1. 1. Introduction

          The rapid development of artificial intelligence technologies over the past decade has raised the question for many professions: will a person be replaced by a machine? [1] The legal profession, traditionally considered intellectually intensive and requiring high qualifications, is also under threat [2]. Today, neural networks can analyze multi-volume case files, draw up contracts, check counterparties and even predict a court decision [13]. In this regard, a natural question arises: will artificial intelligence replace a legal assistant or a lawyer in the foreseeable future, or will it remain just a technical tool? [4] The relevance of this study is due to the need to understand the real boundaries of AI in law in order to predict changes in the labor market and adjust the system of legal education [14].

2. Material and methods

          Artificial intelligence in jurisprudence is a set of technological solutions that allow simulating certain aspects of human intellectual activity when working with legal information [1]. This includes natural language processing (NLP), machine learning on arrays of judicial acts, predictive analytics and generative neural networks (such as ChatGPT, GigaChat) [2]. Unlike a simple search engine, AI is capable of not only finding documents by keywords, but also semantically analyzing them, highlighting legal structures and even offering ready-made legal positions [13].

          The first experiments with the introduction of AI into the legal profession date back to the 1990s, but real progress began in the 2010s with the advent of accessible machine learning algorithms [13]. In the United States, systems such as ROSS Intelligence began to be used, which answered lawyers’ questions in natural language, analyzing the entire body of bankruptcy law [7]. In Europe, the Promethea project has been implemented, which, with high accuracy, predicts decisions of the European Court of Human Rights based on an analysis of previous cases [7].

          Currently, AI is most actively used in the following legal areas [4]:

1. Automation of document review (due diligence). Neural networks can process thousands of pages of contracts and corporate documents in a few minutes, identifying risky clauses and inconsistencies [11]. This task would take a legal assistant several weeks [14].
2. Judicial practice analytics. AI systems (for example, the domestic «Pravosudiye» or CaseLab) analyze court decisions on similar cases and calculate the probability of a particular outcome of the case [11]. For a lawyer, this provides an objective basis for building a strategy [5].
3. Generation of standard documents. Neural networks easily draw up claims, complaints and contracts based on a brief description of the situation [2]. However, such documents require mandatory verification by a specialist, since AI may miss legally significant nuances or rely on outdated versions of legislation [13].
4. Legal chatbots for citizens. AI is already replacing junior legal assistants in primary consultations: the bot answers typical questions (how to draw up an application for a fine, how to collect alimony) and generates templates [14]. This reduces the burden on lawyers in routine matters [14].

However, along with the obvious advantages, serious limitations of AI in law are also revealed [15]. For example, a neural network does not understand the meaning of the text it processes — it operates with statistical patterns [2]. As a result, the AI may ignore an unusual but key circumstance of the case, because it did not occur in the training sample [13]. Also, the neural network is not capable of strategic thinking, building a line of defense taking into account the personal characteristics of the judge or the emotional mood of the client [4]. Finally, the problem of responsibility remains acute: if AI gives the wrong advice and the client suffers damage, there is no subject to sue [15].

Let us consider several illustrative examples from judicial practice (albeit still few) demonstrating both the successes and failures of AI in law.

Case №1: Lawyer fined for using ChatGPT. In 2023, a US federal court fined lawyers who submitted a motion drawn up by ChatGPT to the court [7]. The neural network generated references to completely non-existent judicial precedents, which the lawyers did not check [7]. The judge found that the lawyers had committed bad faith and ordered them to pay a fine [7]. This case showed that AI cannot be a substitute for a lawyer’s professional judgment — it is only a tool, the results of which must be verified [15].

Case №2: Compas recidivism assessment system. In the United States, the Compas algorithm has been used since the 2000s to assess the likelihood of a defendant re-offending [9]. The system predicted that the black defendant would re-offend with a high probability, and the white defendant with a low probability [9]. In reality, the opposite happened [9]. Studies have found that the algorithm contained hidden racial bias because it was trained on historically biased data from previous convictions [9]. This example demonstrates that AI is not inherently objective; it reproduces and amplifies the biases embedded in the training data [15].

Case №3: Successful implementation in the Russian arbitration process. Since 2019, the «Sudebnik» AI system has been operating in the Russian Federation, helping arbitrators to quickly find similar cases and generate the reasoning part of the decision [8][11]. Practice shows that the system reduces the time for preparing documents by 30-40%, but the final decision — the resolution of the dispute on the merits — is always made by the judge [5]. The neural network does not replace the judge’s inner conviction [8].

Thus, from the analysis of the current state of affairs, it follows: AI confidently wins in the speed of processing homogeneous legal information, searching for patterns in judicial acts and preparing standard documents [13]. In these areas, a neural network can already replace a legal assistant or a junior lawyer performing routine work [14]. However, in matters of legal strategy, interpretation of ambiguous norms, representation in court (where personal interaction, rhetoric and emotional influence are important) and ethical decision-making, AI is fundamentally inferior to a human [4][15].

3. Results and discussions

  During the study:

• the concept of artificial intelligence in jurisprudence, its current capabilities and technical limitations have been analysed [1][2];
• the key areas of legal activity that are most amenable to automation have been identified (document review, primary consultation, judicial analytics) [4][11];
• it has been established that neural networks successfully replace a legal assistant in performing routine, repetitive tasks [14];
• it is proved that a lawyer (advocate) cannot be completely replaced by AI due to the need for strategic thinking, moral judgment, empathy and personal responsibility for the result [4][15];
• the risks of introducing AI into law are formulated: «black box» problem (lack of explainability of decisions), algorithmic bias, lack of legal personality of AI and, as a consequence, impossibility of holding it accountable [9][15].

A separate discussion deserves the question of the transformation of the legal profession [14]. Most likely, in the next 5-10 years, we will not see the disappearance of lawyers, but a change in the structure of the profession [4]. The demand for junior staff performing «manual» document processing will fall [14]. At the same time, the demand for specialists who know how to work with legal AI systems, formulate queries for neural networks and verify the results of their work will increase [13]. The so-called «prompt engineers» in law may appear — specialists in setting tasks for AI to solve legal problems [14].

4. Conclusion

The conducted research indicates that artificial intelligence in jurisprudence has passed from a theoretical concept to the stage of active practical implementation [1][11]. Neural networks already today demonstrate high efficiency in the analysis of judicial practice, automation of document flow and primary consulting [13]. In these areas, AI is able to partially replace a legal assistant, especially when performing routine, repetitive operations [14]. However, the complete replacement of a lawyer (advocate) by a neural network seems impossible in the foreseeable future [4]. Lawyering is not limited to knowledge of normative texts and formal logic — it includes strategic thinking, understanding of the human context, ethical choice and personal responsibility for the client’s fate [15]. AI will become a powerful assistant, but not an independent subject of legal activity [2]. The future of jurisprudence is a symbiosis of a human and a neural network, where each solves its own tasks: the machine — for speed and accuracy in routine, the human — for meaning, strategy and justice [4][13].