Norms, Algorithms, and Programs

by Enrico Nardelli

(versione italiana qua)

Periodically, when something goes wrong in the use of computer systems in procedures subject to public scrutiny (this being the most recent case), the media talk about "algorithmic discrimination," "algorithmic dictatorship," and similar expressions coined by journalists hunting for a headline that cuts through — though, unfortunately, with a rather limited grasp of the subject matter. As evidence, please consult figure 2.5 (p.24) of the AGCOM document "Observatory on Journalism" dated November 23, 2020, and read the related press release, which states, verbatim, that "media professionals lack a sufficient level of specialist knowledge (understood in particular in terms of academic training) to cover economic, financial, scientific, and technological events and issues". "Don't shoot the messenger", as they say — I hope my friends and colleagues in that field won't hold it against me; I do respect them.

Having clarified this point, I would like to note that in the digital world the term "algorithm" is open to slightly ambiguous interpretation. This is because, historically speaking, algorithms were invented as a method by mathematicians (for the few who still may not know, I should mention that the term algorithm derives from the epithet al-Khuwārizmī of the ninth-century Arab mathematician Muḥammad ibn Mūsa, who was a native of Khwarizm, a region of Central Asia) to describe with precision, to other mathematicians, the procedures to be followed in solving problems.

In informatics, this term was borrowed to describe a problem-solving procedure in a way entirely independent of any specific implementation in the form of a computer program written in a given programming language. The critical issue — and the resulting ambiguity — arises from the fact that it is the computer program (and not the algorithm from which it is derived) that is actually executed by the computer (technically, by the automaton). And in the transition from algorithm to program, which is a genuine act of translation, much like the translation of literary texts, there is always a risk of betraying the original.

Coming back to where we started, it is essential to bear in mind that legal norms are already algorithms — and in particular, they are the algorithms that regulate and organize social life: if you break this rule, then you receive this punishment. Their executors are jurists — I use this term loosely to refer to legal professionals. Seen this way, a perfectly proportional parallel emerges: the algorithm is to the mathematician what the norm is to the jurist. It is no coincidence that both in academia and in the professional world, hybrid profiles combining informatics and law are becoming increasingly common. This mirrors a similar process that has been underway for much longer at the intersection of informatics and biology. But that is a different story.

In law, this social superstructure of written norms and their interpreters — much like mathematics — has functioned perfectly well for millennia: law based on written statutes has been in use in the "Bel Paese" for many centuries before it became a nation. In other, more ancient civilizations, its history stretches back even further. By and large, there have been no major problems, beyond the natural human tendency — which we might consider inseparable from human nature — of some individuals to regard themselves or others as "more equal than the rest". As is often said of democracy, a social system grounded in a written body of law and well-trained interpreters, for all its imperfections, remains better than any alternative.

Therefore, when legal norms — which are already algorithms — are translated into computer programs and something goes wrong in their execution, it is not accurate to speak of an "algorithmic error". One should instead speak of a "programming error". The device that runs the program, operating in a purely mechanical and entirely predetermined fashion, never makes mistakes. If the result produced by execution does not match what the algorithm intended, the error lies in the computer program. To continue the parallel with what happens when the executors are human: it is akin to a jurist misinterpreting a norm. Except that, in the vast majority of cases, being a human being endowed with intelligence, common sense, and experience, the jurist has a series of "safety nets" and "guardrails" that prevent them from distorting the norms they are called upon to interpret.

Far more common, however, is the case where a set of norms, once translated into a set of computer programs, becomes something subtly different. Sometimes this is because the conversion is typically carried out by informatics professionals who do not possess deep expertise in the subject matter and who are not provided with sufficient information by those requesting the translation. Other times it is because the norms themselves, having been written with human executors in mind — executors endowed with intelligence and experience — do not account for every possible scenario. Nor do they need to, just as when we ask a child to walk toward someone, we do not need to tell them to avoid tripping over obstacles, because they can figure that out on their own. The executors in informatics — PCs, tablets, smartphones and the like — simply do not figure things out on their own.

The problem increasingly afflicting the automated execution of administrative procedures through computer programs is nothing other than the latest manifestation of the difficulty of building a computer system that fully meets its specifications — the greatest unsolved problem in software engineering.

Yet, since these administrative procedures are now woven into all of our daily lives, these challenges — once confined to domain experts — are now something we all have to contend with.

Paraphrasing one of the most famous closing lines in cinema history, we might say: "This is informatics, sweetheart, informatics. And there's nothing you can do about it. Nothing". In reality, there is something we can — and must — do. We need to start teaching informatics in schools as a scientific discipline, just as we do with mathematics, so as to raise generations of citizens capable of understanding the difference between algorithms and programs, and the difficulties and pitfalls lurking in the process of translation.

--
The original version (in italian) has been published by "Key4Biz" on 21 February 2022.