Although we are still far from fully utilizing artificial intelligence (AI) technologies, they are now increasingly integrated with other technologies, which enhances their effectiveness. Among the areas where AI technologies are increasingly in demand is public safety, particularly in video surveillance systems. 

The use of AI in video surveillance creates numerous opportunities for more reliable public safety measures. However, there is another side to technological solutions: challenges related to the improper use of collected data, increased state control, and invasion of      privacy.

Features of AI-based video surveillance systems

Modern video surveillance systems using AI technologies can recognize and analyze images and sound from connected cameras. In general, most modern systems use "machine vision" technology, a series of algorithms or mathematical models that compare      the observed object with many others in databases to accurately identify the desired object. 

In this context, AI can analyze not only the appearance of an object but also its properties, such as speed, movement, position, etc. Thus, AI can be used to obtain maximum information about the observed objects in a given location and identify them accordingly.

Moreover, the most advanced AI technologies in video surveillance can not only identify the objects they see but also independently form their own standard of what a proper      situation on the streets and roads should look like. Through self-learning, AI technologies can determine what constitutes normal behavior, such as that of vehicles or individuals. So, if a car is moving on the sidewalk or a person is standing in the middle of the road, AI can identify this as a deviation from the norm or a violation based on its own autonomous decisions.

Prospects and Risks of Using AI in Video Surveillance

As noted by Rahul Yadav, Director of Milestone Systems, AI-based video surveillance systems offer many opportunities to enhance the safety of citizens, such as:

1. Recognizing dangerous human interactions, such as fights, falls, or criminal activities.

2. Understanding crowd behavior to identify early signs of a conflict.

3. Detecting anomalies, such as intrusions beyond the security perimeter.

4. Analyzing long-term trends to predict risks before a specific incident may occur.

5. Correlating the collected information with other capabilities of various systems, such as access control systems, audio detection systems, etc.

Data obtained through AI-based video surveillance can be used in criminal and administrative proceedings. Artificial intelligence can quickly analyze video or audio materials and provide its assessment, determining what exactly occurred and whether the documented event constitutes a violation of the law. Theoretically, this could save time      conducting investigations and reduce the risk of human subjectivity. 

Data collected by such systems can be organized by AI into a database and used both for self-learning and for identifying subjects who have previously violated the law. Thus, the use of AI in video surveillance systems benefits the development of predictive analytics. 

Such opportunities for using AI have special prospects in the security sector. However, these technologies also raise discussions about the consequences that may arise from the use of AI in video surveillance, particularly concerning privacy. 

Firstly, AI technologies face issues of bias and accuracy concerning unrepresentative data, thereby creating precedents that could exacerbate discriminatory risks. This is a significant problem, as AI may identify some circumstances incorrectly or ignore them altogether. Such concerns stem from a large number of precedents in which AI has misidentified people and confused them with criminals, particularly in the United States, namely in Texas, Detroit, California, and New Jersey. 

Risks associated with the use of AI are also present in the field of predictive analytics. AI-based video surveillance systems can also exhibit bias against certain ethnic groups, misidentifying them on video and incorrectly interpreting their behavior. For instance, the COMPAS system, an algorithm for assessing the risk of recidivism, is prone to erroneously assign high-risk scores to African American defendants. 

The problems include the so-called "black box" dilemma and the issue of transparency of AI decision-making algorithms. The term "black box" refers to the difficulties in determining the cause-and-effect relationship behind AI's decision-making, as the process of identifying the algorithm that influenced a specific decision is often unclear and poorly understood. This creates distrust in the results produced by AI and raises concerns about the quality of the data on which it is trained. 

Secondly, with the rapid development of AI, society is increasingly viewing such technologies with suspicion and prejudice. 

Naturally, in this situation, the state must play a significant role by proposing regulatory standards that protect the population from improper data collection and use.  

However, the state itself can use AI technologies not only for public safety purposes but also for surveillance and control, which is especially dangerous for citizens in countries with authoritarian regimes. 

The Practice of Using AI-Based Video Surveillance Systems Worldwide and in Ukraine

AI-based video surveillance systems have already begun to be widely used in many countries around the world. 

In particular, a high-profile precedent for the deployment of AI-enabled mass video surveillance technologies took place in France on the eve of the 2024 Summer Olympics in Paris. The French government has enacted a series of laws allowing French law enforcement agencies and their technical contractors to experiment with intelligent video surveillance systems and utilize AI technologies for analyzing video materials. 

These innovations have raised concerns among civil society, as questions about the proper and lawful use of such technologies by EU member states have emerged, especially regarding the EU legislation on AI and data protection. Regulation in France contradicts the requirements of the EU AI Act and its provisions on preventing the harmful use of AI technologies. Additionally, French legislation permits experimentation with AI not only by law enforcement agencies but also by technical contractors, which are private companies. This has raised the issue of compliance with the principles of the General Data Protection Regulation (GDPR). 

The French government is not the only one trying to use AI in video surveillance. Before France, during the Tokyo Olympics in 2021, the Japanese government attempted to do the same. Experiments with AI and surveillance also took place in the UK, where these technologies proved to be relatively effective.

However, the most widespread deployment of video surveillance systems with AI elements is taking place in China. It is evident that the Chinese government uses mass surveillance systems not only for public security purposes but also to strengthen surveillance and control of the population. According to 50 publicly available documents analyzed by Reuters, dozens of organizations in China purchased software called "one person, one file" between 2018 and 2022. 

"One person, one file" is a specific program that can accurately identify and archive the appearance of a face, even if it is hidden under a mask or glasses. Furthermore, this system has the ability to self-learn and create the required number of files depending on the amount of data received. Alongside developing and implementing the "one person, one file" software, the Chinese government began the mass installation of surveillance cameras known as "Sharp Eyes," designed to collect data and integrate it into a nationwide platform for data sharing. 

In Ukraine, initiatives are also proposed for the deployment of mass video surveillance systems utilizing AI technologies. Active processes in this direction already took place in the largest cities of Ukraine between 2015 and 2019. For example, in March 2021, the current head of the Lviv police, Oleksandr Shliakhovskyi, spoke about "total video surveillance," stating intentions to unify all existing surveillance cameras in Lviv Oblast into a single system as part of the Safe Lviv Region project and emphasizing the feasibility of using facial recognition technology. 

Even then, questions were raised about the appropriateness and correctness of using video surveillance systems with facial recognition technology, and solutions to these issues were proposed.

In February 2024, a draft law was submitted to the Verkhovna Rada of Ukraine aimed at regulating video surveillance systems in Ukraine within the framework of the Safe City project. Although the draft law does not explicitly mention the use of AI technologies, the "automated analysis of digitized data" provided for in it clearly implies the use of these technologies.

The mentioned draft law aims to implement the automated collection of metadata in real time. Such metadata includes the person's name, date of birth, information about their registered or declared place of residence (or stay), a digitized image of the person's face, and the registration number of the vehicle. Accordingly, this data will be used to identify a particular individual. The highest priority objects for video monitoring will include public places, areas of general use, educational institutions, healthcare facilities, and certain objects and items. 

Such intentions of Ukrainian lawmakers raise concerns among many experts and human rights defenders, as the draft law has numerous issues, particularly those related to the definitions of the types of information subject to collection and the reasonable limits on the collection and analysis of information.  

These "gaps" may be considered somewhat acceptable if video surveillance systems using AI technologies are used exclusively by law enforcement agencies. However, the draft law allows private individuals and companies to install such cameras, provided that the police have continuous access to them. So, this jeopardizes the privacy of individuals captured on video, as the collected data may be misused. 

Another issue is that the cameras being installed in Ukrainian cities are produced in China, specifically by companies like Hikvision and Dahua. For example, the "Skhemy" (Schemes) investigation emphasizes that cameras from Chinese companies are vulnerable to use for espionage and hacking purposes and that Chinese intelligence services may have access to the data collected through them.

After all, concerns about the Chinese government’s control over such technologies are not unreasonable. Following the large-scale protests in Hong Kong in 2019, China has implemented a policy of maximizing control over its citizens by utilizing AI technologies.

In this context, experts from the Institute of Mass Information emphasize the issues that require additional regulation, notably the need to specify the types and categories of data that can be collected during video surveillance, establish reasonable limits on data collection and analysis, and ensure the protection of citizens' rights to respect for their private and family lives.

Conclusions and Recommendations 

The use of AI systems in video surveillance can be a promising tool for ensuring public safety. An effective video surveillance system can strengthen security measures, especially during martial law, when law enforcement agencies deal not only with public order but also with intelligence and sabotage operations of the enemy country. 

However, it is also clear that the prospect of using AI-based video surveillance may cause a number of problems that are not currently addressed in the draft law. 

One of the most important issues is whether the authorities are able to effectively protect the collected data. This is especially relevant in the context of Russia's war against Ukraine. The possibility of data leakage that could be used by attackers or intelligence services of a hostile country to harm Ukrainian citizens cannot be ruled out. 

In addition, secure suppliers of relevant technologies must be selected. As noted above, the Skhemy (Schemes) notes that most cameras in Ukraine are produced by Chinese companies, which potentially gives the Chinese government the ability to obtain information not only about Ukrainian citizens but also about strategic objects in Ukraine. Combined with AI technologies and given the alliance between Moscow and Beijing, this could pose critical risks to Ukraine's national security.

The reaction of civil society representatives to the draft law initiative shows that there are concerns about the constant monitoring and collection of private information. Obviously, such concerns increase the level of public anxiety, especially in the context of martial law and the suspension of democratic processes in the country. 

Problems related to AI technologies themselves, assessment of their application, and resolution mechanisms also remain uncertain. Bias and systemic errors will likely accompany the Ukrainian law enforcement system. Addressing these issues will depend on the overall progress in the development of AI technologies. 

In this regard, the implementation of such technologies must include the following measures:

1. Ensuring accountability: Implementing a large-scale surveillance system is unfeasible without effective oversight mechanisms to monitor its operation and the actions of agencies and officials who will access the collected data. If AI elements are integrated into such a system, it is necessary to provide control mechanisms for these technologies as well.

Institutional mechanisms for overseeing law enforcement activities currently lack the capacity to ensure effective monitoring of the use of such a system, especially in the context of war. The activities of the Commissioner for Human Rights of the Verkhovna Rada in the field of human rights and privacy protection are effective but limited due to a lack of resources and an adequate number of staff. As for the regulation and control of the use of AI technologies, the government plans to introduce appropriate regulation after the martial law is lifted.

Moreover, there are no clear and consistent mechanisms for handling complaints and resolving disputes regarding the implementation of such a system.

In this regard, the introduction of such technologies in the short term seems unlikely to ensure a balance between national security interests and human rights. This can be done on the basis of a risk-based approach after the relevant regulatory and institutional mechanisms are established or improved.

2. Vendor control: It is important to carefully choose the suppliers of both video surveillance cameras and AI software and technologies that will ensure their functioning as a single system. National security interests must prevail over commercial considerations, speed of delivery, or ease of integration. The process must be authorized at the highest level with the involvement of representatives of the expert community and civil society.

3. Compliance with the European integration obligations: Automated processing of personal data on such a scale may constitute a severe interference with the privacy of millions of Ukrainians and residents of Ukraine, which would not meet the EU standards. In this respect, it is necessary to verify the implementation of such a system for compliance with the EU legislation, particularly regarding data protection (GDPR) and artificial intelligence (AI Act), as well as the overall adherence to the European Union's requirements for ensuring human rights and freedoms and the accountability of law enforcement agencies.

4. Cooperation with the Ukrainian IT industry: Issues such as data quality, "black box", and AI bias are already being addressed by leading IT companies. For example, Statworx provides three methods for dealing with black box and bias problems: the partial function importance method, the SHAP function method, and the local effect accumulation method. Each of these methods is aimed at identifying the most important variables for AI that influence its decision-making. 

Trigyn Technologies also describes a number of measures that should be taken to ensure the proper quality of the information that AI uses for its training. The most important are the ways of data labelling, as well as proper tracking, increasing, and management. At the same time, it is important to remove "noise", i.e. information that is irrelevant and harmful to AI learning. 

Ukrainian IT companies can also focus their efforts on solving the most common AI problems by developing new products on the technology market. The development of such projects by IT companies can be supported by government orders or other types of financial support.

Any regulation of AI technologies in video surveillance should be based on proper human rights protection. In this context, it is essential to establish legal protection mechanisms against unfair decisions caused by the use of AI-based systems, as well as accountability for violations of personal data processing principles. In the use of AI-based video surveillance, it is crucial to prioritize transparency and accountability, as these principles will help achieve a balance between national security interests and civil society.