Legal reasoning is an old concept, capable of being found way back in the Roman times. Decisions were justified by reference to exemplar factual situations and reasoning of other jurists, often seemingly guided by own views. The current age decision-making contains slight differences. To understand why a judge argues a case in a certain way, it is necessary to consider the reasons used to justify his reasoning. As Hunh suggests, a judicial decision is capable of being reduced into a syllogistic reasoning by use of deductive logic, thus, first of all, this work will explore how such syllogism comes to existence and what factors are necessary to be considered in its justification. Reference will also be made to the three main judicial decision-making models to provide clarification on how judges reach a solution in relation to formalism and realism. The above-mentioned information will then be incorporated into the context of computers and artificial intelligence to uncover that certain aspects of legal reasoning are not capable of being reproduced by a machine sufficiently to replace human lawyers.
Human Legal Reasoning
When a human lawyer engages in legal reasoning, the process he uses can often be simplified into that of a syllogism. A rule is applied to the facts to create an outcome. Following Sir Edward Coke’s view that “reason is the life of law”, syllogisms may be assumed to be of a purely logical nature. Such argument may be partly true in simple cases, however even there it is often necessary to be able classify the facts based on the rule , due to the uncertainty of language questioning whether a fact is included under the rule of law or not. Classification is labelling of a fact as an instance of a rule. On the other hand, the more difficult cases, where there is a choice of conflicting or ambiguous rules, are resolved by the process of “evaluation and balancing”.
As legal reasoning is, primarily, rule-based decision-making, when faced with multiple relevant rules, it is necessary to be able to distinguish which of them is the most applicable to the facts. Rules can derive from principles that justify them, which are to be weighed against each other when applying law. Dworkin suggests that the difficulty arises when despite the clarity of the rule or the principle, the judges question its validity and overall applicability in terms of fairness. The two-stage proportionality analysis is engaged, as majority, if not all rules deal with restrictions or positive obligations to engage. The limitations of the rule are weighed against costs and benefits of it . Factors such as whether less restrictive means are possible are also considered. The strength of the rules range form being merely indicative to being conclusive and their guidelines can be found in the relevant provisions or case law (Practice Statement, etc.), although they are often still not clear and have to be interpreted.
Interpretation of rules provides further difficulties to lawyers. Not only the law is open textured and provides no clear guidelines, but there are also different methods of interpretation and there is no order of priority. The French philosopher Montesquieu proclaimed that “the national judges are no more than the mouth that pronounces the words of the law, mere passive beings, incapable of moderating either its force or rigour”. However, it became obvious that interpretation was necessary in some cases as to follow the words literally is to believe that rules are written perfectly. The Law Commission Report stating this has also attempted to encourage the use of purposive approach instead of the other three. The purposive approach was endorsed over twenty years later by the landmark case Pepper v Hart which allowed extraneous material constituted prior to the enactment of the rule to be used in cases of ambiguity.
Problems may arise when the freedom of interpretation is put into the context of realism. Realism is a type legal reasoning based on one’s ideology. It inclines a person to interpret information and to reason in a way that it affirms their previous beliefs about an issue. Due to it being a part of human nature, it is safe to say that every lawyer utilises this method, unless he has trained himself to consciously not to. However, as long as the person is capable of admitting failure in finding justification of their answer, then “judicial hunch” may not be so harmful but rather support Hutcheson Jr.’s argument that it is the “true basis of legal reasoning”. Nonetheless, to truly understand how humans perform legal reasoning, the realistic approach should be explored in greater depth, especially in relation to the attitudinal model.
Attitudinal model compares the ideology of judges with how they tend to vote on specific topics to identify intellectual influences at play. It is used to attempt to predict future decisions made by those judges. In the US Supreme Court, a trend can be seen of justices adhering to their beliefs quite often when engaging in decision-making, yet the predictions still are not hundred percent correct as certain cases may involve topics on which the judge only has an unstable opinion that can be swayed and thus vote contrary to his ideology. The UK, on the other hand, faces more difficulties with the Attitudinal model as its Supreme Court not only hears fewer cases, but also not all judges are present, thus there is a possibility of different combinations of judges affecting the outcome of a case.
On the other hand, we have the strategic model which, while not the opposite of the attitudinal model, it does rather focus on the process of legal reasoning, rather than on the input. The goal is not to promote own views, but to have a solid collective decision that will not be overturned. Such outcome is achieved by strategic bargaining between the judges during the judicial conference part of the decision-making. In the UK, the court held a decision that compensation was owed for damage dealt during the war which was consequently overturned by an Act of Parliament.
Computers and Aspects of Legal Reasoning
In order to determine whether computers can replace lawyers, the above-mentioned legal reasoning processes need to be examined in the light of artificial intelligence. Issues discussed here will be classification of facts according to rules, balancing act, interpretation and lastly the concept of prediction of an outcome of a case.
Classification, as aforesaid, is not a logical process. While in one situation, it may be decided that a fact is an instance of a rule, in another it can be held that it is not even though it seems that it a would be perfectly reasonable decision. AI are already capable of being taught to profile information to be able to classify. One such instance is the personalised pricing where depending on information that was collected about a customer, the prices will be adjusted accordingly to suit their trend. However, there are also AIs who were unable to do this task correctly. The Odyssey, court software which records the judgements of judges, for example in terms of warrants, has in this case erroneously classified individuals to be arrested, while some even had to register as sex offenders. Google AI has also been seen to make a mistake when it labelled photos of two darker-skinned people as “gorillas”.
Balancing of sources is another issue to be briefly discussed. Ernst makes a convincing argument in this regard. He states that the process used to balance information, algorithms are not to be underestimated. The data, while vast and seemingly objective due to its variety, is chosen based on values and preferences of the creator, who has to make a decision which criteria bears what weight.
Interpretation can cause difficulties even for humans, especially what rule is to be used to read the source. Number of AIs currently manufactured, such as Alexa or Siri, utilise natural language processing to aid them. Bouazis defines this as “a computer program’s ability to understand spoken and written language”. However, merely understanding the formal and slang language as it is in a dictionary is not enough. Following Baude and Sachs’s argument, legal interpretation is not just concerned with the meaning of the words but what law they signify and what is its position within the wider body of law? Dervanović further articulates this argument by indicating limitations of language in reference to “unused legal provisions, of which validity does not expire by non-usage, while elements of language may cease to exist without usage”.
Lastly, foreseeing a decision of a case by use of a computer may not be far. Blue J Legal created a software that is capable of predicting the outcomes of Employment Law cases. The ‘Classifier’ is said to “uncover hidden patterns in case law by discerning relationships between individual factors and court decision outcomes”. Thus, it is capable of predicting the outcome of a case based on the responses given to the twenty or so questions to be answered with up to 90% accuracy. By using the software, other decisions can also be explored by changing any one answer to the questions given.
To conclude, while AIs have seen vast improvements and are able to ultimately foretell decisions of a court based on facts and precedents given, the process that reaches that stage is uncertain. Computers are still unable to grasp the concept of rules and their place in the law as a whole. Furthermore, classification test has also been unsuccessful in numerous instances and balancing is subject to bias of the creator of the computer. Computers cannot replace lawyers however, by working alongside them but still leaving humans to be the ones to call the shots, productivity of humanity can be greatly increased. This would allow for more focus into the more pressing matters of legal reasoning while still maintaining high standards and accountability.