Predictive Artificial Intelligence and Legislative Gaps in Arabic Legal Texts
This study evaluates the ability of large language models to identify and assess legislative gaps in Arabic legal texts, finding that while they effectively detect omissions and generate arguments, determining the constitutional significance of such gaps requires human legal judgment to distinguish between actual defects and intentional legislative choices.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the world of law, a "gap" is not a hole in the ground, but a silence in the rules. It is a situation where a new problem arises, or an old one changes, and the written laws do not say exactly what to do. For centuries, finding these gaps has been the job of human lawyers and judges, who must read thousands of pages of text, understand the history behind them, and decide if a missing rule is a simple oversight or a dangerous flaw that violates the constitution. Today, a new tool has entered this room: artificial intelligence. Specifically, large language models are computer programs trained on vast amounts of text that can read, summarize, and argue about legal documents. But a critical question remains: can these machines tell the difference between a law that is simply silent and a law that is broken? If a computer says a law is missing a rule, is it actually missing, or is the silence a deliberate choice by the lawmakers?
A researcher named Sahar Al-Atawi from the University of Tabuk set out to answer this by putting three of the most advanced artificial intelligence systems to work on Egyptian law. She did not ask the machines to write new laws or to act as judges. Instead, she asked them to act as investigators. She fed them four specific articles from Egyptian statutes—laws governing university professors, postal workers, and social funds—and gave them a strict set of instructions. The computers had to read the text, find anything that seemed to be missing, such as a warning before firing someone or a clear definition of a term, and then decide if that missing piece was a minor detail or a major constitutional failure. The study compared the work of three different models: ChatGPT, Gemini, and DeepSeek, treating their outputs not as final answers, but as raw data to be checked against human legal reasoning.
The results revealed a fascinating split in how these machines think. All three models were excellent at spotting the silence. They all correctly identified that some laws did not explicitly state that a worker must receive a written warning before being deemed to have resigned, or that a professor must be given a hearing before losing their job. However, the machines disagreed sharply on what that silence meant. One model, DeepSeek, tended to treat every missing procedural step as a serious legal defect, often concluding that the law was broken and needed immediate fixing. The other two models, ChatGPT and Gemini, were more cautious. They recognized that just because a rule was not written down did not mean the law was unconstitutional. They understood that sometimes silence is a deliberate choice, or that other parts of the legal system already cover the issue.
The study found that the most accurate approach was the cautious one. In the first case, involving university professors who stopped showing up for work, the machines found that the law did not explicitly require a warning before the professor was considered to have quit. DeepSeek called this a major gap that could violate rights. The other models, and the human analysis that followed, determined that the law was actually quite robust. It gave the professor six months to return and explain their absence, and it required a council to approve any excuse. The missing warning was a detail that could be improved, but it was not a fatal flaw that made the law invalid. The same pattern appeared in a second case about postal workers. The law allowed a worker to be fired for missing thirty days of work, but it did not explicitly say the worker had to be told why their excuse was rejected. Again, one machine saw a crisis, while the others saw a manageable gap that did not break the constitution.
The most revealing moment came in the fourth case, where the machines were tested on a law about a national social fund. The text of the law described how money was collected and how the fund was run, but it said nothing about firing employees. When the researchers asked the machines if the law was missing a rule about firing people, one model, DeepSeek, struggled. It assigned significant weight to the absence of individual procedural safeguards, suggesting the law was flawed because it didn't protect workers from being fired, even though the law was never about firing workers in the first place. It was like asking a chef to explain why a cake recipe is missing instructions for building a car; the recipe is not broken, it is just about a different subject. Only one model, ChatGPT, correctly realized that the law did not cover employment termination at all, and therefore could not be criticized for missing those rules. This error highlighted a crucial limitation: without deep human context, artificial intelligence can mistake a silence for a mistake, or look for problems in the wrong place.
The final conclusion of the research is that artificial intelligence is a powerful assistant, but not a judge. These computer models can scan thousands of pages in seconds and point out every place where a word is missing or a rule is vague. They can generate arguments for why a law might be unfair. But they cannot decide if that unfairness is real or if it matters in the eyes of the constitution. The study showed that an identified omission is not automatically a legal gap. Sometimes the silence is intentional, sometimes the rule exists in a different part of the law, and sometimes the issue is simply outside the scope of the text being read. The machines are excellent at finding the missing pieces, but humans must still decide if those pieces are actually needed.
This research suggests a new way for lawyers and lawmakers to work. Instead of expecting artificial intelligence to solve legal problems on its own, they should use it as a screening tool. The computer can highlight the areas that need attention, flagging potential issues that a human might miss in a hurry. Then, a human expert can step in to verify the finding, check the broader context, and decide if the silence is a problem that needs fixing or a choice that should stand. The study makes it clear that while technology can speed up the search for answers, the final judgment on what the law means, and whether it is just, remains a deeply human task. The machines can point to the gap, but only a person can decide if it is a chasm or just a shadow.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.