Netherlands lower court use of GenAI in a ruling sparks controversy

A Dutch lower court judge's reliance on information from a generative AI tool in rendering a recent verdict has sparked heated debate on social media and raised critical questions about the reliability and fairness of AI-assisted judicial decisions. Both legal professionals and AI experts have expressed indignation at the idea of a judge using AI. Generative AI, however, cannot be ignored in either society or the legal sector. Still, the question remains: is a verdict based on GenAI unacceptable or could it become an innovative approach to court decisions if used correctly?

GenAI and fair hearing

In the underlying case, the plaintiff sought damages for the loss of output from his solar panels as a result of his neighbor's roof extension. According to the Dutch Code of Civil Procedure, the court, when arriving at an award of damage, should itself estimate the damage or, as a secondary option, refer the estimation to a separate procedure. 

According to established case-law from the Supreme Court of the Netherlands, the court must adhere to the principle of fair hearing when estimating damage. According to this principle, any decision adverse to either party should not be based on evidence that the party has not had an opportunity to comment on. In this case, the court used GenAI to estimate on the lifetime of the solar panels and the average price per kilowatt-hour. Critics of this method state that it is not clear what models the GenAI tool relied on when providing this assessment, which hindered the respondent’s ability to conduct a defense, thereby violating the principle of fair hearing. 

The Judge on the internet

The case brings to mind previous rulings on judicial decisions based on information the judge independently searched for online. Among them is a Supreme Court ruling dated from 2011 that involved a financial guardian who had purchased an administration system at the expense of the protected person. The Court of Appeal had ruled that the guardian should refund the purchase price because the system was used more to support the guardian's administration rather than representing the interests of the protected person. The court independently obtained that information via the Internet. In this regard, the Supreme Court ruled that the court violated the principle of fair hearing as the administrator was not given the opportunity to respond to the facts gathered by the court.

Facts of common knowledge

Thus, a person facing an award of damages should have the opportunity to comment on the facts on which that award will be based. But that set of facts is extensive and not all facts should be brought to the surface through in-depth investigation. Some facts are already apparent to the judge from his own knowledge or can be easily ascertained. Yet a judge is not required to present all independently gathered information to the parties before considering it in the decision. Under Dutch law, this is not the case when it comes to “facts of common knowledge”. 

According to legislative history, facts of common knowledge are defined as “notorious facts that every normally developed person knows or can know from sources accessible to everyone”.  This definition, however, dates back to times when the amount of available sources and the information that could be derived from these sources was in stark contrast to the situation today. In the present day, when any question can be answered in a few seconds by means of generative AI, this definition does not seem appropriate. 

In a 2008 ruling, the Supreme Court held that the circumstance that information came from a generally knowable source, in this case a newspaper, did not mean that the information also constituted facts of common knowledge. This raises the question: what facts found on the Internet can be assumed to be common knowledge? It is conceivable that, for example, the capital of the Netherlands is common knowledge, but the question what city in the Netherlands benefits the Dutch economy most is less likely to be. In this case, what criterion distinguishes the two pieces of knowledge and where is the boundary between what is common knowledge and what isn’t?

In Dutch criminal law, this question has been addressed by the courts, since an identical principle of commonly known facts applies in this type of law. Hence, we can draw connections between how the issue has been addressed in criminal law and its application in civil evidence law. The Supreme Court has held that facts of common knowledge generally involve information that does not presume specialist knowledge and the correctness of which is not reasonably open to dispute. The criterion of “not reasonably being open to dispute” does justice to the principle of fair hearing. One can assume that if there is a plausible way to dispute the fact being alleged against the respondent, he would likely want to do so and must be granted the opportunity to do so. Thus, regardless of whether a judge looks something up on the Internet, consults a GenAI tool, or relies on their own knowledge, when a factual detail forms the basis of a decision, the justice must consider whether the fact is reasonably debatable in order to consider whether the parties should be allowed to comment on it.

The future 

Clearly, the development of GenAI cannot be reversed and increasingly will be used in every sector, even in the judicial system. This, however, does not necessarily mean that the outcomes will be unacceptable or in violation of rights. 

It is important that GenAI not be viewed as a reliable source, but as a method for finding reliable sources. Although GenAI is a whole new source for gathering information, issues of fair hearing regarding factual judgments formed on the basis of GenAI can be approached in the same way as one tends to do with traditional methods of information gathering. GenAI tools often offer the ability to display references to an answer. If a judge uses these references to check the sources of the answer for reliability, considers whether the data is reasonably open to dispute and, if so, submits the data to the litigating parties so that they can assess and comment on them, the principle of fair hearing is observed. Adapting to these developments takes time, but when used properly, generative AI can be a useful way to conduct research and reach conclusions, even judicial ones.

For more information on this debate and the use of GenAI in the legal sector, contact these CMS experts: