I care very little if I am judged by you or by any human court; indeed, I do not even judge myself

(1 Corinthians 4:3)

I. Contemporary Understanding of the Right to a Fair Trial and the Potential Impact of Artificial Intelligence

The notion of the right to a fair trial has evolved over time. Taking Poland as an example, the beginnings of the right to a fair trial were rooted in the privileges of the gentry (or, oversimplified, the aristocracy). This dates back to the XV century and was first expressed in the statutory limitations of the royal power of expropriation and the rule of subjecting the gentry only to adjudication based on a written law (the so-called Czerwińsk Privilege of 1422), which was accompanied by the neminem captivabimus nisi iure victum principle. The right to a fair trial was strengthened by the first Constitution of 1791 and further developed during the Second Republic (1918-1939). Gradually abolished during the communist era (roughly 1944-1989), the right to a fair trial was revived as early as the late 1980s and “flourished” again with the rise of democracy after the collapse of the communist rule in 1989. Sadly, it took a great hit after 2015 under the present far-right government.¹

More generally, the more power is (at least allegedly) vested with the judiciary (both international and national), the more weight the right to a fair trial carries.² This ratio is like a litmus test of democracy where “fair-trial guarantees […] are guided by the aim of upholding the fundamental principles of the rule of law and the separation of powers.”³

Fundamental international documents devoted to human rights share a relatively common definition of the right to a fair trial: the Universal Declaration of Human Rights (UDHR) in Articles 8 and 10, the International Covenant on Civil and Political Rights (ICCPR) in Art. 14, the American Convention on Human Rights in Art. 8, (to a lesser extent) the African Charter of Human and Peoples’ Rights in Art. 7, the European Convention on Human Rights (ECHR) in Arts. 6 and 13, and the Charter of Fundamental Rights of the European Union (ChFR) in Art. 47 – all rather uniformly refer to nine elements of the right to a fair trial, i.e. fairness, public hearing, reasonable time, independence and impartiality of the trial court, lawfulness of judicial appointment, right of a party to be represented and to have legal aid free-of-charge, and (implicitly – e.g. in the ECHR, or explicitly – e.g. in the ChFR) the right to effectiveness of judicial protection. The national constitutions of the European states all refer to the safeguards of the right to a fair trial in much the same way.⁴

Perhaps one may even claim that the notion of a fair trial, as matters stand, amounts to a pre-existent⁵ (or extant) constitutional notion, i.e. a concept that does not require defining because everyone is already familiar with it. As for these nine criteria at the heart of a fair trial, it seems pretty clear that an AI-driven judiciary is likely to have a positive effect on the expeditiousness of proceedings whilst making no difference to the right to legal representation or legal aid, or to the right to a public hearing (i.e. access of the public to the hearing and to the pronouncement of judgments⁶). However, what remains unclear is whether the introduction of an AI-based judiciary might adversely affect the (“sub”) rights to independence and impartiality (and lawful establishment) of the trial court or to fairness of proceedings. This will be discussed further in the next sections.

1. Independence, impartiality, and lawfulness of establishment of the trial court

Legal scholars have raised certain objections against the use of AI in the courtroom concerning judicial independence⁷. For example, Nowotko held that “the court which issues judgments by means of an IT system based on artificial intelligence cannot be independent in the meaning of independence in adjudication”⁸, and Gentile suggested that

“any influence exercised by a state’s executive or the legislative, for instance, over data centres used to digitise judicial decisions, the selection of the training data for neural systems, or the very design of the algorithm used in the courtroom would be liable to raise doubts about the court’s independence”.⁹

These concerns are apparently shared by the European Commission, which proposed a draft Regulation laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) in 2021.¹⁰ The proposal classifies “AI systems intended to assist a judicial authority in researching and interpreting facts and the law and in applying the law to a concrete set of facts” as “high-risk AI systems” and (in Article 14) aims at subjecting them to human supervision “during the period in which the AI system is in use” in order to protect fundamental rights. This approach is based on the noticeable belief in the quality of the human judiciary and – more broadly – on the anthropocentric orientation of contemporary constitutionalism¹¹ and, gradually, international law as well¹².

The idea behind these concerns seems to be that if an algorithm delivers a judgment (or participates therein in the decision making by providing decisive data), it is not “independent” from the humans who developed the algorithm, nor from the data contained and managed by an automated system. The judgment, instead of being the sole consequence of individual judicial appraisal, is predetermined by the human programmer and the dataset provided by him or her.

However, judicial independence means that the decision-making process is not interfered with and not that it is unrelated to data on which it is based. A human judge is (even more so than algorithms, perhaps) “programmed” by his/her knowledge of law, available information (in practice, this nowadays amounts to electronic databases of the case-law and scholarly works), cultural and social background, individual prejudices etc. Independence entails personal and institutional qualities required for impartial decision-making, and it is thus a prerequisite for impartiality.¹³ It characterises both a state of mind which denotes a judge’s imperviousness to external pressure as a matter of moral integrity, and a set of institutional and operational arrangements which must provide safeguards against undue influence and/or unfettered discretion of the other State powers.¹⁴. This makes independence a characteristic of a tribunal’s relations vis-à-vis the other branches of government¹⁵ and the parties to the proceedings¹⁶. What matters is that there must be no external, undue influence from the outside on how justice is administered in a particular case. Even the political nature (or “flavour”) of the judicial appointment process as such does not necessarily mean that the requirement of independence is always impaired¹⁷ – provided that sufficient safeguards exist protecting the sterile judicial decision-making mechanism. Taking into account the concerns regarding the independence of AI judges from the possible undue influence of other branches of government¹⁸, one cannot but note that from a normative perspective they are no different from those concerning any other undue influence on judicial decision-making. The difference lies in the technological implementation of such influence, but not in the influence itself.

Obviously, specific technical safeguards are required to protect the algorithm from external influence while deciding cases. Yet, theoretically, achieving the required standard of independence is not excluded when algorithms are used for adjudicating purposes. The same conclusion applies to the requirement concerning the way in which a court is established. As the CJEU rightly held in A.K. v. Krajowa Rada Sądownictwa:

“although the principle of the separation of powers between the executive and the judiciary has assumed growing importance in its case-law, neither Article 6 nor any other provision of the ECHR requires States to adopt a particular constitutional model governing in one way or another the relationship and interaction between the various branches of the State, nor requires those States to comply with any theoretical constitutional concepts regarding the permissible limits of such interaction. The question is always whether, in a given case, the requirements of the ECHR have been met”.¹⁹

It follows that while there may be different methods of appointing judges, they can entail different degrees of political involvement. Ultimately, however, it comes down to an overall assessment of whether the judicial branch is sufficiently protected from the undue political influence by government, as is also the case for the independence requirement.

Therefore, while relying on AI as a judge clearly raises serious concerns pertaining to the requirement of independence, it appears that these concerns are no more pertinent than in the case of human judges. In a way the opposite might be true – AI promises more transparency in the sense that anyone (obviously provided they have enough technical expertise) can actually check the dataset on which AI judgments are based, whereas no one can consult the mind of a human judge.

2. Fairness

If sufficient technological safeguards exist – which by nature is more a technical than a legal question – that protect the independence of AI judiciary (understood as immunity from undue influence from other branches of government), the question nevertheless remains whether an AI judge is able to ensure fairness as a precondition of the right to a fair trial.

The right to a fair trial does not necessarily need to encompass a right to substantive fairness (a “proper” judgment). In some jurisdictions, it extends to a substantive fairness guarantee²⁰, in others it does not²¹. In the European constitutional space, the notion of a “fair trial” is limited to the question of procedural fairness, i.e. whether the rule of law is respected and therefore the adjudication was free of arbitrariness.²² From the point of view of the ECHR, it thus constitutes a “purely procedural guarantee”.²³

Nonetheless, fairness represents one of the constructional requirements of the right to a fair trial, being “one of the fundamental principles of any democratic society”.²⁴ It presupposes that claims and observations of either party are duly considered by a trial court.²⁵ Fairness places “the tribunal under a duty to conduct a proper examination of the submissions, arguments and evidence adduced by the parties, without prejudice to its assessment of whether they are relevant to its decision”.²⁶ The requirement of fairness encompasses a number of detailed conditions such as equality of arms,²⁷ adversarial trial,²⁸ reasoning of rulings,²⁹ freedom from self-incrimination,³⁰ lawfulness of administration of evidence,³¹ or the principle of immediacy³². But fairness is much more than just observing procedural rules – it is about properly considering the material of a case at hand and about avoiding any arbitrariness in its judicial appraisal. Fairness, in that way, also seems to be substantively interlinked with the right to effectiveness of judicial protection, since the latter is unlikely to be respected if the former is not.

So, can we have a fair trial, i.e. proper examination of the case pending before a court, if this court is an AI one? This leads us to the question of how we should understand proper examination (or due consideration), and whether or not this could be upheld without a human component.

II. On the Notion of “Proper Examination (Due Consideration)” and Why We (Sometimes) Need Human Judges

European Court of Human Rights (ECtHR) case-law has been consistent in its stance that “the right to a fair trial cannot be seen as effective unless the requests and observations of the parties are truly ‘heard’, that is to say, properly examined by the tribunal”.³³ This means that the court is “under a duty to conduct a proper examination of the submissions, arguments and evidence adduced by the parties, without prejudice to its assessment of whether they are relevant to its decision”.³⁴ This duty is not limited to providing sufficient reasoning of the tribunal’s decisions,³⁵ but extends to due consideration of the body of evidence collected before it.

The “due consideration” requirement entails proper in-depth examination of the body of evidence produced before a tribunal. For instance, in criminal cases “a conviction ignoring key evidence constitutes a miscarriage of criminal justice”.³⁶ One must note that properly assessing the gravity of evidence and its significance for the proper examination of a case requires very complex analyses. These do not necessarily need to follow the pre-existing patterns according to which an AI judge would come to a decision. Every single case constitutes a more or less unique bundle of factual findings and “due consideration” requires assessing their individual relevance for the ruling as well as judicial appreciation of their interconnections. This results in a duty to address sometimes very complex factual backgrounds of cases where different elements are mutually interlinked.

In Farzaliyev v. Azerbaijan, for example, the ECtHR reproached national courts for failing to provide sufficient reasoning behind their fact-finding. Yet it seems quite clear that what the courts were really failing to do was to respond to the abusive institution of criminal proceedings for the sole purpose of reviving a claim period when all the statutes of limitations had long expired. While the ECtHR criticism primarily/ostensibly targeted the lack of reasoning, the real issue was the abuse of the law in its application in fraudem legis.³⁷  Similarly, in Cupiał v. Poland, the national courts – in the Court’s view – not only failed to respond to the applicant’s allegations, but moreover “no efforts were made to analyse this issue”.³⁸ The assertions and pleas forwarded by a party to a court trial must be “carefully considered” (“soigneusement examinés” in French) and analysed “thoroughly and seriously” (“approfondi et sérieux” in French), which under the fair trial guarantee is considered a separate requirement from providing a proper reasoning.³⁹

However, “due consideration” also requires properly subjecting a tribunal’s decision to the scrutiny of social reality and the present-day conditions. As aptly pointed out by Jordi Nieva-Fenoll, laws

“must adapt to the times, or end up sending Galileo Galilei to the fire, and for this the work of the judge is essential. In their mission to analyse the specific situation in which the rule is to be applied, they must observe the nuances of that situation and determine the best application for it. While this makes the application of the law less predictable than what a mathematician would accept or what an AI programmer would imagine, it is precisely what will guarantee that the law does not enter as a foreign body into people’s lives, but reasonably regulates their coexistence”.⁴⁰

In light of this, “duly considering” a case appears to be the antithesis of “dully considering” it. It goes far beyond simply applying the law, thus presupposing application of justice. The former cannot be achieved without also taking into consideration (“duly considering” vs. “dully considering”) different social contexts and changing attitudes of a given society. What would have been perceived as just when a law was adopted might be considered appalling when it is applied – or, as the Grand Chamber held in Fedotova and others v. Russia, “what may have been regarded as permissible and normal at the time when the Convention was drafted may subsequently prove to be incompatible with it”.⁴¹

Artificial intelligence, in its current state of development, is capable of analysing (probably much more thoroughly and precisely than humans) very complex data and subsuming a legal norm deduced from the legal system to an established factual situation. But that’s really all it can do. So what is missing? Three observations can be made in this regard.

Firstly, as matters stand, AI seems unable to develop an interpretation of the law that adequately takes into account the ever-changing social landscapes surrounding the processes of “doing justice.” Ignoring this obstacle and pressing ahead with AI-driven judgment risks an algorithm applying the law correctly from a strictly formal point of view, yet completely missing the mark when it comes to the societal sense of fairness and justice (the risk of an overly positivistic AI judge). This sense of fairness and justice relies heavily on perceptions of various societal phenomena. Let us take the example of “socially acceptable criticism and exaggeration” as a concept that comes into play when competing interests are weighed against each other, – in this case the right to respect for personal dignity vs. freedom of expression. On numerous occasions, the ECtHR has been confronted with complaints about the judicial application of national laws limiting freedom of expression in cases of “exaggerated” criticism. It has become consistent case-law of the Court that national authorities are under a duty to assess whether the “generally accepted limits of exaggeration” were exceeded.⁴² So far, it seems unlikely that an an AI judge would be able to appreciate the subtleties of a particular expression in a way that would ensure these accepted limits are adhered to. . This would require a very nuanced knowledge of what is accepted by the general public, while taking into account various current social developments. It should be noted that when the ECtHR scrutinises the proportionality of state interventions (even when the classic elements of a proportionality review – i.e. adequacy, reasonability, and necessity – are not explicitly referred to by the Court⁴³), it assesses the measures applied against the legitimate aims pursued, and takes into account what is perceived to be fair and just in a process of judicial appreciation.

Secondly, defining a norm deduced from the provisions of a given legal system, while considering not only its provisions but also scholarly writings and the jurisprudence, seems a task perfectly tailored to artificial intelligence. After all, it involves the analysis of complex databases, which are closed sets of data with a predetermined scope. Yet appraising the body of evidence is something inherently different because evidence may vary according to factual circumstances. In turn, these factual circumstances cannot reasonably be predetermined in most cases (except maybe for relatively simple cases of unpaid invoices etc.). Again, falling back on AI when deciding cases that present more complicated evidence may compromise the duty – rooted in the right to a fair trial – to carefully, thoroughly, and seriously analyse particular pieces of evidence and their interconnections against the general background of a case. The assessment of evidence, as a whole, must be “fair and proper”,⁴⁴ and this remains a task that many humans are unable to accomplish, let alone artificial intelligence.

Thirdly, as things stand, AI justice appears unsuitable as a classic branch of government. The primary role of the judiciary is, of course, to decide individual cases and apply justice. However, administration of justice goes way beyond specific cases. It is also a part of the checks and balances formula characterising modern democratic societies. As a consequence, the judiciary’s interpretations sometimes need to be more extensive when the political branches of the government plainly fail to respond to society’s needs, or when these political branches turn dysfunctional. Judging when the time has come for the judiciary to quickly respond to state failure or dysfunctional government (or, conversely, when it should demonstrate more self-restraint) entails careful consideration of very complex socio-political processes and, almost physically “feeling” them. Given AI’s reliance on predetermined and available definitions and datasets, this seems still unachievable for an AI judiciary.

III. Conclusions

The use of artificial intelligence as a tool for judicial decision-making is becoming ever more widespread. It goes hand in hand with the idea of replacing human judges with AI. While the use of AI as a judge would largely benefit the efficiency and predictability of the administration of justice, its use would be neutral when it comes to fair-trial criteria such as the right to professional legal representation or legal aid. When it comes to independence and impartiality, AI judges appear no more likely to fall of short of these requirements than human judges, provided that sufficient technical safeguards are in place.

Nonetheless, the use of AI (in its current state of development) as judges does not seem to be reconcilable with the right to “fairness” of a court trial, which includes the duty of the trial court to duly consider the case.

For one, this is because more complicated litigations implying complex analysis of the body of evidence and the interconnections between particular pieces of evidence against a general factual background seem to exceed the technical scope of AI as envisaged by developers. What is more, AI seems equally unable to duly consider concepts which require human intuition, without which justice cannot be administered fairly (i.e. corresponding to the general sense of what is fair and just). Lastly, AI does not seem to be able to genuinely play the role of the judicial branch of State powers maintaining checks and balances on the political branches. It does not seem to be capable of navigating the interpretation of the law in such a way as to properly respond to possible failures and dysfunctions of the legislative and the executive branches (i.e. sometimes applying more dynamic interpretation, other times being more self-restrained).

It follows that in cases that are more complicated than simply ordering payment on the basis of outstanding and undisputed invoices, the use of AI in the judiciary system may compromise the requirement of fairness, which is one of the commonly accepted definitional elements of the right to a fair trial. Instead of “duly considering” such cases, AI – predetermined and limited by the underlying dataset – is likely to “dully consider” them.