EU AI REGULATION: A STEP FORWARD OR AN ATTEMPT TO HALT PROGRESS?

Abstract

On March 13, 2024, the European Parliament approved the draft “Regulation of the European Parliament and of the Council laying down harmonized rules on artificial intelligence”, and on May 21 of the same year, the Council of the European Union endorsed the said act as well. As a result, the regulation will enter into force in July 2024. It will become the world’s first legislative act to regulate all types of artificial intelligence available in the private sector.

Artificial intelligence remains one of the most important challenges of the modern world. Technologies related to it are developing at a high speed, affecting all industries and individuals. As a result, it is necessary to regulate the field as effectively as possible; however, before the aforementioned regulation, there was practically no legal framework, which especially increased its importance. It is necessary to conduct an effective analysis of the given regulation to develop an even more effective regulatory framework in the future.

Keywords: Artificial Intelligence, AI, EU Law, Regulation, New Regulation

Introduction

On March 13, 2024, the European Parliament approved, by a majority vote, the draft “Regulation of the European Parliament and of the Council laying down harmonized rules on artificial intelligence”, and on May 21 of the same year, the Council of the European Union unanimously approved the said act.[1] As a result, the regulation will enter into force in July 2024 and will become the world’s first legislative act that aims to regulate all types of artificial intelligence available in the private sector.[2]

Artificial intelligence remains one of the most important challenges of the modern world.[3] Technologies related to it are developing at a high speed,[4] which affects almost all industries[5] and even the life of an average person.[6] As a result, it is necessary to regulate the field as effectively as possible, however, before the aforementioned regulation, there was practically no legal framework, which especially increases its importance.

The draft regulation was first presented by the European Union Commission on April 21, 2021.[7] As a result, during the following 3 years, many different kinds of opinions were expressed about it, both with positive and negative evaluations.[8] However, despite the polarized reaction, the regulation was adopted without radical content changes.[9]

Considering the above, it can be said that despite the newness of the regulation, there are available academic studies related to it that have been conducted in the last three years. Using these papers and primary research, the purpose of the present paper is to evaluate the adopted regulation and provide a critical analysis of its text and potential implications.

1. The Primary Stipulations of the Regulation

Considering that the EU regulation is, in fact, the first attempt to regulate the new field, it is natural that many new provisions are gathered in it.[10] This means that there are various question marks regarding the text of the regulation, to which it is not possible to give a firm answer so far, given the lack of practical application thereof.[11]

First, it should be noted that the scope of the regulation includes the use of artificial intelligence both in the private sector and for public purposes.[12] At the same time, several exceptions are allowed in the normative act, which includes the use of artificial intelligence in the military and defence fields, as well as its use exclusively for research and scientific purposes.[13]

As for the content of the regulation, it has introduced the so-called A “risk-based” approach, in which the greater the risk arising from artificial intelligence, the stricter the rules that apply to it.[14] According to the regulation, artificial intelligence was divided into three categories based on the risk arising from the relevant AI:

Unacceptable risk - systems that are perceived as a threat to humanity. Among them, according to the examples given in the regulation, are cognitive behavioural manipulation (both of the entire population and any of its vulnerable parts), social assessment (categorizing people and/or scoring them according to their behaviour, social status or personal characteristics), as well as biometric identification and artificial intelligence focused on determining the location of people. The creation or use of artificial intelligence that includes any type of such unacceptable risk is fully prohibited throughout the EU.[15]

High risk - according to the regulation’s text, all artificial intelligence that negatively affects the basic rights of people or their safety is considered high risk. These include all types of artificial intelligence used in aviation, medical devices, automobiles, elevators, toys, infrastructure operations, education, labour and employee management, utilities, policing, migration, and legal aid. All types of high-risk artificial intelligence, before they become available to the public, must be evaluated by the relevant authority and, subsequently, subject to periodic inspection.[16]

Low risk - Every AI that does not fall into the above two categories is considered low risk. Among them is, for example, generative artificial intelligence (for example, ChatGPT).[17] Low-risk artificial intelligence is not subject to additional regulation, although it is still subject to transparency requirements.[18]

According to the regulation, when any kind of product is created using artificial intelligence, it must be identified as having been created by artificial intelligence. In addition, the manufacturer is obliged to ensure that it is impossible to create any illegal product using artificial intelligence.[19] It is also worth noting that the regulation introduces mandatory rules of conduct for all persons developing artificial intelligence, the purpose of which is to further increase the level of security.[20]

Another important issue regulated by the regulation is the creation of relevant bodies related to artificial intelligence issues. An artificial intelligence office and a scientific panel of independent experts will be created within the commission. Additionally, an Artificial Intelligence Council formed by representatives of the member states will be created, with which a forum of interested parties will work, which will periodically submit recommendations to the Council.[21]

If an AI producer violates the rules set out by the regulation, it will be subject to a fine, the amount of which will be determined by the Commission’s Artificial Intelligence Office,[22] which may represent a set percentage of the company’s global turnover, which in some cases may be an exceptionally large amount, increasing the Office’s Probability of successful enforcement of decisions and the regulation requirements in practice.[23]

Finally, it should be noted that the regulation also aims to promote innovation. According to the regulation, member states are obliged to support independent businesses and start-ups, as well as  private individuals, in researching and creating artificial intelligence in such an environment that it can be presented to the general public only after it is confirmed that the risks associated with it are minimized.[24]

The regulation enters into force gradually and will be fully effective within 24 months after its publication. However, there are special provisions related to high-risk artificial intelligence, which will enter into force within 36 months after publication.[25]

2. The Set Purpose and the Fulfillment thereof by the Text of the Regulation

In the preamble of the regulation, it is emphasized that its purpose thereof is to ensure the protection of basic human rights and to limit as much as possible their violation using artificial intelligence.[26] However, it is important to highlight that the regulation deals only with some issues related to artificial intelligence and cannot be considered a normative act that regulates all topics associated with the creation and use of artificial intelligence.[27]

Today, there is an active debate on the creation of artificial intelligence and the so-called “training” thereof.[28] According to the current standard practice, when programming artificial intelligence, it is provided with a particularly large amount of information to skim over, as a result of which artificial intelligence develops and acquires new capabilities due to the processing thereof.[29] For example, an artificial intelligence, the function of which is to create paintings, learns by “reading” pictures drawn by humans, as a result of which it can create a similar painting itself.[30] Therefore, the artists whose works are used by artificial intelligence created by various companies believe that their copyrights are being infringed.[31] The situation is essentially identical in all other fields of art, in which artificial intelligence is utilized.[32] This issue appears to be one of the most important parts of the academic and legal discussion about artificial intelligence and its impact, although the regulation does not even touch on it and leaves it beyond regulation.

Another topic that does not appear in the text of the regulation is crimes committed using artificial intelligence. Artificial intelligence is being used more and more frequently in the online spaces for illegal purposes,[33] although the regulation does not propose to regulate this issue either.

Hence, it can be said that, on the one hand, the steps taken by adopting this regulation are important, and that the European Union is the first to create a legislative framework to regulate this extremely important area.[34] At the same time, it should be emphasized that the issue is not exhausted, and the regulation of artificial intelligence should be continued to create a normative system that ensures that the goal defined in the preamble of the adopted regulation is truly and fully achieved.

3. Potential Influence on the Development of Artificial Intelligence

As with almost all new arrangements, the EU’s first submission of a draft regulation on the introduction of harmonized rules on artificial intelligence was met with a negative reaction from members of the relevant industry.[35] Opinions were expressed that by regulating the field, the European Union would only ensure that the development of the said field would be limited within its framework and that the global competitors of the European Union would be given an advantage.[36]

In connection with the above, it is necessary to emphasize that the regulation, in fact, happens to be merely a general framework.[37] The vast majority of the rules therein can be divided into two categories: the outline, within the framework of which a more detailed regulation needs to be created later, and the general rules that people working on artificial intelligence, within the framework of the principle of good faith, should already be following.[38] In fact, there are only two hard, concrete rules that would fundamentally change the situation: creating a regulatory body and prohibiting the development of artificial intelligence that poses an unacceptable risk.[39]

Due to the importance and complexity of artificial intelligence, it is virtually impossible to regulate it without the existence of a specialized body.[40] Admittedly, bad regulation can often be worse than no regulation at all,[41] but as things stand, there is no reason to believe that the yet-to-be-formed EU Artificial Intelligence Office will do its job poorly. At the same time, considering that a number of companies working in the field of artificial intelligence have already had scandals related to security and human rights,[42] it can be said that the existence of the regulation is necessary. As a result, it is not justified to say that the establishment of the Artificial Intelligence Office by the Regulation of the European Union Parliament and the Council will in any way have a negative impact on the given field.

As for artificial intelligence containing an unacceptable risk, as already mentioned above, work on such tools (except for specialized parties working in the field of defense and security) has been prohibited for everyone. Critics of the regulation point out that with this step, the European Union lags behind, for example, the People’s Republic of China, in which human monitoring and social classification is an important direction of the development of artificial intelligence.[43]

The preamble of the regulation explicitly states that artificial intelligence containing unacceptable risk is associated with actions that cannot be entrusted to artificial intelligence.[44] All of the areas identified as containing unacceptable risks in the regulation are related to issues that have a particularly large impact on people’s daily lives.[45] Therefore, interference therein, which is often conducted via artificial intelligence,[46] can in no way be deemed acceptable.[47] Therefore, the decision taken by the European Union to ban the creation of such artificial intelligence cannot be considered an undesirable step.

As a result, it can be stated that the provisions of the regulation will not have any significant negative impact on the development of artificial intelligence. On the other hand, it can also be said that creating a legal framework can have a net positive effect on the mentioned field.

First of all, it is often desirable for businesses to establish rules of conduct and clearly define prohibited actions, as it eliminates ambiguity, which is usually considered to be a hindering factor to business.[48] In addition, provisions related to transparency will increase the number of sources available to start-ups and average businesses, which will also have a positive impact on the level of development of artificial intelligence in the future.[49]

Considering all of the above, it can be said that the criticism of the regulation from the point of view that it will hinder the development of artificial intelligence in the European Union is without foundation. On the contrary, its introduction will reduce ambiguity and provide transparency, which will positively impact the entire field both within and outside the EU.

Conclusion

Adopting the draft “Regulation of the European Parliament and of the Council laying down harmonized rules on artificial intelligence” is an important step forward in regulating the field of artificial intelligence. This is the first normative act of comparable scale to regulate the research, creation, and use of artificial intelligence by the private sector. Adopting it creates a framework within which artificial intelligence and the people working on it can evolve. The regulation does not prevent the persons working in the field from effectively performing their activities; on the contrary, it will positively affect them.

However, the mentioned regulation remains only the first step. Based on this, it is necessary to develop a more detailed set of rules that will regulate all important areas related to the use of artificial intelligence. The special bodies created within the framework of the regulation are obliged to ensure the correct application of the rules in practice. Only in this case it will be possible to develop artificial intelligence in such a way that the main goal specified in the preamble of the regulation - the well-being of humankind - is achieved.

Bibliography

1. Alon-Barkat, S. Busuioc, M. Human–AI Interactions in Public Sector Decision Making: „Automation Bias” and „Selective Adherence” to Algorithmic Advice. (2023) Journal of Public Administration Research and Theory, 33(1). pp. 153-169.


2. Baronchelli, A. Shaping new norms for AI. (2024) Philosophical Transactions of the Royal Society, 379. pp. 1-6.


3. European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024].


4. Gaffar, H. Albarashdi, S. Copyright Protection for AI-Generated Works: Exploring Originality and Ownership in a Digital Landscape. (2024) Asian Journal of International Law, 24. pp. 1-25.


5. Gibney, E. What the EU’s tough AI law means for research and ChatGPT. (2024) Nature, 626. pp. 938-939.


6. Gornet, M. Maxwell, W. The European approach to regulating AI through technical standards. (2024) The HAL Open Science Journal, 24. pp. 1-25.


7. Hacker, P. A Legal Framework for AI Training Data - from First Principles to the Artificial Intelligence Act. (2021) Law, Innovation and Technology Journal, 13(2). pp. 257-301.


8. Hakan, C. Criminal Liability of Artificial Intelligence from the Perspective of Criminal Law: An Evaluation in the Context of the General Theory of Crime and Fundamental Principles. (2024) International Journal of Eurasia Social Sciences, 55(15). pp. 276-313.


9. Laux, J. Wachter, S. Mittelstadt, B. Three Pathways for Standardisation and Ethical Disclosure by Default under the European Union Artificial Intelligence Act. (2024) Computer Law & Security Review, 53. pp. 1-11.


10. Luk, A. The Relationship Between Law and Technology: Comparing Legal Responses to Creators’ Rights under Copyright Law through Safe Harbor for Online Intermediaries and Generative AI Technology. (2023) Law, Innovation and Technology Journal, 16(1). pp. 148-169.


11. Mendez-Suarez, M. Virginia, S.M. De Prat, J.M. Do Current Regulations Prevent Unethical AI Practices? (2023) Journal of Competitiveness, 15(3). pp. 207-222.


12. Mitchell, A.D. Let, D. Tang, L. AI Regulation and the Protection of Source Code. (2023) International Journal of Law and Information Technology, 31(4). pp. 183-301.


13. Novelli, C. Casolari, F. Rotolo, A. Taddeo, M. Floridi, L. AI Risk Assessment: A Scenario-Based, Proportional Methodology for the AI Act. (2024) The Digital Society Journal, 3(13). pp. 1-29.


14. Pirozzoli, A. The Human-centric Perspective in the Regulation of Artificial Intelligence. (2024) Insight European Papers, 9(1). pp. 105-116.


15. Sganga, C. Is There Still a Policy Agenda for EU Copyright Law? (2023) International Review of Intellectual Property and Competition Law, 54. pp. 1407-1417.


16. Tataru, S.R. Cretu, A.C. Decoding the EU Artificial Intelligence Act: An Analysis of Key Concepts and Provisions. (2024) Journal of Public Administration Finance and Law, 31. pp. 463-474.


17. Voss, W.G. AI Act: The European Union’s Proposed Framework Regulation for Artificial Intelligence Governance. (2021) Journal of Internet Law, 25(4). pp. 8-17.

Footnotes

[1] Gornet, M. Maxwell, W. The European approach to regulating AI through technical standards. (2024) The HAL Open Science Journal, 24. pp. 1-25.

[2] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024].

[3] Voss, W.G. AI Act: The European Union’s Proposed Framework Regulation for Artificial Intelligence Governance. (2021) Journal of Internet Law, 25(4). pp. 8-17.

[4] Alon-Barkat, S. Busuioc, M. Human–AI Interactions in Public Sector Decision Making: “Automation Bias” and “Selective Adherence” to Algorithmic Advice. (2023) Journal of Public Administration Research and Theory, 33(1). pp. 153-169.

[5] Hacker, P. A Legal Framework for AI Training Data - from First Principles to the Artificial Intelligence Act. (2021) Law, Innovation and Technology Journal, 13(2). pp. 257-301.

[6] Tataru, S.R. Cretu, A.C. Decoding the EU Artificial Intelligence Act: An Analysis of Key Concepts and Provisions. (2024) Journal of Public Administration Finance and Law, 31. pp. 463-474.

[7] Voss, W.G. AI Act: The European Union’s Proposed Framework Regulation for Artificial Intelligence Governance. (2021) Journal of Internet Law, 25(4). pp. 8-17.

[8] Laux, J. Wachter, S. Mittelstadt, B. Three Pathways for Standardisation and Ethical Disclosure by Default under the European Union Artificial Intelligence Act. (2024) Computer Law & Security Review, 53. pp. 1-11.

[9] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024].

[10] Hacker, P. A Legal Framework for AI Training Data - from First Principles to the Artificial Intelligence Act. (2021) Law, Innovation and Technology Journal, 13(2). pp. 257-301.

[11] Laux, J. Wachter, S. Mittelstadt, B. Three Pathways for Standardisation and Ethical Disclosure by Default under the European Union Artificial Intelligence Act. (2024) Computer Law & Security Review, 53. pp. 1-11.

[12] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Article 2.1.

[13] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Article 2.2, Preamble Article 16.

[14] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Preamble Article 14.

[15] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Chapter II.

[16] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Chapter III.

[17] Pirozzoli, A. The Human-centric Perspective in the Regulation of Artificial Intelligence. (2024) Insight European Papers, 9(1). pp. 105-116.

[18] Tataru, S.R. Cretu, A.C. Decoding the EU Artificial Intelligence Act: An Analysis of Key Concepts and Provisions. (2024) Journal of Public Administration Finance and Law, 31. pp. 463-474.

[19] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Chapter IV.

[20] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Chapter IX.

[21] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Chapter VI-VIII.

[22] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Chapter X.

[23] Tataru, S.R. Cretu, A.C. Decoding the EU Artificial Intelligence Act: An Analysis of Key Concepts and Provisions. (2024) Journal of Public Administration Finance and Law, 31. pp. 463-474.

[24] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Chapter V.

[25] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Article 85.

[26] European Union Law Electronic Journal (Eur-Lex). Online Database. Available online at: https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=celex:52021PC0206 [Last Access: 29.05.2024]. Preamble.

[27] Tataru, S.R. Cretu, A.C. Decoding the EU Artificial Intelligence Act: An Analysis of Key Concepts and Provisions. (2024) Journal of Public Administration Finance and Law, 31. pp. 463-474.

[28] Gibney, E. What the EU’s tough AI law means for research and ChatGPT. (2024) Nature, 626. pp. 938-939.

[29] Luk, A. The Relationship Between Law and Technology: Comparing Legal Responses to Creators’ Rights under Copyright Law through Safe Harbor for Online Intermediaries and Generative AI Technology. (2023) Law, Innovation and Technology Journal, 16(1). pp. 148-169.

[30] Sganga, C. Is There Still a Policy Agenda for EU Copyright Law? (2023) International Review of Intellectual Property and Competition Law, 54. pp. 1407-1417.

[31] Gaffar, H. Albarashdi, S. Copyright Protection for AI-Generated Works: Exploring Originality and Ownership in a Digital Landscape. (2024) Asian Journal of International Law, 24. pp. 1-25.

[32] Luk, A. The Relationship Between Law and Technology: Comparing Legal Responses to Creators’ Rights under Copyright Law through Safe Harbor for Online Intermediaries and Generative AI Technology. (2023) Law, Innovation and Technology Journal, 16(1). pp. 148-169.

[33] Hakan, C. Criminal Liability of Artificial Intelligence from the Perspective of Criminal Law: An Evaluation in the Context of the General Theory of Crime and Fundamental Principles. (2024) International Journal of Eurasia Social Sciences, 55(15). pp. 276-313.

[34] Voss, W.G. AI Act: The European Union’s Proposed Framework Regulation for Artificial Intelligence Governance. (2021) Journal of Internet Law, 25(4). pp. 8-17.

[35] Laux, J. Wachter, S. Mittelstadt, B. Three Pathways for Standardisation and Ethical Disclosure by Default under the European Union Artificial Intelligence Act. (2024) Computer Law & Security Review, 53. pp. 1-11.

[36] Tataru, S.R. Cretu, A.C. Decoding the EU Artificial Intelligence Act: An Analysis of Key Concepts and Provisions. (2024) Journal of Public Administration Finance and Law, 31. pp. 463-474.

[37] Voss, W.G. AI Act: The European Union’s Proposed Framework Regulation for Artificial Intelligence Governance. (2021) Journal of Internet Law, 25(4). pp. 8-17.

[38] Pirozzoli, A. The Human-centric Perspective in the Regulation of Artificial Intelligence. (2024) Insight European Papers, 9(1). pp. 105-116.

[39] Tataru, S.R. Cretu, A.C. Decoding the EU Artificial Intelligence Act: An Analysis of Key Concepts and Provisions. (2024) Journal of Public Administration Finance and Law, 31. pp. 463-474.

[40] Novelli, C. Casolari, F. Rotolo, A. Taddeo, M. Floridi, L. AI Risk Assessment: A Scenario-Based, Proportional Methodology for the AI Act. (2024) The Digital Society Journal, 3(13). pp. 1-29.

[41] Baronchelli, A. Shaping new norms for AI. (2024) Philosophical Transactions of the Royal Society, 379. pp. 1-6.

[42] Mitchell, A.D. Let, D. Tang, L. AI Regulation and the Protection of Source Code. (2023) International Journal of Law and Information Technology, 31(4). pp. 183-301.

[43] Mendez-Suarez, M. Virginia, S.M. De Prat, J.M. Do Current Regulations Prevent Unethical AI Practices? (2023) Journal of Competitiveness, 15(3). pp. 207-222.

[44] Novelli, C. Casolari, F. Rotolo, A. Taddeo, M. Floridi, L. AI Risk Assessment: A Scenario-Based, Proportional Methodology for the AI Act. (2024) The Digital Society Journal, 3(13). pp. 1-29.

[45] Mendez-Suarez, M. Virginia, S.M. De Prat, J.M. Do Current Regulations Prevent Unethical AI Practices? (2023) Journal of Competitiveness, 15(3). pp. 207-222.

[46] Mitchell, A.D. Let, D. Tang, L. AI Regulation and the Protection of Source Code. (2023) International Journal of Law and Information Technology, 31(4). pp. 183-301.

[47] Novelli, C. Casolari, F. Rotolo, A. Taddeo, M. Floridi, L. AI Risk Assessment: A Scenario-Based, Proportional Methodology for the AI Act. (2024) The Digital Society Journal, 3(13). pp. 1-29.

[48] Mendez-Suarez, M. Virginia, S.M. De Prat, J.M. Do Current Regulations Prevent Unethical AI Practices? (2023) Journal of Competitiveness, 15(3). pp. 207-222.

[49] Tataru, S.R. Cretu, A.C. Decoding the EU Artificial Intelligence Act: An Analysis of Key Concepts and Provisions. (2024) Journal of Public Administration Finance and Law, 31. pp. 463-474.