On Wednesday, March 13, the European Parliament approved the AI Act, the first regulation of its kind encompassing a broad area of artificial intelligence. The aim of the new regulations is to build a legal environment that will allow for the development of this technology, while respecting basic rights, protecting democracy, and maintaining the rule of law in the face of high-risk systems. – “Given the many articles that may not necessarily be easily and coherently interpretable, I would suggest that the European Commission, but this could also be Poland’s role during the presidency, prepare an implementation plan for regulations on artificial intelligence,” assessed former Minister of Digitalization Michał Boni.
The Artificial Intelligence Act anticipates, among other things, categorizing prohibited practices within AI solutions, particularly those that threaten the rights of citizens. These include the use of subliminal techniques as well as discriminatory, manipulative ones that exploit human weaknesses. The new rules will ban the use of artificial intelligence systems for so-called social scoring, or a point classification of citizens. They will also limit the use of biometric identification systems for law enforcement – in this case, some exceptions are provided, which are narrowly defined in a closed list. In addition, there will be an obligation to clearly mark non-authentic or manipulated images, audio or video content.
“Artificial intelligence in itself is neither bad nor good, it’s neutral. The applications of artificial intelligence can be varied and often depend on people, countries, and companies. Regulations are needed to analyze the risks that specific applications of artificial intelligence may bring,” says Michał Boni, a lecturer at SWPS University and former Minister of Digitalization. A key factor will be assessing these risks and their potential impact on citizen safety.
“If the risks are too great, there may need to be a limitation on the implementation of a solution, and if they are smaller, but still exist, then the company should undergo a test to check whether these risks will impact people, society, and the external environment,” says Michał Boni. “Such tests have been prepared by a group of experts dealing with the ethical dimensions of artificial intelligence. So it is controllable, but we need to practically approach it from the perspective of applications and risks, not from a general discussion about artificial intelligence.”
The AI Act sets out special obligations for high-risk systems. The areas in which they are used include critical infrastructure, education, vocational training, employment, and basic private and public services (e.g., healthcare, banking). Some law enforcement systems also rely on them. They are used to manage migration and borders, for the justice system, and democratic processes, such as elections. Such systems must assess and limit risk and maintain a registry of events. They must also be transparent and accurate, and subject to human oversight. In addition, citizens will gain the right to file complaints about AI systems.
As the expert emphasizes, creating this type of legislation is a challenge due to the universality of individual provisions. The goal is to make them applicable to future, unknown risks. Hence, many provisions may seem unclear.
“With many articles that may not necessarily be easily and coherently interpretable, I would suggest that the European Commission, but this could also be Poland’s role during the presidency, prepare a key document, namely a plan to implement AI regulations. This is complicated because some things from this act will come into force a few months after adoption, others in a year, others in eighteen months, others in two years, up to three, so it will take time. In order for it to be done sensibly, such an implementation plan should be prepared,” assesses Michał Boni. “Where different things are not entirely clear, we should go in the direction of a code of good practice. This means that businesses should agree among themselves, along with other partners, such as users, which solution is good and respects the spirit of the law on artificial intelligence, which is among other things intended to protect our subjectivity.”
According to Deloitte, the progressing digitization of the economy means that more and more organizations are deciding to use tools based on artificial intelligence. The “Unlocking the power of AI” report shows that over 90% of companies aim to achieve a higher level of maturity in AI within the next three years. The pace and realization of their plans, however, depend on factors such as staff qualifications and the ability of an organization to implement machine learning solutions. The authors of the cited report estimate that the business value created by AI could amount to as much as 4.4 trillion dollars over the next two years.
Many sectors expressed concerns at the drafting stage of the new legislation that the introduction of the mentioned regulations could slow down the development of the artificial intelligence market. Michał Boni does not share this assessment.
“I think the market will continue to develop as dynamically as it is now because many companies already apply these rules and mechanisms, such as risk analysis, which are included in this regulation. However, good advice will certainly be needed, especially for small businesses, on how to interpret certain provisions, but also primarily whether the applications of artificial intelligence that they want to implement, some prediction, foreseeing based on the data what might happen to the business, whether it creates any risks or not,” says the former Minister of Digitalization. “In many cases, it will not create any risk at all. Remember, this regulation points to high-risk systems, but they are between 10 and 15 percent. Eighty-five percent of artificial intelligence systems operate without any risk, so there will not be any radical, universal change.”
The EU regulation envisages the establishment of a European Council for Artificial Intelligence. In addition, national bodies dealing with this issue are to be established in member states. The main task of these institutions will focus on the effective enforcement of the provisions of the regulation in each EU country. On the national level, there will also be so-called regulatory sandboxes, which aim to facilitate companies in developing innovations and conducting tests of new solutions before they are put into use.
Artificial Intelligence is a very broad concept that first appeared in a theoretical approach in science in 1956 thanks to American computer scientist John McCarthy. Over time, it has been defined as the ability of a system to correctly interpret data from various external sources, and to automatically learn from collected information and then perform specific tasks. Currently, the term “artificial intelligence” is associated with various popular solutions in the field of automatic text creation, graphics, translations, and audio-visual materials. In reality, it is a much broader concept touching on areas such as neural networks, machine learning, data processing (including Big Data), image, sound and speech recognition, and artificial creativity.
