1 Introduction

Brazil has advanced its digital agenda since the 1990s. The first effort in this direction was the Information Technology Acts—Law 8248 [1] and 8387 [2], from 1991. Both acts instrumentalized digital development through tax incentives for companies that implemented research and development in hardware and automation. In addition, this legislation instrumentalized the creation of networks between companies and universities for digital development, with a particular interest in developing hardware and beginning automation processes for industry.

Before that, Brazil already had an extensive digital infrastructure and a long history in data collection, storage, processing, and sharing. In 1964, the Data Processing Service (SERPRO) was created, a public company of the Federal Government that collects and processes various data, ranging from census data, economic and commercial transaction data, social data, and data from different public policy domains and administrative registers. In addition to SERPRO, since 1973, Brazil has had Dataprev, another public company that collects, stores, and processes data from Brazil's entire social protection system.

The Brazilian government formulates policies to promote long-term digital development based on this data infrastructure and tax incentives for developers. In addition, the Ministry of Science, Technology, Innovations, and Communications provides funding for network research and development, such as, for example, the National Network for Education and Research (RNP), constituting collaborative networks and access to computational and data infrastructure. These policies have achieved outcomes for developing hardware and software, with a particular interest in industrial automation processes, deployment of digital technologies in governments and markets, and encouraging and accelerating technological development. In addition to these policies, Brazil accelerated the deployment of the Internet in the 1990s, ranging from the privatization of public communication companies and governance rules for global network connection.

In the mid-2010s, a disruptive legislation in this regard was the Civil Rights Framework for the Internet—Law 12965 of 2014 [3]—which introduced innovative mechanisms for a multistakeholder governance and which became a regulatory model, since it allowed a balance between innovation and secure development of this digital world. The construction of the Civil Rights Framework for the Internet followed a whole mode of participation by society and stakeholders, based on holding public hearings and public consultations to listen to the population before and after the publication of the act.

Considering this long trajectory of policies for digital development, Brazil has changed different elements of public policies to package these policies around the idea of digital transformation. Digital transformation is a broad idea that emerged from the 2010s and that guides the adoption of disruptive digital technologies to change business models in markets and industries, as well as the provision of public services and public policies in governments [4]. Brazil revised the Information Technology Acts, removing bureaucratic barriers that facilitate the acceleration of digital transformation, with a particular interest in research and development on artificial intelligence. The New Information Technology Acts—Law 13674 of 2018 [5] and Law 13969 of 2019 [6] —expanded the scope of tax incentives, including the development of artificial intelligence, autonomous systems, the internet of things, blockchain, and other data-based technologies, going beyond an idea of automation.

In addition to implementing tax incentives driven by the idea of digital transformation, the Brazilian government began to redesign digital development policy based on benchmarks formulated in the international arena, especially with the guidelines of the Organization for Economic Co-operation and Development (OECD) [7, 8]. The OECD disseminated comprehensive digital transformation policy with the Brazilian government in order to create incentives for digital development, adoption of artificial intelligence and platformization of governments, guides for data governance, cooperation with the private sector, and models of technological governance involving stakeholders from the private sector, academia, and governments.

Another interesting recent policy promoted by the Brazilian government to stimulate the safe development of new technologies is available in the Legal Framework for Startups. This legal instrument provides regulatory sandboxes as a support mechanism and reliable testing of new technologies. Regulatory sandboxes for testing artificial intelligence-based solutions have been applied by the Brazilian Central Bank, especially for fintechs (financial technology, startups that develops different solutions to the financial sector) and insurance regulation.

On artificial intelligence, there are different effects regarding the one development. Although this technology has become pervasive in different sectors of the Brazilian economy and society, the country ranks 19th in the AI Vibrancy Ranking, organized by Stanford University.Footnote 1 AI policies in Brazil are motivated by increasing economic competitiveness, outlining strategic actions that expand the presence of AI in different sectors of society. Brazil has advanced policy agendas for artificial intelligence going beyond tax incentives. The Brazilian government has created centers of excellence for AI development, public funds to accelerate research at universities, collaborative networks, development ecosystems, and change institutional frameworks for data access and sharing, incorporating privacy and data protection concerns [9].

These policy initiativesFootnote 2 have produced results that gradually make artificial intelligence pervasive in the Brazilian society. For example, cities like Fortaleza, São Paulo, Rio de Janeiro, Florianópolis, and SalvadorFootnote 3 have gradually deployed systems based on facial recognition technology with direct applications in security policy. Another example is how the Brazilian Central Bank has accelerated the development of fintechs, creating regulatory mechanisms that encourage technologies applied to the financial and insurance market. Another example is how Brazil has encouraged national e-commerce, with platforms based on artificial intelligence, to produce personalization and optimization of commercial relationships. Governments have adopted machine learning algorithms to optimize decision-making and task accomplishment, which change the entire governmental process of policy formulation and implementation through government platformization [10]. The Judiciary has been using systems based on machine learning algorithms to recommend decisions and optimize judicial procedures [11]. In Brazilian society, artificial intelligence is gradually pervasive in governments, the economy, industry, and markets, changing various aspects of society.

However, artificial intelligence policy in Brazil tend to reproduce a more corporatist governance style, in which some organizations benefit from participating in the governance of technology development.Footnote 4 This corporatist governance, that is, the set of processes, policies, laws, procedures, regulations and institutions that regulate the way an organization is run, reproduces a long trajectory of modernization and industrialization of Brazilian society since the 1930s. In the case of policies related to the technology domain, they reproduce a long pathway of modernization policies with the business organizations based on a corporatist model [12, 13]. This corporatist dynamic tends to reinforce situations of path dependence and increase barriers to the entry of new actors in this artificial intelligence development market [14]. This path dependence produces increasing returns and situations that hinder new actors in the digital development scene. Additionally, policies for artificial intelligence present incoherent and inconsistent elements, producing sparse results for digital transformation [9]. Although Brazilian government implements different initiatives and policies to promote digital development, these policies need to be more coherent and consistent concerning concrete outcomes that favor the framing of artificial intelligence for the Brazilian social reality.

Considering this institutional context and the policy trajectory for digital development, in 2018, the Brazilian government opened a public consultation on the achievement of a Brazilian National Strategy for Artificial Intelligence (EBIA). This Brazilian strategy aims to produce a more robust institutional framework that integrates different policy initiatives coherently and consistently with digital development objectives. Moreover, the Brazilian strategy is associated with developing an institutional framework that establishes the fundamental lines of AI regulation, defining ethical frameworks, and proposing guiding principles for technological development.

Throughout the article, we seek to demonstrate how the EBIA fails because it proposes fragmented policies that are characterized by being ineffective, but popular (while the unpopular ones encounter resistance); that eventually serve to put out fires, but not to avoid crises and provide development and because it does not present an important factor in this type of document, that is, the evaluation of previous public policies [15].

In other words, the EBIA intends to produce policy change based on integrating different initiatives and the definition of institutional frameworks. However, although the objective of this Brazilian strategy recognizes a long trajectory of digital development and that policies in this area are fragmented, the result of the strategy fails to produce engagement among the stakeholders, identify the problems and propose solutions. The EBIA reproduce situations that maintain the status quo ante. That is, the Brazilian strategy fails to produce changes, representing a missed opportunity to design policies that are more coherent and consistent with the needs of digital transformation.

This article aims to analyze why the Brazilian strategy for artificial intelligence fails to produce policy integration and for the development of artificial intelligence. In the first section of the article, we deal with the concept of strategy that we adopted, considering that national strategies comprise policy design dynamics. In the second section, we examine the context, actors, and governance styles from which the EBIA emerges. In the third section, the article deconstructs the components of the EBIA. Finally, in the fourth section, the article explains why EBIA fails to build strategic action for AI development. The article concludes by highlighting lessons learned and challenges involved in national strategies for artificial intelligence. This article contributes to the analysis of national strategies for artificial intelligence, composing a topic of interest in AI policy. It also contributes to the definition of national strategies, the dilemmas involved in these strategies, and the challenges related to the design dynamics concerning AI policy.

2 Background—policy design and national strategies for AI

National strategies compose policy documents that establish principles and guide policymakers to design policies based on values, visions, and short, medium, and long-term objectives. National strategies thus create a perspective for policy problems and represent an institutional building process focused on engaging policymakers and actors around proposed solutions. In this sense, a national strategy deals with an institutional design that establishes policies to solve a problem. In other words, national strategies are problem-solve perspectives.

Institutional theory is useful for analyzing strategies, considering the types of institutions. According to Elinor Ostrom [16], institutions comprise norms, rules, and shared strategies that coordinate and guide the actors’ behaviors in action situations. Institutions comprise statements that can be deconstructed grammatically. Norms comprise principles for action, implying deontic operators that establish rules for behavior. Rules, according to Ostrom, express norm-based collective choices. Finally, the shared strategies, the object of interest in this article, comprise shared ideas guided by norms and rules for coordinating collective action without necessarily implying sanctions. That is, strategies are shared ideas that guide collective action, engaging actors to carry out and instrumentalize a collective enterprise [16].

Public policies are institutional regimes that organize the rules of the game for society to produce public goods and solve collective action dilemmas [17, 18]. This definition by Ostrom points to some central issues. First, strategies are shared among actors. Strategies imply collective action, according to the engagement of actors in designing solutions. Second, institutional regimes govern policy situations, defining demands, places, and people. Individuals and groups deliberately craft these institutional regimes to make the actors' interactions more predictable and reduce uncertainties and risks [19]. Third, the construction of solutions to collective problems demands design dynamics oriented towards crafting solutions, mobilizing resources, defining objectives, and relating instruments to change the behavior of the policy’s target population.

National strategies are, therefore, policy documents that aim to coordinate collective action to achieve specific policy objectives. Strategies represent an intention [20], being designed to instrumentalize means to achieve a goal. As a collective action enterprise, national strategies represent an intention to engage policymakers and diverse actors to carry out a collective objective. Following this concept, national strategies represent an intention and guide the policy design to solve policy problems through collective action. National strategies, therefore, link actors to shared objectives, defining instruments to achieve the purpose and deliver effective outcomes to promote social change.

National strategies for artificial intelligence thus identify the factors causing the problems, enumerate the public objectives to be achieved, the instruments through which these objectives will be achieved, monitoring mechanisms, governance perspectives and stakeholder involvement [21, 22]. National strategies are problem-solving and enumerate public challenges in a given context and time to coordinate and guide collective action. National strategies are designed to realize this intention and therefore incorporate policy design dynamics.

Policy design reflects a dynamic combination of goals and instruments that develop over time [23]. The interests and concerns of policy design researchers and those who work in neo-institutionalism theory sometimes converge because both are committed to explaining policy changes, outcomes, and implications [24]. Policy design, therefore, implies strategic action in which policymakers struggle to build coherent designs between problems and proposed solutions, that the instruments are consistent for the policy to achieve its objectives, and that the instrumentation is congruent with the intended objectives [25].

Public policies are designed from a mix of objectives and instruments used during the design dynamics [23], guiding all efforts so that the policy is effective and socially and politically accepted. Policy design is not a rationalist activity in which designers connect instruments and objectives to produce outcomes. Policy design is a dynamic activity in which designers constantly choose and calibrate instruments responding satisfactorily to dynamics that arise in design spaces [25, 26]. Policy design is a political activity [27] and responds to action situations related to the conditions of the political regime and the governance modes and styles from which policy is steered. That is, policy design occurs in situations of limited rationality, building policy situations that are satisfactory for the actors and not necessarily the most rational. The multiple actors involved in policy design dynamics select and choose the instruments necessary to achieve the formulated goals. The actors choose the instruments in the context of governance modes that guide decision-making and implementation. Policy design is not a technocratic activity but essentially a political one [27, 28].

Scholarship on national strategies for artificial intelligence points to different analytical perspectives. National strategies for artificial intelligence analyze policy initiatives grouped into different policy areas. For example, Van Roy [29] analyzes European strategy in five major groups such as human capital, from the lab to the market, networking, regulation, and infrastructure. There are also analyzes as Radu [30] that aim at a global analysis of these policy documents. It is also possible to analyze, in these national strategies, the sociotechnical imaginaries as presented in the discourse of four national AI strategies [31], group strategies together with AI ethics guidelines produced by the private sector [32], focus on the comparison between two countries like China and India [33] or concentrate on an in-depth analysis of the national AI strategy of a single country [34,35,36]. Van Berkel et al. [37] analyze 25 national AI strategies of countries from a quantitative approach. In all these approaches, the analysis of policy documents aimed to identify the components and elements that define objectives, instruments, ideas, and processes of collective action.

National strategies reflect the proposition of policy change so that policies can be formulated and implemented to solve public problems. Strategies involve not only the definition of problems and objectives but also the design of public policies through integrated strategic actions aimed at proposing solutions.

In this article we make an in-depth analysis of the Brazilian National Strategy for Artificial Intelligence (EBIA), considering the strategic actions and how they list objectives and instruments. We will analyze the policy design dynamics and how the strategy operates to engage stakeholders in the collective enterprise of digital development through artificial intelligence in Brazil.

3 Context, actors and governance styles to design EBIA

The first task for analyzing the EBIA is to understand the actors who intentionally shaped the goals and perspective for artificial intelligence in Brazil. Considering that strategies must produce shared objectives and perspectives, the broad involvement of stakeholders is fundamental. Furthermore, artificial intelligence is a general-purpose technology that can be implemented in different policy domains and sectors of public administration. Thus, creating horizontal coordination mechanisms between the actors is essential [18].

In the context in which the strategy is produced, the design of the EBIA found a design space marked by vague ideas about artificial intelligence and a populist political context. Add to this political context a bureaucratic management of the Ministry of Science, Technology, Innovations, and Communications marked by a governance style characterized by institutional weakness and aversion to interaction with society. The political dimension and the availability of the bureaucracy to interact with society shape governance styles and how policies will be steered [38].

The first point we must highlight concerns the context. The functioning of a populist political regime led by right-wing President Jair Bolsonaro, associated with a bureaucracy that is resistant to interacting with society, creates a context of institutional weakness and an incapacity to design public policies. A similar event occurs with other policies, such as, for example, the nuclear energy policy [38]. In the case of policies for digital development, this context of institutional weakness emerges from a governance style based on a populist political regime associated with a bureaucracy that is refractory to any interaction with society, reproducing technocratic patterns not supported by stakeholder involvement.

In this context, the Brazilian National Strategy for AI (EBIA) was published in 2021, but its design process took place during the years 2019 to 2020. The Brazilian Government was based on three ways of understanding the issues that involve the theme of AI technology, both in its ethical aspect, as well as its technical aspect. These three ways to design the EBIA are the consultancy provided by UNESCO, the production of benchmarks on international experiences, and the public consultation initiated by the Ministry of Science, Technology, Innovations, and Communications.

From 2019 to 2020, therefore, a specialized consultancy hired by the Ministry of Science, Technology, Innovations and Communications was carried out on the subject of AI through the international technical cooperation project with UNESCO. It should be noted, however, that the result of these three stages prior to the official publication of the policy document was not brought or shared publicly.

UNESCO played a crucial role in helping the Ministry of Science, Technology, Innovations, and Communications to formulate the first set of problems and thus produce a benchmark of international initiatives to solve these problems. The perspective adopted is that the strategy can be designed based on the dissemination of international experiences without involving stakeholders in the construction of strategic actions. The first element is possible to be identified. EBIA's design dynamics were guided by a technocratic perspective, with little stakeholder involvement in identifying problems and solutions. The initial corollary is that the Ministry of Science, Technology, Innovations, and Communications considered design-oriented dynamics as a process of creating collective action from technical work conducted in association with UNESCO, not as political work aimed at creating actors’ engagement.

The methodology for building the EBIA was based on hiring a consultancy from UNESCO, producing an international benchmark, and then collecting suggestions and perspectives from different stakeholders. It is important to note that public consultations in Brazil are carried out on a digital participation platform—Participa.br—which facilitates online participation and society's influence on policies [39]. Although public consultation was implemented, the suggestions and notes made by the industry or civil society organizations were not systematized. They were not incorporated or used as critical mechanisms in the EBIA document.

The actors in the design dynamic of the EBIA, therefore, were the Ministry of Science, Technology, Innovations, and Communications, associated with UNESCO consultancy. In identifying the causes of the problems and the construction of the objectives, actors of the industry, the market, other governmental agencies, the academy, and civil society organizations were excluded from the process. The definition of objectives and strategic actions was formulated technocratically, without any form of interaction with society. Moreover, the political context disfavored the construction of a robust policy for digital development. The emergence of the EBIA occurred in a populist government, which distrusts coalitions, produces democratic regression, and encourages the policy dismantling [40].

The EBIA was published in 2021 after a long consultation process, representing a formal strategy on how Brazil will formulate policies and actions for digital development with artificial intelligence, but lacking a background of stakeholder participation and engagement. The EBIA was produced as a technical mechanism for state strategic actions, but with huge collective action problems. Without thinking of effective mechanisms to produce a collective enterprise, the EBIA fails, being an ungrounded perspective on AI policy and development. Despite the long process of public consultation and contracting of a collaborating institution, the result of these documents was not presented to society, which makes it difficult to understand what was or was not incorporated into the final policy document.

This topic will seek to present the EBIA, to situate the reader to understand why it fails as a strategic document for the development of public policies. Despite having previously consulted international organizations, whoever reads the text notes that the formulation of the EBIA completely ignored a fundamental aspect of any strategy, namely, the proposition of an agenda [41]. The EBIA is designed disregarding what the international actors heard consider as relevant for decision makers at the most diverse levels of administration, as well as what society and the multiple actors involved in the development of AI have pointed out.

4 EBIA’s vision, components, and objectives

In this section, we deconstruct the vision, objectives and strategic actions listed in the EBIA. The first element that we deconstruct in the Brazilian strategy is the vision it embodies to resolve the identified problems. A vision creates the initial elements of a national strategy, constituting what kind of approach will be taken to policy problems. The Brazilian National Strategy for Artificial Intelligence (EBIA) sets out the following vision:

This Strategy assumes the role of guiding the actions of the Brazilian State in favor of the development of actions, in its various aspects, that stimulate research, innovation and development of solutions in artificial intelligence, as well as its conscious, ethical use and in favor of a better future [42].

This view of strategy starts from two implicit problems. First, the EBIA needs to integrate policies for the development of AI, building an integrative perspective of different policies already implemented. This objective of integrating into a single set of strategic actions aims to generate coherence between different initiatives already existing in Brazil. Among the initiatives, Brazil implements public funds to finance research and development, tax incentives given by the New Information Technology Acts, creating centers of excellence, training and transitioning work and labor, and digital government initiatives. Integrating these different policies comprises the first primary objective of the Brazilian perspective. The second major objective is the need to create an ethical and regulatory framework that ensures the development of artificial intelligence. The problems identified in the EBIA reproduce a common framework of AI policy concerning economic growth and societal challenges [43].

The EBIA has six primary objectives arising from the identification of the problem as the poor results of national economic productivity. The EBIA addresses this problem but does not ask how to constitute objectives coherent with the problems mentioned. These two significant problems converge in the enumeration of six primary general objectives covered by this vision, as bellow:

  • Contribute to the elaboration of ethical principles for the use of responsible AI.

  • Promote sustained investment in AI research and development.

  • Remove barriers to AI innovation.

  • Promote capacity building and train professionals for the AI ecosystem.

  • Stimulate innovation and development of Brazilian AI in an international environment.

  • Promote an environment of cooperation between public and private sectors, industry, and research centers for the AI development.

By defining these objectives, the policy document enunciates an extensive set of strategic actions to be implemented or coordinated by the federal government. The document created and published was stablished, then, with the proposition of nine thematic axes: three transversal and six vertical axes. At each axis shape strategic actions to Brazilian government steer the policy. These axes emerge from the international benchmark provided by UNESCO but without clearly listing and contextualizing the problems to be solved. The way the strategy was built, the engagement of the stakeholders is poor, consolidating a situation of non-design [21]. The context in which the EBIA policy document was constructed reproduces a situation of a populist government with a technocratic style to steer the policy, in a context of political turbulence. The document does not define responsibilities and partnerships, nor mechanisms for monitoring and evaluation.

The transversal axes are characterized by crossing all the vertical axes and, therefore, concern the ethical content that permeates the vertical axes, as in Fig. 1, below:

Fig. 1
figure 1

Source: elaborated by the authors

Design process of EBIA.

Full size image

Once the problems are identified and components defined, the EBIA lists a series of strategic actions to be implemented by the Brazilian federal government, particularly by the Ministry of Science, Technology, Innovations, and Communications. These strategic actions are related to the set of objectives identified above, as shown in Table 1 below:

Table 1 Strategic actions by EBIA
Full size table

The first horizontal axis is legislation, regulation and ethical use. The first axis demonstrates the concern and understanding on the part of the Brazilian government with the various international and national institutions that concern the subject of AI and data protection. The first axis in some way seek to limit and prevent some risks inherent in the AI development, such as biases, hyper surveillance and opacity. Here it is worth mentioning Brazilian laws such as the General Data Protection Law, the Positive Registration Law and the Decree No. 8.771/2016, which establishes the Open Data Policy of the Federal Executive Branch. In this axis, it is proposed to carry out research actions and normative formats that reconcile state incentives for the AI development in the most varied areas and, at the same time, deeply understand its probable risks and assess negative impacts. The idea is to try to converge existing laws in Brazil created to control and stimulus with a view to open data and new technologies and concerns with the effectiveness of human rights. In this first axis, the EBIA lists a series of legislative dilemmas without listing how stakeholders will be involved in the legislative process. Next, it needs to define the policy advisory mechanisms and the agencies responsible for achieving these objectives.

The second horizontal axis is AI governance. In this axis, fifteen objectives are listed that seek to create networks between governments and the private sector, enumerate data governance objectives, disseminate good practices for managing the design and validation of systems based on artificial intelligence.

Here, the actions intend, therefore, to establish mechanisms, formats and methods of observance of the principles for the development of an ethical AI. However, this axis does not define responsibilities and does not involve essential actors such as the National Authority for Data Protection or agencies responsible for digital development in the government. It also does not involve stakeholders from the private sector or civil society organizations, reducing the scope of governance. The AI governance tends to assume polycentric designs punctuated by concrete problems [47]. However, as a result, the EBIA misses an opportunity to build stakeholder engagement in building AI governance.

The EBIA lists four strategic actions regarding insertion of Brazil in the international AI market on the horizontal axis. These strategic actions deal with the construction of policies oriented towards the international market and the competitiveness of the Brazilian industry. However, in this axis, the EBIA is silent regarding the instruments that will be used to achieve the proposed objectives to compose a consistent framework of instruments related to digital development. Furthermore, it does not address issues such as digital sovereignty and the international data flow. These three horizontal axes create six vertical axes, which define actions aimed at specific sectors of society. These horizontal axes are intersected with the vertical axes, which define strategic actions aimed at industry, markets and governments, as shown in the figure below (Fig. 2).

Fig. 2
figure 2

Source: Brazilian Strategy for Artificial Intelligence [42]

Horizontal and vertical axes of EBIA.

Full size image

The focus of the first vertical axis is the education and qualification promotion of Brazilian citizens for AI production. An important concept should also be highlighted here: digital literacy, “ability to use computers”, which according to the document should be encouraged from childhood. However, the EBIA does not instrumentalize actions to promote digital literacy, nor does it state how labor training and higher education training actions will be achieved. Furthermore, EBIA does not declare the involvement of essential actors in this field, such as the Ministry of Education and the Ministry of Labor, creating objectives without involving the responsible actors.

The major concern of this axis on “labor and capacity building” is training and promoting diversity for the formation of skills capable of making Brazilian citizens able to contribute to the development of this technology in Brazil. The fourth horizontal axes define strategic actions for the application of AI in the private sector, to increase industry productivity, and in the public sector, reinforcing various elements of the Brazilian Digital Government Strategy. Finally, the last horizontal axis involves the creation of strategic actions aimed at the application of AI in the area of public security. In the three cases, the EBIA does not define instruments for coordinating and agreeing on actions with the Ministries of Industry and Commerce, the Ministry of Economy, and the Ministry of Justice. Although the Ministry of Science, Technology, Innovations, and Communications intends to support digital development in strategic sectors of government policies, actions are steered in a technocratic style without involving stakeholders and creating mechanisms to interact with society for policy steering.

EBIA never states which policy instruments or mix of instruments will be used. That is, it does not instrumentalize the objectives listed above and does not formulate policy portfolios to face the problems identified. The identification of the problems is limited to the location of Brazil in different international rankings of competitiveness and innovation, associating the problem with a question of competitiveness and not of policy change. In addition, many strategic actions relate to other policy domains such as education, health, work, and public security. The EBIA did not involve the organizations responsible for these policy domains in its design dynamics and did not even create instruments that ensure policy coordination and authority. The technocratic style of conducting the EBIA design reflects a politically inoperative vision. Without establishing coordination instruments, strategic actions sound like objectives thrown to the wind.

EBIA's vision has difficulties formulating the problem and connecting solutions contextualized in the Brazilian reality. Moreover, the EBIA disregards the entire portfolio of existing policies in Brazil, wanting an integrative and innovative approach to artificial intelligence but needs more means to implement the objectives, nor does it bring an agenda such as the Chinese strategy that brings a 2030 Agenda. The outcome is to produce a fictional policy document that does not engage and guide the actors but reproduces a situation of policy fragmentation and reproduction of a status quo that favors the actors participating in technology governance.

5 Why EBIA fails?

As can be seen, the EBIA’s axes bring important but vague strategic actions. There is a lack of precision and concrete forecast of public policies with an agenda to be fulfilled by the Brazilian State. The EBIA is fruitful in enunciating broad objectives in different topics of strategic action for AI policy. However, it does not instrumentalize these actions in order to create stakeholder engagement, policy advisory, the instrumentation of objectives, the transversal combination of instruments, the definition of monitoring mechanisms and policy evaluation, and policy integration mechanisms.

The main problem of the EBIA is the instrumentation of policies. Instrumentation involves choosing tools and modes of operation to produce policy change. According to Lascoumes and Le Galès [44], every instrument constitutes a condensed form of knowledge about social control and ways of exercising it; instruments at work are not neutral devices: they produce specific effects independently of the objective pursued. Instrumentation is essential in policy design dynamics since it expresses the political choices and power dynamics embodied in the instruments [27]. The way the EBIA was built makes instrumentation fail because it does not involve stakeholders in polycentric institutional designs and does not define governance mechanisms that allow the engagement of actors in a large collective enterprise. As a result, the EBIA was perceived as a mere managerial requirement and not as a political enterprise that could promote collective action for AI-based digital development.

The EBIA fails because it still needs to resolve the collective action problems. First, it was perceived by policy communities as a managerial instrument rather than a political choice. Second, it did not involve stakeholders in the policy design process. Research and development communities, civil society organizations, actors from public and private sectors, and actors from the Legislative and Judiciary branches were not invited to be involved in the process. Third, the public consultation that resulted from the reviewed document contained suggestions for improvement and approaches were not even systematized and presented to society, resulting in a piece that lists vague objectives without instrumentalizing the policy action. Complex policymaking faces substantial risks of failure when horizontal or vertical dimensions of policy-making are not well integrated [45].

Another problem is that the EBIA discards previously existing policies and needs to integrate them as an objective of coherence and consistency of the AI policy. As we said earlier, Brazil has robust funding mechanisms and incentives for research and development in AI at universities and the private sector, funds centers of excellence in AI, encourages access to government data through an open data policy, has a strategy of digital government, designed polycentric and multistakeholder mechanisms for internet governance. However, the Brazilian government has yet to respond to the challenge of integrating these policies in ways that accelerate and drive AI-enabled digital development. AI policy integration is key to accelerating digital development. Policy integration comprises the construction of portfolios of instruments that are coherent and consistent with the policy's objectives so that the needs of markets, industry, and governments can be met and encourage actors in collective actions [45].

The lack of integration of previously existing policies leads the actors to a position of non-cooperation, making the EBIA not a shared strategy capable of creating engagement and solving collective action problems but an ineffective policy document. Instead, the private sector, research and development communities, civil society actors, and the government seek individual solutions, reinforcing their political positions or reproducing the partial benefits of a corporatist position. For example, the private sector benefits from tax incentives without offering any counterpart in digital development. Likewise, research and development communities are committed to producing solutions that need a sense of priorities or actions that can reposition the Brazilian economy in global competition.

In the general context, the Brazilian Strategy for Artificial Intelligence results from a political context in that it is challenging to build stakeholder engagement, reproducing a technocratic governance style associated with a populist government. In this context, a design space is created [26] that does not engage actors to cooperate and package and integrate previously existing policies, produce improvements, and policy change. The EBIA reflects a situation of non-design, in which ideologies contaminate design dynamics without actors being able to participate in this process. In non-design situations, as reflected by the EBIA, it does not produce an idea that can guide policymakers but ineffective actions that no mean to the actors.

6 Conclusions and discussion—when AI policy fails

The case of failure of the EBIA in Brazil provides a lesson on national strategies for artificial intelligence. National strategies comprise policy designs to solve social problems, containing vision, objectives, instruments, and strategic actions. These components comprise politically oriented policy documents to produce stakeholder engagement, assignment of responsibilities, and integration of existing policies to make effectiveness. National strategies for artificial intelligence are politically shaped and should mirror governance styles capable of solving collective action problems and generating motivation for digital development. Policy failures should produce learning, which was this article's primary objective.

Brazil has followed, since the 1988 Constitution, the tradition of encouraging social participation in the debate on the legislative construction of public policies aimed to regulating new technologies. This was the case with the first regulations of the 1990s, the Civil Rights Framework for the Internet and, more recently, the Artificial Intelligence Regulation Bill. The EBIA, however, contrary to any strategy that creates public policies in such a complex matter, did not listen to the main sectors involved and the public consultations carried out were not even publicized or incorporated.

In summary, the EBIA fails when it proposes to be a strategy without being strategic. That is, it is a document that could receive any name other than strategy, which even creates a certain expectation in developmental sectors and citizens that is soon frustrated when reading the policy document. Considering that the scenario is that the Artificial Intelligence Regulation Bill will soon be voted on in Brazilian Congress, the risk that the lack of balance between innovation and protection is high. There is, therefore, a complete mismatch between instruments and objectives, which prevents the strategies from moving forward.

Brazil is a case of non-perspective on AI development. The EBIA fails to design the policy to integrate previous policies, choose the instruments and combine them, define mechanisms for monitoring and evaluating the policy, and establish forms of cooperation and coordination between research and development communities, the private sector, and public sector actors. Furthermore, the dynamic design of the EBIA fails to disregard legal instruments under discussion in the National Congress, such as the AI Regulation Bill. Non-design situations comprise ideological factors based on intuitive decisions that ignore learning and knowledge [46]. The EBIA is a situation of non-design as it recognizes ideological frameworks understood in the form of a benchmark, such as the need for an ethical framework and the idea of digital transformation. However, it does not make these ideas instrumentalize concrete action situations because it needs to solve the problems of collective action inserted in constructing a national strategy.