and the right to good administration

: The use of AI in public administration is becoming a reality, although it is still a long way from large-scale undertakings. The right to good administration, well-established in EU legal order, is equally real, however, it must be borne in mind that this right has so far been defined only in relation to traditional administration. Therefore, the purpose of the paper is to examine whether the use of AI in public administration would allow individuals to fully exercise their right to good administration. To achieve this purpose, it is reconstructed, on the basis of EU law provisions in force and the case-law of the CJEU, the meaning and scope of the right to good administration, and analysed, taking into account a definition of AI systems and planned legislative changes, whether and to what extent the reconstructed understanding of this right enables the use of AI systems in public administration. In the course of research the hypothesis that the right to good administration does not preclude the use of AI systems in public administration is verified. As the conducted analysis shows, the right to good administration as interpreted in traditional administration enables the use of AI systems in public administration, provided that the appropriate quality of these systems and the level of knowledge and skills of the parties and authorities are ensured. reinforcement learning, using a wide variety of methods including deep learning; 2. logic- and knowledge-based approaches, including knowledge representation, inductive (logic) programming, knowledge bases, inference and deductive engines, (symbolic) reasoning and expert systems; 3. statis-tical approaches, Bayesian estimation, search and optimisation methods 19 .


Introduction
The 2018 European Initiative on Artificial Intelligence (also known as the European Strategy on AI) aimed to boost the EU's technological and industrial capacity and AI uptake across the economy, both by the private and public sectors 1 . With regard to the latter sector, the European Commission assumed that AI can significantly improve public services and contribute to the objectives set out in the 2017 Tallinn Declaration on eGovernment 2 , for example, when it comes to analysing large amounts of data and helping check how single market rules are applied 3 . In the 2018 Coordinated Plan on AI, the European Commission stated that AI tools are crucial to the future work of public administrations. At the same time the Commission indicated that when AI is implemented, for example, for security and law enforcement, particular legal and ethical challenges arise, considering that public administrations are bound to act as prescribed by law, that they need to motivate their decisions and that their acts are subject to judicial review by administrative courts 4 . Awareness of these challenges did not prevent the Commission from setting an ambitious goal. The Coordinated Plan was to bring together a set of actions at EU, national and regional levels in view of making public administrations in Europe frontrunners in the use 1 Cf. Joanna Mazur, "Unia Europejska wobec rozwoju sztucznej inteligencji: proponowane strategie regulacyjne a budowanie jednolitego rynku cyfrowego, " Europejski Przegląd Sądowy, no. 9 (2020): 14. 2 Text of the Declaration is available at https://www.newsd.admin.ch/newsd/message/attachments/49838.pdf. 3 Communication from the Commission to the European Parliament, the European Council, the Council, the European Economic and Social Committee and the Committee of the Regions. Artificial Intelligence for Europe, Brussels, 25.4.2018, COM(2018 237 final (hereinafter COM(2018) 237), p. 3. 4 Communication from the Commission to the European Parliament, the European Council, the Council, the European Economic and Social Committee and the Committee of the Regions. Coordinated Plan on Artificial Intelligence, Brussels, 7.12.2018, COM(2018 795 final (hereinafter COM(2018) 795 final), p. 7. Cf. Jan Etscheid, "Artificial Intelligence in Public Administration, " 18th International Conference on Electronic Government (EGOV), Sep 2019, San Benedetto del Tronto, Italy, accessed February 24, 2022, https:// hal.inria.fr/hal-02445801/document; Adrien Bibal, Michael Lognoul, Alexandre de Streel and Benoît Frénay, "Legal requirements on explainability in machine learning, " Artificial Intelligence and Law, no. 29 (2021): 153, https://doi.org/10.1007/s10506-020-09270-4. of AI 5 . In the 2020 White Paper on AI, the European Commission stated that it is essential that public administrations rapidly begin to deploy products and services that rely on AI in their activities 6 .
In this context, mention should be made of the Recovery and Resilience Facility (RRF) that entered into force on 19 February 2021 and finance reforms and investments in Member States from the start of the pandemic in February 2020 until 31 December 2026. This financial instrument provides an unprecedented opportunity to accelerate the uptake of AI in public administration across Europe through its Flagship "Modernise" which aims at boosting investments and reforms in digitalisation of public administration 7 . However, concrete actions related to the application of AI in public administrations were undertaken earlier. For instance, the European Commission's AI-powered eTranslation portal 8 was introduced to public administration in Member States in November 2018. Two years later, 6600 civil servants across the Member States were utilising the eTranslation web portal. Some Member States have also not been idle. For example, Estonia's AI strategy has exceeded expectations and Estonia has seen wide adoption and use of AI -with over 50 AI use-cases deployed by the public sector 9 .
The use of AI in public administration is thus becoming a reality, although it is still a long way from large-scale undertakings 10 . The right to good administration, well-established in the EU legal order, is equally real, 5 Annex to the COM (2018)  however, it must be borne in mind that this right has so far been defined only in relation to traditional administration. Therefore, the purpose of this paper is to examine whether the use of AI in public administration would allow individuals to fully exercise their right to good administration. To achieve this purpose, it will be reconstructed, on the basis of EU law provisions in force and the case-law of the Court of Justice of the EU (CJEU), the meaning and scope of the right to good administration, and analysed, taking into account a definition of AI systems and planned legislative changes, whether and to what extent the reconstructed understanding of this right enables the use of AI systems in public administration. In the course of the research undertaken for this paper, the hypothesis that the right to good administration does not preclude the use of AI systems in public administration is verified. The analysis is primarily legal-dogmatic, but, to the extent necessary, also takes into account the socio-legal and legal-theoretical perspective.

EU definition of AI systems
There is still no definition of AI in EU law. According to the EU "soft" definition, included in the European Strategy on AI, AI systems mean "systems that display intelligent behaviour by analysing their environment and taking actions -with some degree of autonomy -to achieve specific goals" 11 . As further explained in this document, "AI-based systems can be purely software-based, acting in the virtual world (e.g. voice assistants, image analysis software, search engines, speech and face recognition systems) or AI can be embedded in hardware devices (e.g. advanced robots, autonomous cars, drones or Internet of Things applications)" 12 . Of course, there are more "soft" definitions, but it should be noted that non-EU definitions are omitted here. The European Commission is trying, with the participation of many entities, such as groups of experts, to develop an autonomous definition for the purposes of the EU legal order. As indicated in the literature on the subject, the efforts made so far in this field have resulted in a dominant definition in terms of the public governance of AI use 13 .
The Glossary at the end of the Ethics Guidelines for Trustworthy AI, written by the High-Level Expert Group on AI (AI HLEG) and made public on 8 April 2019, provided a definition of AI systems for the purpose of this document. According to this definition, AI systems are software (and possibly also hardware) systems designed by humans (humans design AI systems directly, but they may also use AI techniques to optimise their design) that, given a complex goal, act in the physical or digital dimension by perceiving their environment through data acquisition, interpreting the collected structured or unstructured data, reasoning on the knowledge, or processing the information, derived from this data and deciding the best action(s) to take to achieve the given goal. Such systems can either use symbolic rules or learn a numeric model, and they can also adapt their behaviour by analysing how the environment is affected by their previous actions 14 .
This definition is further elaborated on in a dedicated document prepared by the AI HLEG that accompanies the Ethics Guidelines. Interestingly, the AI HLEG started from the definition proposed by the European Commission in the European Strategy on AI. Then, they expanded this definition to clarify certain aspects of AI as a scientific discipline and as a technology 15 . According to this more advanced definition, the term 13 Anneke Zuiderwijk, Yu-Che Chen, and Fadi Salem, "Implications of the use of artificial intelligence in public governance: A systematic literature review and a research agenda, " Government Information Quarterly 38, issue 3 (July "AI system" should be understood as meaning any AI-based component, software and/or hardware wherein usually AI systems are embedded as components of larger systems, rather than stand-alone systems. An AI system is first and foremost rational. It achieves rationality by: perceiving the environment in which the system is immersed through some sensors, thus collecting and interpreting data, reasoning on what is perceived or processing the information derived from this data, deciding what the best action is, and then acting accordingly, through some actuators, thus possibly modifying the environment. In this context, the term "decision" should be considered broadly, as any act of selecting the action to take, and does not necessarily mean that AI systems are completely autonomous. A decision can also be the selection of a recommendation to be provided to a human being, who will be the final decision maker. The AI HLEG clearly pointed out that rational AI systems do not always choose the best action for their goal, thus achieving only bounded rationality, due to limitations in resources such as time or computational power. However, rational AI systems are a basic version of AI systems. They modify the environment but they do not adapt their behaviour over time to better achieve their goal. Learning rational systems are rational systems that, after taking an action, evaluate the new state of the environment (through perception) to determine how successful its action was, and then adapt its reasoning rules and decision-making methods 16 .
For the purpose of the proposed Regulation laying down harmonised rules on AI (in the 2021 version) 17 , AI system means software that is developed with one or more of the techniques and approaches listed in Annex I to this Regulation and can, for a given set of human-defined objectives, generate outputs such as content, predictions, recommendations, or decisions influencing the environments they interact with 18 . In accordance with the aforementioned Annex the relevant AI techniques and approaches are: 1. machine learning approaches, including supervised, unsupervised and 16 A Definition of AI, p. 1-3. 17 Proposal for a Regulation of the European Parliament and the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) and amending certain Union legislative acts, Brussels, 21.4.2021, COM(2021) 206 final. 18 Article 3, point 1 of the proposed Regulation. reinforcement learning, using a wide variety of methods including deep learning; 2. logic-and knowledge-based approaches, including knowledge representation, inductive (logic) programming, knowledge bases, inference and deductive engines, (symbolic) reasoning and expert systems; 3. statistical approaches, Bayesian estimation, search and optimisation methods 19 . Although Annex I explains a lot, contrary to recital 6 of the Regulation, the proposed legal definition cannot be considered as precise.
If all three elements, i.e. Article 3, point 1 of the Regulation, recital 6 of the Regulation and Annex I to the Regulation, are combined, the resulting definition is similar to the one proposed by the AI HLEG. However, from the point of view of the need for clarity of the legal definition, such a procedure does not seem to be a recommended solution. It should also be emphasised that experts' explanations similar to an academic lecture have a different function than a legal act. The definition contained in the draft Regulation under discussion has been considered too broad by many commentators. As they have indicated, Article 3, point 1 in conjunction with Annex I covers almost every computer program. Such a broad approach may lead to legal uncertainty for developers, operators, and users of AI systems, especially when it comes to high-risk AI systems. Many associate the term "artificial intelligence" primarily with machine learning, and not with simple automation processes in which pre-programmed rules are executed according to logic-based reasoning. The mandatory requirements envisaged for high-risk AI systems are based on the observation that a number of fundamental rights are adversely affected, in particular, by the special characteristics of machine learning, such as opacity, complexity, dependency on data, autonomous behaviour. Since these characteristics are either not or only partly present in simple (logic based) algorithms, the broad definition of AI systems can lead to an overregulation. However, a wide definition may be justified in light of the prohibited AI practices delineated in Article 5 to offset the threats posed by different kinds of software to the fundamental rights of individuals because there is little difference 19 Annexes to the Proposal for a Regulation of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (artificial intelligence act) and amending certain union legislative acts, Brussels, 21.4.2021, COM(2021) 206 final, Annexes 1 to 9, p. 1.
to the rights of affected citizens whether the banned practices (subliminal manipulation, exploitation of vulnerabilities, social scoring, or remote biometric identification) are enabled by machine learning or logic-based reasoning 20 . In conclusion, from the perspective of defenders of fundamental rights, the proposed legal definition of AI systems is not a major problem.

Public administration as a high-risk area in the context of AI systems
In the revised Coordinated Plan on AI, the Commission indicated that the use of AI creates risks that need to be addressed. Certain characteristics of AI, such as the opacity of many algorithms that makes investigating causal relationships difficult 21 , pose specific and potentially high risks to fundamental rights. For example, it is often not possible to determine why an AI system has arrived at a specific result. As a consequence, it may become difficult to assess and prove whether someone has been unfairly disadvantaged by the use of AI systems, for example in an application for a public benefit scheme 22 . Also, poor training and design of AI systems can result in significant errors that may undermine fundamental rights. The use of AI systems may leave affected people with significant difficulties to correct erroneous decisions. Therefore, AI-enabled robots and intelligent 20 Martin Ebers, Veronica R. S. Hoch, Frank Rosenkranz, Hannah Ruschemeier, and Björn Steinrötter, "The European Commission's Proposal for an Artificial Intelligence Act-A Critical Assessment by Members of the Robotics and AI Law Society (RAILS), " Multidisciplinary Scientific Journal, 4 (2021): 590, https://doi.org/ 10.3390/j4040043. Cf. "ZVEI Comments on the EU Commission's Proposal for a Regulation laying down harmonised Rules on Artificial Intelligence ("AI Act"), " ZVEI -Zentralverband Elektrotechnik-und Elektronikindustrie e.V. Abteilung Innovationspolitik, accessed February 24, 2022, https://www. zvei.org/fileadmin/user_upload/Presse_und_Medien/Publikationen/2021/September/EU-KI-Gesetz/ZVEI-Comments-on-AI-Proposal_2021-08.pdf; "DIGITALEUROPE's initial findings on the proposed AI Act, " DIGITALEUROPE, accessed February 24, 2022, https:// www.digitaleurope.org/resources/digitaleuropes-initial-findings-on-the-proposed-ai-act. systems must be engineered and designed to meet the same high standards of protection of fundamental rights provided for by EU law as for traditional technologies. However, existing EU legislation is unable to provide this level of protection 23 . The Commission's proposal for a regulatory framework on AI was intended to be "a key juncture in the journey towards protecting (...) fundamental rights and hence ensuring trust in the development and uptake of AI" 24 . Work on it, however, is delayed due to disagreement on many issues that are the subject of it. As the European Commission itself pointed out, the stakes are high for the EU -spearheading the development of new ambitious global norms 25 .
The Commission's proposal contains a set of harmonised rules applicable to the design, development and use of certain high-risk AI systems. The purpose of these rules is to enhance transparency and minimise risks to fundamental rights before AI systems can be used in the EU. In line with a risk-based regulatory approach, the proposed legal framework is designed to intervene only where this is strictly needed 26 . In this context, the 'high-risk' AI use means that the risks posed by the AI systems are particularly high. Whether an AI system is classified as high-risk depends on its intended purpose of the system and on the severity of the possible harm and the probability of its occurrence. As can be seen from recital 28 of the proposed Regulation, the extent of the adverse impact caused by AI systems on the rights protected by the Charter of Fundamental Rights is of particular relevance when classifying an AI system as high-risk. According to Article 6(2) of the proposed Regulation, AI systems referred to in Annex III shall be considered high-risk. As Annex III states, high-risk AI systems are the AI systems listed in one of the areas, including: 1. access to and enjoyment of public services and benefits; 2. migration, asylum and border control management. In both of these areas there are "classic" administrative proceedings and accordingly the annex indicates "AI systems 23 COM (2021)  intended to be used by public authorities or on behalf of public authorities to evaluate the eligibility of natural persons for public assistance benefits and services, as well as to grant, reduce, revoke, or reclaim such benefits and services" 27 . In the second area, the Commission listed four AI systems, including "AI systems intended to assist competent public authorities in the examination of applications for asylum, visa and residence permits and associated complaints with regard to the eligibility of natural persons applying for a status" 28 .
When analysing the content of Annex III, a question should be asked why it only refers to "public services and benefits", by which -it seemsthe authors of this act understand the rights of the addressee of the decision, not the obligations. Meanwhile, as Mateusz Pszczyński points out, AI systems can also be used in administrative proceedings aimed at imposing an obligation or an additional charge. An example is the decision setting real estate tax for natural persons. Every year between January and March, thousands of such tax decisions are issued by the executive bodies in all Polish municipalities. They use computer programs which, on the basis of the taxpayers' records, including data on the subjects of taxation, rates and tax base, determine the amount of real estate tax. When printed, the decisions are signed by the office holder of the body or a person authorised by him/her. The correctness is verified at the initial stage when the updated rates are checked. There is no time nor technical possibility for each decision to be checked before it is signed by an authorised entity. The same applies if an additional charge is imposed as a result of a parking fee not being paid. Therefore, in Poland already today a computer program basically replaces human beings and makes a decision on their behalf, while the signature is a legal fiction, which, due to the formalism of tax or other proceedings, must be observed 29 . It is similar in other countries, such as Denmark and Finland 30 . Nowadays, it is easy to imagine the use of AI sys- tems to issue decisions on imposing the obligation to provide personal or material benefits for defence in the event of mobilisation and during war or for natural disaster prevention and recovery, if only because of the haste necessary in such situations. Even if not all potential obligations imposed by an "AI official" fall within the scope of application of EU law, certainly some of them, such as e.g. obligations relating to value added tax. In any case, as noted by Błażej Kuźniacki, tax law has a very large potential to use AI compared to other branches of law, due to the high complexity and technical nature of tax standards and their detachment (to a large extent) from everyday human life 31 . Obviously, AI systems intended to be used to assess the eligibility of individuals for public assistance benefits and services and to grant, reduce, revoke or reclaim such benefits and services may pose a high risk to fundamental rights, however, inferring a minore ad maius (from the less to the greater), AI systems intended to be used to establish or assert natural persons' obligations can pose an even higher risk. It is therefore a misunderstanding to leave this group of administrative matters out of sight, and the short and closed catalogue of Annex III should be considered a mistake already from the perspective of technological development in 2021.

Right to good administration at the EU level
The right to good administration is guaranteed in Article 41 of the Charter of Fundamental Rights of the EU 32 . According to Article 41(1) of the Charter, every person has the right to have his or her affairs handled impartially, fairly and within a reasonable time by the institutions, bodies, offices and agencies of the Union. As provided by Article 41(2) of Charter, this right includes: 1. the right of every person to be heard, before any individual measure which would affect him or her adversely is taken; 2. the right of every person to have access to his or her file, while respecting the legitimate interests of confidentiality and of professional and business secrecy; 3. the obligation of the administration to give reasons for its decisions. The word 'includes' in that latter provision shows that the right to good administration is not confined to the three abovementioned guarantees 33 . Article 41(2) of the Charter lists a set of rights to be observed by the Union's administration, including the rights of defence, which include the right to be heard and the right to have access to the file 34 . It should be emphasised that the requirements pertaining to the right to good administration, which reflects the general principle of EU law, and in particular the right of every person to have his or her case handled impartially within a reasonable time, are applicable in procedures where Member States apply EU law 35 .
The right to good administration (also the general principle of good administration) requires administrative authorities, when carrying out their inspection duties, to conduct a diligent and impartial examination of all the relevant matters so that they can be sure that, when they adopt a decision, they have at their disposal the most complete and reliable information possible for that purpose. Consequently, where a party makes errors in her/his application, and neither the party nor the authority concerned subsequently identifies those errors, that authority may not be held responsible for doing so, unless the errors are easily noticeable, in which case the authority should be able to detect them under his obligation of verification under the principle of good administration 36 . That requirement of impartiality encompasses, on the one hand, subjective impartiality, in so far as no member of the institution concerned who is responsible for the matter may show bias or personal prejudice, and, on the other hand, objective impartiality, in so far as there must be sufficient guarantees to exclude any legitimate doubt as to bias on the part of the institution concerned 37 . Subjective impartiality is presumed in the absence of evidence to the contrary 38 . Whether the time taken for a procedure is reasonable must be assessed in relation to the individual circumstances of each case, and in particular its context, the conduct of the parties during the procedure, what is at stake for the various undertakings concerned and its complexity, and also, where relevant, to information or justification which the Commission may provide concerning the measures of investigation carried out during the administrative procedure 39 .
The right to be heard guarantees every person the opportunity to make known his views effectively during an administrative procedure and before the adoption of any decision liable to affect his interests adversely 40 . The right of access to a personal file means that the institution in question must give to the person concerned the opportunity to examine all the documents in the investigation file that might be relevant for his defence. These include, in particular, both incriminating and exculpatory evidence, save for internal documents of the institution in question and other confidential information 41 . The statement of the reasons for the decision is particularly important in so far as it allows persons concerned to decide in full knowledge of the circumstances whether it is worthwhile to bring an action against the decision and the court with jurisdiction to review it, and it is therefore a requirement for ensuring that the judicial review guaranteed by Article 47 of the Charter is effective. The statement of reasons must be adapted to the nature of the legal act at issue and to the context in which it was adopted. In that regard, it is not necessary for the reasoning to go into all the relevant facts and points of law, since the question whether the statement of reasons is sufficient must be assessed with regard not only to its wording but also to its context and to all the legal rules governing the matter in question and, in particular, in the light of the interest which the addressees of the act may have in obtaining explanations. Consequently, the reasons given for an act adversely affecting a person are sufficient if that act was adopted in a context which was known to that person and which enables her/him to understand the scope of the act concerning her/him 42 .

Exercising the right to good administration in AI-based proceedings
As indicated in the literature on the subject, it is uncontroversial that a clear definition of AI is necessary to create an effective legal framework 43 . Such a definition is definitely missing in the Commission's proposal. The (too) broad definition of AI systems means that an "AI official" already frequently deals with administrative cases, and she/he will do so more often, both in terms of the number of cases (mass cases) and the categories of cases (cases belonging to various areas of public administration activity). Therefore, it is urgent and increasingly important to answer the question whether the use of AI in public administration allows individuals to fully exercise their right to good administration. From the point of view of the need to protect this right, we can put together traditional administrative procedures, administrative procedures using Bayesian estimation and administrative procedures using deep machine learning. However, the scale of potential threats or real detriment to the exercise of the right to good administration will be significantly different. The European Commission has admitted that the use of AI can affect the right to good administration 44 . "Influence" is a broad concept and not necessarily a negative one. AI systems in administrative procedures can have both threats (or damages) and benefits to the right to good administration. In the Charter of Fundamental Rights in the context of Artificial Intelligence and Digital Change 45 , the representatives of the 26 Member States 46 meeting in the Council recognised the potential of digital technologies, including AI applications, to improve the protection of the right to good administration. Unfortunately, they did not explain what they meant.
It is believed that AI systems are able to help decision-makers to reach better decisions, to enhance officials' analytic abilities and to intensify creativity adequate to modern challenges 47 . However, this is one perspective, focusing rather on good (comfortable) work in administration, albeit with a high-quality decision for the addressee as its result. On the other hand, the addressee of the decision may face problems of a rudimentary nature. By way of example, the Haut Conseil du Travail, an advisory body to the French Ministry for Social Affairs, has estimated that 1 in 5 people in France encounter difficulties trying to complete administrative procedures online, and has warned that digitalisation can jeopardise the principle of equal access to public services if alternative means of access are not maintained 48 . It is therefore legitimate to ask whether the right to access the case file, which is an element of the right to good administration, if the case is handled by an "AI official", will not be illusory for an average participant in the proceedings. The group of authors is right when they write that with a lack of appropriate monitoring, it is challenging for a variety of stakeholders to identify the risk of harmful repercussions of AI systems after 45 Note from Presidency on 12 October 2020 to Delegations. Presidency conclusions -the Charter of Fundamental Rights in the context of Artificial Intelligence and Digital Change, item 2 (text of the Charter is available at https://www.consilium.europa.eu/ media/46496/st11481-en20.pdf). 46 One Member State (Poland) objected to the use of the term "gender equality". 47 Mohammad I. Merhi, "A Process Model of Artificial Intelligence Implementation Leading to Proper Decision Making, " in Responsible AI and Analytics for an Ethical and Inclusive Digitized Society: 20th IFIP WG 6.11 Conference on e-Business, e-Services and e-Society, I3E 2021, Galway, Ireland, September 1-3, 2021, Proceedings, ed. Denis Dennehy, Anastasia Griva, Nancy Pouloudi, Yogesh K. Dwivedi, Ilias Pappas, Matti Mäntymäki (Cham: Springer, 2021 COM(2021) 819. deployment 49 . In this context, it should also be noted that the exercise of the right to good administration is not possible without digital sovereignty, which, as recalled by the Berlin Declaration, is of key importance for ensuring the ability of citizens and public administrations to make decisions and act in a self-determined manner in the digital world 50 .
The analysis of the normative content of the EU's right to good administration suggests one simple conclusion -AI systems should facilitate the consideration of the case within a reasonable time. Whether the requirements of impartiality (in particular, subjective impartiality) and fairness are met, will directly depend on the quality of the AI system -the quantity and reliability of data entered into it and the appropriate training of the AI algorithms. In turn, the exercise of the right to be heard (to present one's views, but also to provide evidence in her/his favour) and the right to access the files (to read them, but also to demand changes or supplements to their content in one's favour) will be conditioned by the knowledge and skills of the proceedings' party. Therefore, sufficient information and effective training will be a key factor in this regard. The implementation of the obligation to give reasons for the decision will be the result of the quality of the AI system and the knowledge and skills of the official, assuming that all stages will not be carried out by an "AI official". Since, as is clear from the case-law, only the key elements of the right to good administration are mentioned, not all of them, the possibility of exercising the right to good administration should always be analysed in the light of the circumstances of the case. Meeting the detailed requirements related to the interpretation of the CJEU, such as diligence and verification of documents for the presence of noticeable errors, should be facilitated by the use of AI. On the other hand, the usefulness of the justification of the decision for the purposes of appealing and conducting judicial review may be increased by a good quality AI system, but will also depend on the knowledge and skills of the (real) official who uses it. In this context, close attention should be paid to the CJEU's liberal approach to justifying the decision. In the light of its guidelines, a rationale for the decision prepared with the participation of the "AI official" or by the "AI official" will not be defective or insufficient in principle. The reconstructed understanding of the right to good administration enables the use of AI systems in public administration, provided that the appropriate quality of these systems and the level of knowledge and skills of the parties and authorities are ensured.
European Declaration on Digital Rights and Principles for the Digital Decade, solemnly proclaimed by the European Parliament, the Council and the European Commission on 26 January 2022, addresses the issue of interactions with algorithms and AI systems only in the context of freedom of choice, which is not entirely in line with the specificity of how public administrations handle matters. As regards the form and manner of resolving the case, the parties to the administrative procedure do not enjoy freedom of choice. According to the Declaration, everyone should be empowered to benefit from the advantages of AI by making their own, informed choices in the digital environment, while being protected against risks and harm to one's fundamental rights 51 . Taken literally, this should apply to all stages of the administrative procedure, and therefore not only to the submission of the application, but also to the decision. This should also mean maintaining the traditional form of dealing with cases as an alternative to cases handled by an "AI official". It is important, however, that in this Declaration, the EU institutions have committed to: 1. ensuring transparency about the use of algorithms and artificial intelligence, and that people are empowered and informed when interacting with them; 2. ensuring that algorithmic systems are based on suitable datasets to avoid unlawful discrimination and enable human supervision of outcomes affecting people; 3. ensuring that technologies, such as algorithms and artificial intelligence are not used to 51 European Declaration on Digital Rights and Principles for the Digital Decade, Brussels, 26.01.2022, COM(2022) 28 final (hereinafter COM(2022) 28), p. 4 pre-determine people's choices, for example regarding health, education, employment, and their private life; 4. providing for safeguards to ensure that artificial intelligence and digital systems are safe and used in full respect of people's fundamental rights 52 . It should be assumed that the fulfilment of three of these commitments (the first, the second and the fourth), which are relevant to administrative proceedings, would be a necessary and sufficient condition for the effective exercise of the right to good administration at the EU level.

Conclusion
As the above analysis has shown, the right to good administration as interpreted in traditional administration does not exclude the use of AI systems in public administration, e.g. when issuing tax decisions, decisions in the field of social security or social assistance, or even decisions regarding construction processes or environmental protection, but it requires the fulfilment of a number of conditions. Therefore, all three EU institutions that have adopted the European Declaration on Digital Rights and the Principles of the Digital Decade, and in particular the EU legislator, i.e. the European Parliament and the Council, should ensure that AI Regulation is in line with the commitments made. In the revised Coordinated Plan on AI, the European Commission announced that it will continue its efforts to ensure that AI developed and put on the market in the EU is human-centric, sustainable, secure, inclusive, accessible and trustworthy. The Commission is right when it writes that a regulatory framework to ensure trust in AI systems is essential to achieve these goals 53 . However, it is not just the regulatory framework for AI that is at stake. It is also necessary to supplement and clarify the rights of the parties and the obligations of the authorities in administrative proceedings, so that all subjects in these proceedings have the necessary information and skills.