; or
(b) has a meaning that is expressly or by necessary implication to a different effect than the corresponding old law provision; or
(c) is enacted, amended, or repealed by legislation that is not a revision Act (for example, an amendment Act intended to change the effect of the old law).
Section 61 requires that the old law is used to interpret the new law if the meaning of the new law “is unclear or gives rise to absurdity”:
If the meaning of a provision of a revision Act is unclear or gives rise to absurdity, the wording of the old law that corresponds to the provision must be used to ascertain the meaning of the provision.
Revisions are made by preparing a revision Bill. Section 96(2) creates a limited list of revisions that “may” be made. Revisions may:
(c) omit redundant and spent provisions:
(e) make changes in language, format, and punctuation to achieve a clear, consistent, gender-neutral, and modern style of expression, to achieve consistency with current drafting style and format, and generally to express better the spirit and meaning of the law:
(f) include new or additional purpose provisions, outline or overview provisions, examples, diagrams, graphics, flowcharts, readers’ notes, lists of defined terms, and other similar devices to aid accessibility and readability:
(g) include new or additional provisions alerting users of the revision to legislation that is not incorporated in the revision but is relevant to the subject matter of the revision:
Section 96(2) therefore presents notable opportunities to use the revision programme to incorporate insights from a better rules approach. Those insights could be incorporated in the form of basic improvements to readability through to the incorporation of flowcharts, all the way to changes “to express better the spirit and meaning of the law”.
On the plain and ordinary meaning of s 96(2), it would be possible to incorporate the outputs of a better rules process into a revision Bill. For example, a concept model is a diagram or graphic that could be appended to the legislation as an interpretive aid and legitimately used as a guide to “ascertaining meaning” of the legislative text. Further, it would be possible to use the powers at s 96(3) to make changes that go beyond bare clarification of the text as it is, for example by including “new or additional purpose provisions”.
Section 96(3) separates other powers from the powers in s 96(2), presumably because they are taken to have a greater risk of changing legislative meaning. Importantly, sub (4) states that a revision Bill “must not change the effect of the law, except as authorised by subsection (3).” Subsection (3) states a revision Bill “may also”:
(a) make minor amendments to clarify Parliament’s intent, to resolve ambiguity, or to reconcile inconsistencies between provisions (or to do all of those things):
(c) make minor amendments to update how provisions can be complied with, or operate, in a way that takes account of changes in technology if those amendments are consistent with the spirit and meaning of the law:
Notably, the “spirit and meaning” of the law is a guide in s 96(2) and (3) and “Parliament’s intent”, resolution of “ambiguity”, and reconciliation of “inconsistencies” are all legitimate reasons to use the revision powers in sub (3).
Of particular note for the better rules approach and its emphasis on digital service provision is that amendments consistent with the “spirit and meaning” of the law can be used to take account of changes in technology, including, presumably, to allow legislation to “be complied with, or operate” in digital systems. Subsection (3) therefore presents a clear opportunity to incorporate the insights of better rules approaches to the statutory revision programme without the need to enact computer code as legislation.
Several provisions also indicate a clear Parliamentary intent that any changes to the wording of the law – even if they are not intended to change its meaning – must be transparent and identifiable to unwitting readers.
Section 97(2) requires that a revision Bill’s explanatory note must “include a statement setting out, in general terms, the inconsistencies, anomalies, discrepancies, and omissions that were identified in the course of preparing the revision, and how they have been remedied in the Bill.” Section 96(2)(g) also enables the inclusion of provisions for the purpose of “alerting users” of legislation.
The Legislation Act 2019 empowers the Chief Parliamentary Counsel to authorise PCO to make “Editorial changes” to legislation, but only in tightly constrained circumstances. Section 86(2) reiterates that ss 87 and 89 “do not permit any change to the text of a provision of any legislation that, if enacted, would change the effect of the provision.“ Section 91 requires all such changes to be noted in the legislation.
It is essential to note that revision Bills go through a process of certification outside of the Parliamentary process.
Section 98 requires that revision Bills are submitted by the Chief Parliamentary Counsel to identified “certifiers”, who “may” certify a Bill “if they are satisfied that” both:
(a) The revision powers in section 96 have been exercised appropriately in the preparation of the revision; and
(b) The revision Bill does not change the effect of the law, except as authorised by section 96(3).
The certifiers “may require the Chief Parliamentary Counsel to make whatever changes they consider necessary” before certifying a revision Bill. This certificate must be provided with the Bill to the Attorney-General and a copy of such a certificate is available here by way of illustration.103 At the time of writing, only two certificates were available: one in relation to the Contracts and Commercial Law Act 2017; the other in relation to the Partnerships Act 2019.104
Certification is conducted by a group of people set out in the legislation, being:
The certifiers clearly reflect a diverse range of constitutional actors with statutory obligations to the administration of the law and the legal system itself. Further, none of the certifiers are elected by popular vote. Notably, even the judicial member of the group is someone who no longer practices a constitutional role as a member of the Judiciary, to avoid any suggestion of judicial intervention in a legislative process. “Revising” statutory wording, even for innocuous purposes and with no intent to change a provision’s meaning, is constitutionally fraught.
Notably, the way the revision programme is framed in the statute reflects Parliament’s acknowledgement that it is possible to change the legal effect of the law, even if only minor revisions are made, and even if there is no intent to change the law’s effect. It resolves this doubt by declaring its intent not to change the legal effect of the original drafting and attaching considerable safeguards from a range of constitutional actors throughout the revision process.
When law is modelled with the intent that it be machine-executable, not just machine-readable, it is the anticipated legal effect of the law that is modelled, not its drafting. To the extent its drafting is the thing being modelled, that is achieved by reformulating the statutory language, which again risks incidentally influencing the law’s effect as originally intended.106
Isomorphic modelling practices aim to create clear links between the law as drafted in its source material, and the way that source material has been modelled in a coded system, in the same way that “old law” can be used to interpret “new law” in revision Bills. This allows any gaps or changes in meaning to be independently scrutinised.
The revision bill programme is an interesting opportunity to put the better rules programme into practice: it is a good opportunity to use the program to illustrate conceptual incoherence or logical inconsistency in existing statutes by following a better rules methodology, and to suggest proposed changes to the law that would better reflect Parliamentary intent. It would be possible for better rules outputs to also then be incorporated into the statute as a guide to statutory interpretation, short of an actual “rules-as-code” operational output.
The revision bill programme is also an indication of the importance of transparent constitutional procedure when the wording of an Act is going to be reformulated, in order to independently assure that its intended effect is not changed.
If legislation were to be “translated” into code and intended to have the same legal status and effect as that legislation, we would expect that a process of similar gravity be followed. There would also have to be clear Parliamentary indications about the primacy of the natural language text over the coded model akin to the relationship between “old law” and “new law”, although only one could have primacy.
Of course, if coded models of legislation are only to be treated as an interpretation of the legislation, and able to be knocked out by declarations of inconsistency or unlawfulness, then these concerns are less significant.
The New Zealand public service deals frequently with the legacy of New Zealand’s colonial history, including breaches of te Tiriti o Waitangi – the Treaty of Waitangi. As such, public servants should be well aware of the risks of assuming that a legal instrument in one language comprehensively reflects that same legal instrument in another language.
The Treaty is New Zealand’s original lesson in the potential consequences that can flow from multilingual parallel drafting (although, we note some would argue a concept model might have helped). Equally, we recognise that breaches of the Treaty did not flow solely from the way it was drafted: there were intentional and flagrant breaches regardless of how either version is interpreted.
Legislation in New Zealand has been an instrument whereby Parliament legalises colonial acts and breaches of the Treaty. Legislation is also an instrument used to formalise treaty settlements between iwi and the Crown. New Zealand uses te reo Māori in various ways in legislative drafting.107
Some Acts have been passed in both English and Māori in their entirety, requiring a Parliamentary direction as to how any conflict in meaning should be resolved. In the Te Ture Whenua Māori Act 1993 (Māori Land Act 1993), the preamble to the Act is stated in English and Māori. The interpretation section requires the Act to be interpreted in light of the preamble, and that the Māori version of the preamble “shall prevail […] in the event of any conflict in meaning”:
2 Interpretation of Act generally
(1) It is the intention of Parliament that the provisions of this Act shall be interpreted in a manner that best furthers the principles set out in the Preamble.
(2) Without limiting the generality of subsection (1), it is the intention of Parliament that powers, duties, and discretions conferred by this Act shall be exercised, as far as possible, in a manner that facilitates and promotes the retention, use, development, and control of Maori land as taonga tuku iho by Maori owners, their whanau, their hapu, and their descendants, and that protects wahi tapu.
(3) In the event of any conflict in meaning between the Maori and the English versions of the Preamble, the Maori version shall prevail.
As previously stated, the power to interpret the law is to have the final power to say what it means and what its effect will be. In the provision above, interpretation is used as a deliberate tool to “best further the principles set out in the Preamble” to the Act, which is some acknowledgement of the way that interpretation could be used in order to undermine those principles too.
New Zealand is entering a period of greater recognition of the role, influence and value of tikanga Māori in the common law and the wider colonial legal system.108 In the Ellis criminal appeals during the 2019-2020 period, the New Zealand Supreme Court was asked to determine whether it had jurisdiction to continue with a criminal appeal after the appellant had died. The appellant argued the appeal should proceed, contrary to usual practice, as the reputation of an appellant lasts beyond death in tikanga Māori, despite the English common law approach. While predominantly a question of common law rather than statutory interpretation, counsel for Mr Ellis made the following legal submission on the Treaty and statutory interpretation in support of the proposition that tikanga is a part of the common law.109
… the Treaty of Waitangi has been held to be of such constitutional importance that it’s been read into areas of law, even when there’s no legislative reference to it. So it’s been well acknowledged that the Treaty of Waitangi has that significant place, and so it can be a relevant consideration or an interpretive aid. The Treaty, of course, imports tikanga considerations. Tikanga is both a taonga under Article 2 of Te Tiriti O Waitangi, and also tikanga being highly relevant to rangatiratanga as well. The two are interconnected. So in that sense tikanga has also been read in, in a number of different cases. Now the Crown in their submissions refers to public law. I would just point out that that’s public law in that very wide sense of the Treaty of Waitangi meaning, and so tikanga has been read in in environmental law cases, in family law and recently in respect of immigration. So tikanga principles have been imported in that respect as well.
There is every reason to think that tikanga, as part of the common law, will play an increasing role in shaping the interpretation of legal instruments. It is another source of principles that must be taken into account when modelling legislation.
Where we create computational models of the law, we encode a particular interpretation of it. We thereby risk removing the right of others to advance competing interpretations based not just on text and purpose, but also other legal instruments, including the Treaty, human rights instruments, and international conventions on the rights of indigenous peoples.
Where law is executed automatically through semi-automated decision-making tools, people subject to a decision can be deprived of a right to be heard: equally, their right to be heard at an appeal can have transformative effects for the way the law is understood.
One of the claims in Better Rules promotional materials is that legislation could be drafted “in any language”, although there is only reference to English and “code”. There is an international literature on multilingual drafting as drawn from the experiences of the European Union and from Canada in particular.110 These bodies of scholarship and practice give a useful starting point for the complexity of drafting and interpreting legal instruments set out in multiple natural languages.
Te reo Māori requires close attention to the richness of its language and vocabulary, including the whakapapa of particular words and phrases. If restricted to the public service, a better rules approach would have to account for this in the diversity and expertise of its multidisciplinary teams.
If policy is developed through a better rules approach, and rules as code models are created without the input and influence of Māori, it risks simply presenting a computational means of perpetuating the Crown’s colonial relationship with Māori and implementing the Crown’s interpretation of the law at computational scales in digital systems. As such, there is reason for extreme caution when it comes to “code as law” systems and te Tiriti.
In the preceding analysis, we have been speaking about the ability to “translate” legislation generally, but more specifically to translate it into “code”. It is important for policy-makers to understand that “code” is a term that covers a wide range of computational languages, tools, and techniques.
Advocates frequently speak of “translating” law into “code” or “machine executable languages”. With a full appreciation of what code is, it is easier to understand how “modelling” law in computational languages is a better route to comprehension.
Traditionally, for legal expert systems, the inability to comprehensively code the one true meaning of the law was perceived as being an insurmountable obstacle. Like Morris, we agree that it is not an obstacle to coding effective “interpretations” of the law:111
Here’s a thought experiment: Think of a law that you are pretty sure you understand the meaning of. Now, remove human beings from the universe. Does that law still have a significant meaning? For my part, I’m comfortable that the answer is no. Laws do not have meaning in and of themselves. They are given meaning by interpretation. It is therefore not possible to encode the meaning of the law and not merely an interpretation of it. Every possible meaning of the law, including the correct one if such a thing exists, is an interpretation. So what we are encoding is interpretations. Full stop.
The metaphor of a “model” only becomes more appropriate when it becomes clear how code works and how it differs from human “language”, despite the way computational “languages” are described.
When presented with the concept of “legislation as code”, one of the first questions anyone with expertise in computer programming will ask is: “what kind of code?”
As legal researchers, we have had to come to some working understanding of what computer code is, how it works and what it does.112 There is a long history of scholars with expertise in both law and computer science attempting to merge these two areas, or to use computers to achieve legal tasks.
At the outset, we state our belief that a significant part of the confusion around “translating law into machine languages” stems substantially from the metaphor of code as language. Again, the academic writing on what does or does not constitute a language is significant. It is sufficient to say that there are important differences between natural language, as used by humans and given meaning through a shared and developing system of social usage, and programming languages, which are a formalised system for instructing a computer system to process data algorithmically.
We have found it useful to understand “code” from the starting point of what a computer is and what it does. The essential elements of a computer system are inputs, a processor, memory, and outputs. The processor and the memory are used to take a data input and produce a data output.
It is also helpful to briefly state a definition of data, including digital data, and binary or machine code. A core problem in computer science deals with how to represent knowledge about the world around us in a way that a computer system can process. Knowledge is represented through the use of digital data. Computer systems receive information inputs at a fundamental level in binary notation. Any number or letter can be represented in base-2 format, as a string of 1s and 0s. This means any knowledge that can be represented numerically can be represented in a computer system.
Binary notation is captured in a computer system, at its most fundamental level, in the computer’s hardware, through a system of switches, or transistors, that close and open circuits. This system of transistors can be used to represent a 1 or a 0 by being switched to “on or off” (by closing or opening a circuit and controlling whether an electrical current can flow).
Representing numbers in binary can result in extremely long strings of digits. For this reason, programming languages use abstraction: programmers seldom engage with computer code at the level of 1s and 0s. Instead, there are formalised systems of instructing computers (processing units with access to memory) how to process those 1s and 0s, or data. Different systems of formalising those instructions are referred to as different “languages”. Unlike natural languages, if they are used incorrectly, there is very little if any leniency for error.
In this project, we have found it more helpful to think of code as a tool, rather than a language. This leads to much more nuanced thinking when it comes to the notion of “translating” between one language and other without loss of meaning.
In the past, systems of computation have also been attempted without the use of electronic circuits. In the Reserve Bank headquarters in Wellington, New Zealand, there is a restored hydraulic model of an economic system. This is a useful way of thinking about the idea of modelling law using code. Code, and machine executable languages, are simply a system of knowledge representation and processing using digital systems. In order to use code to model a legal system, decisions have to be made about how best to decide what parts of the system are being modelled and how the model operates.
One important distinction to appreciate is between legal materials that are machine-readable, and legal materials that are machine executable. This distinction is also important for considering the academic history of the idea of “law as code”.
Machine readable: this refers to the ability to use automated processes for the structuring, storage and retrieval of legal text. The use of machine-readable languages in law generally involves no fundamental change to the basic text-based nature of the law. A dominant example is the use of mark-up languages, such as XML, which create “wrappers” for text that enable a computer system to understand the semantic or formal significance of a string of text. Scholars attempt to create formalised standards that allow the various components of legal materials to be signified using machine languages in ways that are predictable and formalised within or between jurisdictions. Materials that are machine readable are still predominantly intended to be used as text, not as code. Despite the important distinction between machine readable and machine executable law, many research programmes deal with both. That is because the creation of machine executable law from legal text is manually intensive. The goal of many law-as-code investigators is to automate the process of extracting legal norms from legal text: to do that, there has to be some formalisation of that legal text to enable machine processing. To some extent, the better rules approach as we define it could be used to produce standardised law that is more amenable to machine readability using standards such as LegalDocML / Akoma Ntoso, but this is not the focus of contemporary rules-as-code or better rules advocates.
Machine executable: making law machine executable refers to developing or modelling laws into forms that enables computers to model the effect of the law. By machine executable laws, it is intended that computational systems can be given data inputs that represent legally significant events, and that the effect of the law can be modelled in response to that event. The computational rules that comprise the modelling system are intended to mimic the effect of legal rules in the system being modelled. Machine executable law is therefore intended to answer legal questions.
We note some also insist on a distinction between computational, programming and machine languages. We do not think that distinction has any implications for our findings but we raise it for completeness.
Code languages are formalised. They are working tools that like any other human-designed tool have limitations, or intended purposes, that make them more or less suitable for particular uses than others. They are usually subject to some kind of oversight or standard that formalises the way they are used.
One important limitation to consider is that a computational model of the law written in “code” can only really be operationalised in a computer system. That sits in contrast to law written in natural language, which applies everywhere within its jurisdiction at all times, and is enforced by humans and institutions who exercise discretion about whether or not enforcement will be pursued. Law written in natural language can be relatively brief because many assumptions and background factors can be taken for granted. The same ruleset drafted in computational languages cannot always take such matters for granted, and paradoxically, becomes much longer and more prescriptive by comparison.
Once code is understood as being a “coded model” using a range of tools and techniques, this makes the illusory equivalence between “law in natural languages” and “law as code / machine executable languages” more nuanced. We have found it useful to consider the following hypothetical:
While perhaps fanciful, this hypothetical scenario does reveal important assumptions that are imported when we speak of “translating” law into other “languages”. Those assumptions are easier to avoid when the specific differences between natural languages and machine-readable languages are clarified through the example.
In this part we have explained in some detail why we believe code should not be given the status of legislation in New Zealand based on a range of practical and principled objections.
Pragmatically speaking, the notion of translating only one piece of legislation cannot be sustained, given the way legislation sits as one thread in a wider fabric of legal instruments.
We have explained how “interpretation” is an unavoidable and inherent process for ascertaining what the law means. That interpretation, as a matter of pragmatism and principle, must be able to be contested. Where a coded model of the law is created, an interpretation of the law is adopted, which may not always be able to be scrutinised or contested.
Further, New Zealand’s constitutional arrangements require different institutions to be responsible for setting legislative text as opposed to conclusively interpreting it. A key plank of this constitutional arrangement is that it is the Judiciary, not the Executive, who has authority to declare the final and correct interpretation of the legislation.
We have made the argument that legislation cannot be prepared in two languages at once without serious risks to changing the meaning and effect of the law and the way that Parliament needs to give explicit direction as to which version of the same law expressed differently is to have the greatest weight.
We have also pointed to the way that the insights generated by following a better rules approach can already be implemented by “revising” legislation through a revision programme that incorporates careful constitutional safeguards. One of these reasons for revising the legislation is to enable it to be implemented in digital systems or to take account of changes in technology.
Finally, we explained how the idea of “translating the law into machine readable languages” tends to be misleading, and that a better metaphor for understanding is to “model an interpretation of the law”.
In Part Four, we explain how the notion of “legislation as code”, a better rules approach, or rules as code can nevertheless be incorporated into the New Zealand legal system while respecting these constitutional features.
Legislation Act 2019, s 5. ↩︎
Ibid, and definition per the New Zealand government Legislation website: <https://www.legislation.govt.nz/glossary.aspx#l>. ↩︎
Legislation Act 2019, s 5. ↩︎
Parliamentary Counsel Office “Principles of Clear Drafting”. See “Other Drafting Tools”: <http://www.pco.govt.nz/clear-drafting#otherdraftingtools>. For a basic legislative example, see Accident Compensation Act 2001 Schedule 1 clause 34. ↩︎
See for example: “Legislation Manual: Structure and Style” (May 1996) NZ Law Commission, r 35; “Review of the Statutes Drafting and Compilation Act 1920” (May 2009) NZ Law Commission r 107; “Presentation of New Zealand Statute Law” (October 2008) NZ Law Commission in conjunction with Parliamentary Counsel Office, r 104. ↩︎
Legislative Design Advisory Committee “Legislation Guidelines: 2018 edition” (2018): <http://www.ldac.org.nz/guidelines/legislation-guidelines-2018-edition/>. ↩︎
See “Instructing the PCO” Parliamentary Counsel Office website <http://www.pco.govt.nz/instructing-the-pco/>. ↩︎
Department of Prime Minister and Cabinet “Cabinet Manual” (2017): <https://dpmc.govt.nz/our-business-units/cabinet-office/supporting-work-cabinet/cabinet-manual> ↩︎
Dean R Knight and Edward Clark Regulations Review Committee Digest (6th ed, New Zealand Centre for Public Law, Wellington, 2016). ↩︎
See: Regulatory Impact Statement “Regulating For Better Legislation – What Is The Potential Of A Regulatory Responsibility Act?” (2 February 2011), New Zealand Treasury. ↩︎
Mireille Hildebrandt Law for Computer Scientists and Other Folk (2020, Oxford University Press, Oxford, UK). Open access available: <https://fdslive.oup.com/www.oup.com/academic/pdf/openaccess/9780198860884.pdf>. ↩︎
Ibid at p 4. ↩︎
Ibid at p 5. ↩︎
Ibid at p 6. ↩︎
Dickerson, R, “The Diseases of Legislative Language” (1964) 1 Harv J on Legis 5. ↩︎
See, for example, Declaratory Judgments Act 1908 and Resource Management Act 1991 at s 310-313. ↩︎
Lawyers and Conveyancers Act (Layers: Conduct and Client Care) Rules 2008, s 2. ↩︎
See Ashley, K D Artificial Intelligence and Legal Analytics: New tools for Law Practice in the Digital Age (2017, Cambridge University Press, Cambridge, United Kingdom) 38-42. Chapter 2 of Ashley’s book is also notable for the way it discusses ambiguity, vagueness, statutory interpretation, computational modelling before turning to the topic of business rules systems and business process modelling, a core part of the heritage of a better rules approach. ↩︎
Hildebrandt, M “Algorithmic regulation and the rule of law” (2018) Phil. Trans. R. Soc. A 376:20170355 at 5-6. ↩︎
Ibid. We note that the role of treaties is a matter that varies among jurisdictions and depends upon whether they are monist or dualist as a matter of international law and their domestic constitutions. ↩︎
We note that the Interpretation Act 1999 has been updated and incorporated into the Legislation Act 2019. We discuss this next. ↩︎
Notably, the Court was being asked to consider evidence from an academic witness about what the plain and ordinary meaning of a word should be (the word “capital” in a contract). This is illuminating for any suggestion that parallel-drafted code might dominate the interpretation of natural language legal instruments. See Justice Susan Glazebrook “Statutory interpretation in the Supreme Court” an address given to the New Zealand Parliamentary Counsel Office in 2015 based on S Glazebrook “Do they say what they mean and mean what they say? Some issues in statutory interpretation in the 21st century” OtaLawRw 7; (2015) 14 Otago LR 61. ↩︎
See Burrows and Carter Statute Law in New Zealand as cited in Glazebrook, ibid. ↩︎
Ye v Minister of Immigration  NZSC 76,  1 NZLR 104 at ; and New Zealand Air Line Pilots’ Assoc Industrial Union of Workers Inc v Director of Civil Aviation  NZCA 27,  3 NZLR 1 at –. ↩︎
Chamberlain v Minister of Health  NZCA 8;  2 NZLR 771. ↩︎
Above n 70. ↩︎
See the helpful summary of the litigation culminating in Teitiota v Ministry of Business Innovation and Employment  NZSC 107 (20 July 2015) available from Library of Congress “New Zealand: ‘Climate Change Refugee’ Case Overview”, last updated 30 December 2020 <https://www.loc.gov/law/help/climate-change-refugee/new-zealand.php>. ↩︎
Hon Justice M J Beazley AO “100 Years of Women in Law in NSW” 18 October 2018, Sydney, Australia. President, New South Wales Court of Appeal. At para 6. Her Honour relies on Justice Virginia Bell AC, ‘By the Skin of Our Teeth –The Passing of the Women’s Legal Status Act 1918’ (Speech delivered at the Francis Forbes Lecture, NSW Bar Association Common Room, 30 May 2018). ↩︎
Citing Donselaar v Donselaar  NZCA 13;  1 NZLR 97 (CA) at 114. ↩︎
Citing Daniel Greenberg (ed) Craies on Legislation (10th ed, Sweet & Maxwell, London, 2012) at [220.127.116.11]: “it would clearly be most undesirable for the courts to begin to attach significance to what Parliament does not do, the manner in which it does not do it and the reasons for which it does not do it.” ↩︎
The Court relied on Ye v Minister of Immigration  NZSC 76,  1 NZLR 104 at ; and New Zealand Air Line Pilots’ Assoc Industrial Union of Workers Inc v Director of Civil Aviation  NZCA 27,  3 NZLR 1 at – for these propositions. ↩︎
Concluding observations on the initial report of New Zealand: Committee on the Rights of Persons with Disabilities (Geneva, 31 October 2014) Adopted by the Committee at its 12th session: <https://digitallibrary.un.org/record/791464?ln=en>. ↩︎
Legislation Guidelines: 2018 Edition, Legislation Design Advisory Committee Chapter 28 “Creating a system of appeal, review and complaint”, Part 1 “Does the legislation seek to exclude or limit the right to apply for judicial review?” ↩︎
See, for example, “How laws are made”, NZ Parliament: <https://www.parliament.nz/en/visit-and-learn/how-parliament-works/how-laws-are-made/>. ↩︎
Legislation Guidelines: 2018 edition, Legislation Design Advisory Committee, Chapter 4: “Fundamental constitutional principles and values of New Zealand law”: <http://www.ldac.org.nz/guidelines/legislation-guidelines-2018-edition/constitutional-issues-and-recognising-rights/chapter-4/>. ↩︎
Hildebrandt, M “Legal protection by design: objections and refutations” (2011) 5(2) Legisprudence 223 at 234. ↩︎
Hildebrandt M. “Algorithmic regulation and the rule of law” (2018) Phil. Trans. R. Soc. A 376:20170355 at pp 2-3 ↩︎
Ibid at p 2. ↩︎
Ibid at 2. We note the same point is cited above in Part 2 in our reference to Greenleaf et al (2018). ↩︎
Ibid at 3. ↩︎
Ibid at 3. ↩︎
Hildebrandt (2011) “Legal protection by design” at 236. ↩︎
David Harvey “Collisions in the Digital Paradigm: Law and Rulemaking in the Internet Age” (Hart Publishing, Oxford, 2017). ↩︎
See for example E. Eisenstein, The Printing Revolution in Early Modern Europe (2nd ed, Cambridge University Press, Cambridge 2005). ↩︎
Ibid at 236. ↩︎
Hildebrandt (2011) Legal Protection By Design, at 239. ↩︎
See Diver, L E “Digisprudence: the affordance of legitimacy in code-as-law” (April 2019, PhD Thesis, University of Edinburgh). ↩︎
Ibid at p 95. We note the parallels between Diver’s summary and the statement in Appendix: Selected Statements from Publicity Materials that “people don’t want legislation in and of itself. They want the results of the legislation.” ↩︎
Ibid at p 32. ↩︎
Ibid at p 141. ↩︎
Ibid at p 301. ↩︎
Ibid at p 304. ↩︎
Miers, D, Page, A “Teaching legislation in law schools” (1980) 1(1) Statute Law Review 23 at p 25. ↩︎
Legislation Act 2019, section 92. ↩︎
Link to the Certification of the revision Bill for the Contract and Commercial Law Act 2017: http://www.parliament.nz/en-nz/pb/presented/papers/51DBHOH_PAP69057_1/legislation-act-2012-contract-and-commercial-law-bill. ↩︎
The revision programme is described on the Parliamentary Counsel Office website: http://www.pco.govt.nz/revision-programme/. ↩︎
Law Commission Act 1985. ↩︎
We note that some approaches involve creating knowledge bases directly from the language of legislative text. Two examples are DataLex, and the RASE Require1 tool from AEC3 shared with us by Nick Nisbet. ↩︎
See by way of illustration and analysis Tai Ahu “Te Reo Māori as a language of New Zealand Law: the Attainment of Civic Status” (2012, LLM Thesis, Victoria University of Wellington) at 58 and 70 et seq, and at 64: “If Māori is to eventually develop into a language of statutory enactment alongside English, there must be some legal mechanism to resolve potential conflicts in meaning between the two languages”. ↩︎
For example see Takamore v Clarke  NZSC 116. ↩︎
R v Ellis  NZSC Trans 19 at p 7 per Counsel N R Coates: <https://www.courtsofnz.govt.nz/assets/cases/Hearing-date-25-June-2020-tikanga-hearing.pdf>. ↩︎
See for further reading by way of example: Robertson, C “Multingual legislation in the European Union. EU and National legislative-Language Styles and Terminology” (2011) 9(1) Research in Language 51; Leung J “Statutory interpretation in multilingual jurisdictions: typology and trends” (2012) 22(5) Journal of Multilingual and Multicultural Development 481; Sin, K K “Out of the Fly-Bottle: Conceptual Confusions in Multilingual Legislation” (2013) 26 Int J Semiot Law 927; Schafer, B “Formal Models of Statutory Interpretation in Multilingual Legal Systems” (2017) 38(3) Statute Law Review 310; Cao, D “Inter-lingual uncertainty in bilingual and multilingual law” (2007) 39 Journal of Pragmatics 69. ↩︎
Morris, J “10 Questions and Answers about Rules as Code” (26 May 2020) Medium: <https://roundtablelaw.medium.com/10-questions-and-answers-about-rules-as-code-a26ecc091828>. This approach is reflected in Morris’ LLM Thesis, cited above. ↩︎
We reiterate the multiplicity of “code” as a homonym. It can refer variously to legal codes, cryptographic codes, and the research methodologies of “coding” text into categories. ↩︎