Forensic support for the use of artificial intelligence technology in the detection and investigation of crimes

Khamidullin Ruslan Sibagatullovich PhD in Law Senior Lecturer of the Criminalistics Department of the V.F. Yakovlev Ural State Law University 620072, Russia, Sverdlovsk region, Yekaterinburg, Syromolotov str., 7, sq. 11 sledgsugu@mail.ru Other publications by this author

In the history of technology, there are many examples of how designers and inventors have tried to develop machines that will be able to facilitate individual mental operations, usually of a computational nature. Some of these inventions include W. Schickard's counting clock (1623), B. Pascal's summing machine (1642), G. F. Leibniz's arithmometer (1673), and the logic demonstrator Ch. Stanhope (1777) and the difference machine (the machine of differences) by C. Babbage (1822). Although these inventions were not intelligent computer systems, they had a significant impact on the development of computer technology and led to the emergence of such an important technology as artificial intelligence, which D.V. considered in more detail in his doctoral dissertation research. Bakhteev ^[1].

The use of artificial intelligence in the detection and investigation of crimes represents significant progress in the field of criminology. This technology makes it possible to change the traditional working methods of law enforcement agencies, enriching their arsenal of tools. The forensic aspects of the application of artificial intelligence include the analysis of large amounts of data, pattern recognition, automated processing of evidence and forecasting the commission of new crimes. E-digital learning allows you to create models that can identify patterns in criminal events, determine the characteristics of potential criminals and identify key factors affecting the dynamics of the situation in this context.

However, the ethical and legal aspects of the use of artificial intelligence in criminology should be taken into account, such as the transparency of algorithms, maintaining data confidentiality and preventing possible problems with decision-making. This paper will examine how the use of artificial intelligence affects the effectiveness and reliability of crime detection and investigation, as well as what challenges and prospects this process is subject to in the context of forensic activities.

Artificial intelligence helps to improve the quality of law enforcement agencies by automating routine operations and data analysis. Machine learning systems can quickly process huge amounts of information, revealing hidden connections between various events and persons, which significantly speeds up the investigation process ^[2].

According to the National Strategy for the Development of Artificial Intelligence for the period up to 2030 (Decree of the President of the Russian Federation dated 10.10.2019 No. 490 On the development of artificial Intelligence in the Russian Federation (together with the "National Strategy for the Development of Artificial Intelligence for the period up to 2030"), this means a set of technological solutions that allows simulating human cognitive functions (including self-learning and search for solutions without a predetermined algorithm) and obtain results comparable to at least the results of human intellectual activity when performing specific tasks. The complex of technological solutions includes information and communication infrastructure, software (including those that use machine learning methods), processes and services for data processing and solution search.

Important aspects that should be taken into account when using artificial intelligence systems are, first of all, ensuring the protection of constitutional human and civil rights and freedoms. This also applies to everyone's right to privacy. It is also necessary to emphasize the inadmissibility of using artificial intelligence to intentionally harm citizens and legal entities.

Another important aspect is the prevention and minimization of risks that may arise as a result of the use of artificial intelligence technologies. In this regard, it is important to organize the work of artificial intelligence algorithms in such a way that they are open and understandable. This will allow you to control the process of achieving results and prevent possible negative consequences ^[3].

In recent years, the works of A.A. Bessonov ^[4], V.F. Vasyukov ^[5], R.I. Dremlyugi ^[6], I.A. Zavyalov ^[7], A.M. Kustov ^[8], A.I. Shemetov ^[9] and D.V. Bakhteev ^[10] have investigated and reflected various aspects of the application artificial intelligence in the detection and investigation of crimes. These studies make a significant contribution to the development of the field of artificial intelligence and its application in the field of law enforcement.

In the light of the increased crime rate, the use by criminals of new methods and means of committing crimes, as well as insufficiently effective detection rates, it becomes necessary to search for new ways to improve activities related to the identification, disclosure and investigation of crimes. An insufficiently effective system of indicators, a low level of training of law enforcement officers, further bureaucratization of the investigative and judicial system, subordination of the subjects of investigation, corruption risks of criminal procedural activities and directly criminal procedural legislation and a number of other factors negatively affect the quality of this activity. The intellectual capabilities of individual law enforcement officers, investigators, operatives, prosecutors, as well as the heads of these structures play a key role in changing the current situation. However, this approach turns out to be insufficient for a radical change in the current situation. In this regard, a qualitatively new system of intellectual support for operational investigative, forensic and criminal procedural activities, including the use of artificial intelligence systems, becomes necessary. Artificial intelligence is an essential attribute in modern and effective operational investigative and criminal procedural activities, and not a factor leading to the exclusion of the human element. The system of intellectual support of the criminal process should include the following interrelated and complementary elements:

- training of employees involved in the identification, disclosure and investigation of crimes;

- legislative support of the criminal process;

- scientific and research support of criminal procedural activities;

- information and technical support of criminal proceedings;

- the use of artificial intelligence systems in criminal proceedings, criminology and operational investigative activities.

At the same time, the introduction of artificial intelligence systems opens up prospects for the successful development of criminal proceedings ^[11].

In our opinion, the most important element in the detection and investigation of crimes using artificial intelligence is the forensic support of such activities.

In the course of the study, we will not dwell on the definition of the concept and essence of forensic software in general. In the dissertation research conducted in 2018 on the topic: "Forensic support for the activities of an investigator ...", we considered in detail the issues of forensic support for the activities of an investigator in the disclosure and investigation of crimes. As a result, we came to the conclusion that the forensic support of the activities of the subjects of the disclosure and investigation of crimes is a complex of knowledge, skills and abilities, including the use of technical means, techniques and tactics in the disclosure of crime investigation ^[12].

It should be noted that in this study, the criminalistic provision of disclosure and investigation of a crime through the use of artificial intelligence should be understood as a set of knowledge and skills of subjects of disclosure and investigation of crimes on the effective use of artificial intelligence systems in solving official law enforcement tasks.

Currently, forensic support for the disclosure and investigation of crimes is becoming an increasingly urgent and difficult task for law enforcement agencies. The difficulty lies in the need to analyze large amounts of information, identify hidden connections and patterns, as well as predict possible scenarios for the development of the case. In this regard, the use of artificial intelligence (AI) can become an effective tool to improve the results of an investigation.

One example of the use of AI in criminology is the use of GPT (Generative Pre-trained Transformer). GPT is an artificial intelligence model that learns from large amounts of text data and is able to generate coherent answers to questions asked or materials provided.

Within the framework of the conducted research, GPT training was carried out on materials studied by students of the criminology course in higher educational institutions. With the help of the AI chat, tasks and information necessary to solve these tasks are provided. AI provided answers efficiently and accurately, offering hypotheses, compiling programs and algorithms of actions necessary for the detection and investigation of crimes at various stages.

To test the effectiveness of GPT in real conditions, we conducted an experiment using materials from a criminal case under investigation by the Main Investigative Department of the Ministry of Internal Affairs of Russia in the Sverdlovsk region. Using the GTP chat, case materials were provided, and based on this information, the AI put forward versions and developed an investigation program at the initial stage. After that, the proposed actions and activities of GPT are compared with the actions carried out by investigators and operatives in the framework of a real investigation.

The results of the experiment showed that the artificial intelligence presented by GPT is able to analyze in detail and objectively and plan the activities necessary for the detection and investigation of crimes. It can also help in the formation of examination questions and the definition of various types of examinations. This allows us to conclude that it is possible to effectively use artificial intelligence in law enforcement not only to solve crimes, but also to prevent their commission, determine the circumstances and the crime scene.

However, it should be noted that the use of artificial intelligence in forensic software has its limitations. Firstly, AI does not have human intuitive understanding and emotional experience, which may affect the accuracy of its conclusions. Secondly, AI can be error-prone, especially if the training was conducted on incomplete or incorrect data.

Nevertheless, if applied correctly, artificial intelligence can become a valuable tool for criminalistically ensuring the detection and investigation of crimes. It can significantly speed up the process of analyzing information and organize the planning of an investigation at its various stages. As a result, the use of artificial intelligence can improve the efficiency of law enforcement agencies and improve the quality of crime investigation.

Before revealing the essence of one of the experiments conducted with GTP, let's consider the essence of this artificial intelligence. The history of GPT (Generative Pre-trained Transformer) has its roots in research conducted at OpenAI, an artificial intelligence laboratory, in order to create a powerful and versatile model for processing and generating natural language. When conceiving GPT, the researchers sought to create a model that could generate high-quality and natural text based on the studied samples of text data.

Let's also say that artificial intelligence is a general concept, and a neural network is a specific implementation of artificial intelligence. The first model, GPT-1, was introduced in 2018 and showed promising results in text generation. GPT-1 worked on the principle of pre-training and precise adjustment to specific tasks. This model used Transformer technology, which is a neural network with multiple layers of attention, capable of processing context and connections within text, improving the quality of generation.

In 2019, the GPT-2 model was introduced, which optimized previous results, expanding the power and capabilities of text generation. GPT-2 was trained on a huge amount of data, including online articles, books, news and other texts. Due to its unique architecture and multi-level processing, GPT-2 was able to generate long and high-quality texts, as well as overcome semantics-related tasks such as questions and answers.

After that, in 2020, the GPT-3 model was introduced, which is considered the most advanced and scalable version. GPT-3 is a huge neural network containing billions of parameters, which allows it to generate texts with amazing accuracy and stylistic naturalness. It has a wide range of applications, including automatic content creation, answering questions, translating text, developing dialog systems, and more.

As a result of numerous studies and technological improvements, GPT models have become a sought-after and successful technology in the field of natural language processing and generated text. OpenAI continues to develop and improve its models, striving to create even more sophisticated and efficient language processing systems based on artificial intelligence.

The analysis of the functionality of artificial intelligence, in this case, the GTP chat, allows us to conclude that it can be used in the process of putting forward versions in the detection and investigation of crimes. GTP, thanks to its software equipment and analytical capabilities, is able to take into account the investigative situation that arises at various stages of the investigation, as well as based on the achievements of the science of criminology, information about the presence of suspects or suspected persons and their characterizing materials to put forward reasonable versions about the person who committed the crime.

It is important to note that the use of artificial intelligence, especially GTP, in the forensic support of the process of disclosure and investigation of crimes has not been previously studied, which is the scientific novelty of this work.

Artificial intelligence, unlike human thinking, is not inclined to miss or distort the factual circumstances that have been provided to it, and also uses a variety of forensic investigative practices accumulated over many years to develop patterns and detect trends in the detection and investigation of crimes. When putting forward a version, artificial intelligence excludes subjective factors and objectively takes into account all circumstances, which contributes to the reliability of its conclusions.

Here is an example of the successful use of AI in criminology, in an attempt to verify how much artificial intelligence, in this case GTP, is able to correctly and objectively put forward versions regarding the perpetrators of a crime based on the data provided. The hypothesis was that artificial intelligence, specially designed to solve forensic tasks like our investigative exercises, is able to offer versions based on established patterns.

In this study, training and interaction with GTP takes place in Telegram via chat. All training materials are provided through this chat, and GTP tasks are set through it based on the information provided so that it can put forward versions and make descriptive fragments of protocols. This makes it possible to effectively use GTP in criminology and increases its capabilities in investigating and countering crimes.

For this purpose, we will demonstrate a dialogue in Telegrams between the author and the artificial intelligence GTP (Fig. 1):

Author: Have you learned how to put forward theories?

GPT + Midjourney: Yes, I can submit different versions or proposals according to the given context or requirements.

Author: Put forward a version of who could have committed the crime in accordance with the given context: "At about 3.00 o'clock Andreev woke up to the sound of the alarm of his car VAZ 21073. Andreev looked out the window and saw three young men running from his car, one of whom was carrying a backpack on his back. When Andreev went out to the parking lot, he noticed that the driver's window of his car was broken, a car radio, a phone holder, and a DVR were stolen from the cabin. Andreev did not notice any special signs of criminals, since it was dark."

GPT + Midjourney:

Version 1: According to the given context, it is likely that the three young men whom Andreev saw may be suspects in the commission of a crime. One of them was carrying a backpack on his back, which may indicate that he may have been responsible for stealing items from the car. However, due to the darkness, Andreev could not notice the special signs of the criminals, so a more accurate description of the identity of the suspects is not available.

Version 2: Based on the information, the criminal committed theft from Andreev's car. It is likely that the young people he saw could have been involved in this crime. The broken driver's window and stolen items such as a car radio, a phone holder and a DVR indicate that the crime was targeted. However, due to the darkness, Andreev could not notice the special signs of the criminals, complicating the identification of suspects.

Fig. 1.

During the research, artificial intelligence was trained based on GPT, the capabilities of AI were noted to analyze forensic tasks and develop their solutions in the same way as students studying a course in criminology do. AI was provided with tasks and materials that can be used to solve these tasks. GPT was effectively and accurately deployed, putting forward hypotheses, compiling programs and algorithms for the detection and investigation of crimes at different stages, depending on the progress of the investigation.

As part of this study, the possibility of using an artificial intelligence neural network to write a fragment of the descriptive part of the object inspection protocol was tested. To do this, the neural network was provided with detailed forensic photography of the knife, on the basis of which artificial intelligence compiled a fragment of the description of the object. Then the investigator used this fragment to compile a complete protocol, indicating the features of the examined knife (Fig. 2, 3).

Figure 2.

Fig.3

In addition, after training the neural network, she was given the opportunity to compile a fragment of the descriptive part of the document inspection protocol with signs of forgery. It is important to note that artificial intelligence drew attention to these signs and described them.

Currently, an experiment is being conducted in cooperation with the Main Investigative Department for the Sverdlovsk region. Within the framework of which the capabilities of artificial intelligence, namely GPT, are used to uncover real crimes, criminal cases for which are under investigation by investigators. The artificial intelligence was provided with materials of the criminal case, on the basis of which it was asked to put forward versions and develop a program for investigating the crime at the initial stage. Then we compared the proposed actions of the GPT chat with the actions carried out as part of the investigation, and found that artificial intelligence analyzes and plans activities necessary for the disclosure and investigation of crimes in sufficient detail and objectively. It can also help in the formation of examination questions and the definition of various types of examinations. Based on this, we can conclude that artificial intelligence can be effectively used in law enforcement not only to solve crimes, but also to prevent their commission, determine the circumstances and the crime scene. The results of the experiment will be described in more detail after the consideration of criminal cases in court and the pronouncement of sentences that have entered into force.

Also, during the study, the possibility of using a neural network after its training was tested in order to compile a fragment of the descriptive part of the incident scene inspection protocol based on orientation and survey photographs. AI coped with this task much better than most students studying the course of criminology.

In order for artificial intelligence to effectively describe objects and documents, as well as signs of forgery and features of the subject, samples of the descriptive part of protocols from real criminal cases were initially uploaded to it. After that, the AI was able to compile a descriptive part of the document inspection protocol, where it described the document in detail, as well as the signs of forgery.

Another experiment was to train GTP artificial intelligence to describe injuries based on forensic medical literature, in particular the Atlas of Forensic Medicine edited by Yu.And Pigolkin. Based on this atlas, GTP provided nodal and detailed images of corpses with various injuries, such as gunshot, knife, stabbed, cut, etc. Then the GTP was given the task to compile a descriptive part of the protocol of examination of the corpse based on photographs taken at the scene of the actual incident. The descriptive part of the protocol compiled by GPT was no worse than the descriptive part of the protocol compiled by the investigator, and in some details it was more detailed.

It is worth noting that similar issues have been raised by foreign authors. In most cases, without analyzing to one degree or another the use of artificial intelligence in criminalistic support for the disclosure and investigation of crimes, research is mainly related to analytical work in preparation for investigative actions or the investigation of certain types and categories of crimes ^{[13, 14, 15, 16, 17, 18, 19, 20]}

Criminology, as a science, was originally created to use the achievements of other branches of science and technology, such as mathematics, physics, chemistry, biology, medicine and computer science, for the suppression, detection, disclosure, investigation and prevention of crimes. In this regard, criminology should actively use advances in the field of artificial intelligence systems to solve problems of effective crime prevention.

As a conclusion, it is worth noting that artificial intelligence is able to provide significant assistance to law enforcement entities, but not replace them. The possibilities of using artificial intelligence systems for the purpose of solving and investigating crimes have not been determined at the moment. Neural network architectures effectively demonstrate the capabilities of artificial intelligence to analyze situations that arise during the disclosure and investigation of crimes and to put forward reasonable versions based on them on the circumstances of the commission of an illegal act. It is also necessary that special attention should be paid to high-quality training of the neural network on the use of modern forensic tools and methods in detection, disclosure and investigation, only after that it can be applied in real life.