HLA Hart’s Concept of Law and the Problem with Computerized Officials

By: Sonia Saroj Anand Knowlton


In a world where technology is becoming increasingly integrated in the law and its enforcement, it is realistic to wonder whether legal officials could ever be completely computerized. In HLA Hart’s renowned The Concept of Law, the officials of a legal system play a critical role in upholding the system.[i] However, Hart’s concept of law likely assumes that legal officials are human. If the officials were computerized, would they be able to operate effectively within Hart’s concept of law?[ii] The answer is no, for the main reason that computerized officials do not have the capacity to “accept” Hart’s Secondary Rules. I am not arguing that we should have a legal system that is operated entirely by computerized officials, nor that Hart’s concept of law is inaccurate. I am instead using Hart’s concept of law to demonstrate potential challenges with a legal system run by computerized officials.

Hart’s concept of law

In Hart’s concept of law, there are two tenets that are necessary for a legal system to function: (1) private citizens must generally accept rules of obligation (the Primary Rules), and (2) officials must accept the Rules of Recognition, Change, and Adjudication (the Secondary Rules) as “common public standards of official behaviour.”[iii] Hart explains that Primary Rules are rules about what conduct is acceptable and what conduct is unacceptable in society. They create an obligation for people to act in a certain way. Primary Rules are exemplified by the illegality of murder, breaching contractual duties, and tax evasion. Secondary Rules, in contrast, are rules about the Primary Rules themselves. For example, Secondary Rules confer power and create procedures to validly add new and change existing Primary Rules.[iv] When someone takes an “internal point of view” of either of these types of rules, they believe in the validity of the rule and use it as a standard of conduct; they internalize the legal rule as a rule of their own.[v]

The problem with computerized officials

Computerized officials could not properly “accept” the Secondary Rules, one of the tenets of Hart’s concept of law, because they cannot take an internal point of view. In my interpretation of Hart’s position, people only accept a rule because either they have an internal point of view towards the rule or because they are afraid of the consequence of not following the rule.[vi] I will assume that computerized officials do not have the capacity to experience fear, and thus do not have the capacity to fear the consequences of not following a rule. Therefore, to accept the Secondary Rules, computerized officials would have to take the internal point of view towards them. However, computerized officials cannot take an internal point of view, regardless of the rule, because they cannot adopt a self-critical attitude, a necessary condition of having an internal point of view according to Hart.[vii] This is largely due to the fact that computerized officials are not capable of self-reflection. Self-criticism cannot be programmed into a machine because it is a quality that involves reflecting upon one’s own behaviour, thoughts, and beliefs.[viii] A machine cannot, on its own, reflect on why it made an inaccurate decision, even if it is able to acknowledge that its output was, itself, inaccurate.[ix] Ultimately, computerized officials’ lack of internal point of view would undermine the officials’ ability to properly execute Secondary Rules in Hart’s concept of law.

To illustrate this difficulty, I will present a hypothetical involving computerized legislators that, without taking an internal point of view, amend the law as external observers. In Hart’s concept of law, the process of adding new rules is done in accordance with the “Rules of Change,” among the Secondary Rules, which include amendment procedures for how rules are allowed to evolve validly.[x] In the scenario of computerized officials, let us assume that the external programmers initially program all of the Primary Rules into the system such that they would be enforced by the computerized officials, and similarly, program the Secondary Rules into the system such that they would be followed by the computerized officials. Since, as I have argued, computerized legislators cannot take an internal point of view towards rules, the amendment procedure for adding new rules would be based off of the computerized legislators acting as external observers. Programmers could enable computerized legislators to add new rules arising from widely intersecting patterns of behaviour in members of the public. Hart acknowledges that an external observer can accurately predict patterns of behaviour in a group of people by “correlat[ing] deviation with hostile reaction.”[xi] A predictive mechanism of observing widely intersecting behaviour, combined with a statistical analysis of when the deviation from the norm is met with sufficiently “hostile reactions,” could signal to a computerized legislator that it is time to incorporate a behaviour into the law. If rules could be updated in this way, we would have an amendment procedure for the Rules of Change. An internal point of view of these Secondary Rules would not be required because the method involves scientific, fact-stating, and predictive mechanisms, reflecting only the rule’s external aspect.

However, Hart would argue that this procedure would not be an effective Secondary Rule. Hart explains that externally observable deviations from the normal standard of behaviour will only be a sign that a consequence or sanction will follow, but will fail to convey why people follow the normal standard of behaviour in the first place. To explain why this is problematic, Hart gives the example of people stopping at a red light. Hart explains that, for an external observer, a red light will be a signal that “people will behave in [a] certain [way]” (i.e. stop at the red light), but will not indicate that the red light is actually the very reason why people behave in a certain way (i.e. that the red light causes the norm of stopping).[xii] Therefore, an externally observing computerized legislator, operating through predictions based only upon socially convergent behaviour and expected hostile reactions, would “miss out [on] a whole dimension of the social life of those whom he is watching”[xiii] by not understanding why people conform to the given social standard. Without an internal point of view from the legislators, this amendment procedure would not be effective in Hart’s concept of law.


Because computerized officials cannot accept Hart’s Secondary Rules, a society in which officials are computerized would not function in accordance with Hart’s concept of law. I have demonstrated the difficulties that arise in lawmaking where a computerized legislator operates without an internal point of view. Perhaps, if the programmers of the computerized officials were also considered officials, human sentience would be involved, and the human programmers could compensate for the computerized officials’ inability to self-reflect. Then, some of the officials could take an internal point of view and could accept Hart’s Secondary Rules. However, we are then back to “square one,” with the necessity of non-computerized officials for Hart’s concept of law remaining intact. It may one day be possible for computerized officials to adopt human sentience (and thus accept Hart’s Secondary Rules), but that day has not yet come. Regardless of what Hart would think, I hope that I have raised doubts about the efficacy of our legal system if it were operated solely by computerized officials. That said, a system only partially operated by computerized officials seems like a much more plausible next step in the evolution of our legal system as it exists today.

About the author: Sonia Saroj Anand Knowlton is a Juris Doctor candidate at the University of Ottawa (2023). Sonia has a Bachelor of Science (Honours) in Mathematics from McMaster University (2020).

The opinions expressed in this text belong to the author and do not reflect those of their position(s) nor of their affiliated institution(s).

[i] See HLA Hart, The Concept of Law, 3d ed (Oxford: Oxford University Press, 2012). In this paper, I will refer to Hart’s understanding of the law outlined in The Concept of Law as “Hart’s concept of law.” I will call the officials of a legal system “legal officials,” or simply “officials.” I will take this term to include judges, legislators, law enforcement, and elected officials, although Hart does not necessarily specify which of these roles are included in the umbrella term “officials.”

[ii] These questions are also partly answered in Brian Sheppard, “Warming Up To Inscrutability: How Technology Could Challenge our Concept of Law” (2018) 68:Supplement 1 UTLJ 36.

[iii] Hart, supra note i at 116.

[iv] Ibid at 94–95.

[v] Ibid at 57, 89.

[vi] Indeed, to Hart, the roles of officials executing Secondary Rules should not just be seen as “scientific, fact-stating, or predictive,” because that belief only “reproduces their external aspect” and does not do justice to their internal aspect. Ibid at 99.

[vii] To Hart, a necessary condition to having an internal point of view is a “critical reflective attitude to certain patterns of behaviour as a common standard, and that this should display itself in criticism (including self-criticism), [and] demands for conformity.” Ibid at 57.

[viii] That is, a machine learning system cannot properly reflect on its own behaviour. It uses a set of data to increase the accuracy of its predictions, but when a prediction goes wrong, the machine must be told that it is incorrect by an outside source (the programmer).

[ix] Alternatively, Sheppard claims that computers cannot take an internal point of view towards a social rule because “[t]o acknowledge a reason means, at the very least, that the acknowledger has some capacity to disregard the reason, compare its weight against countervailing reasons, or follow the reason,” which machines do not have. Sheppard, supra note ii at 55. It is also argued that there is a real potential for artificial intelligence to develop human sentience. See e.g. Hartini Saripan & Nurus Sakinatul Fikriah Mohd Shith Putera, “Are Robots Human? A Review of the Legal Personality Model” (2016) 34:6 World Applied Sciences J 824. For the purposes of this argument, I will assume that this development has not yet occurred.

[x] Hart, supra note i at 95–96.

[xi] Ibid at 89. Sheppard refers to this mode of lawmaking as a form of statistical correlation. Sheppard, supra note ii at 55.

[xii] Ibid at 90.

[xiii] Ibid.