Big Data and artificial intelligence in customs control: convergence between strategy, law and ethics

Brazilian customs has been using algorithms and artificial intelligence in customs control activities for at least ten years, with special emphasis on the Customs Selection System by Machine Learning (SISAM), which has made Brazilian customs a great example of technological modernization. 

Through the use of algorithms and trial and error learning, the system analyses the history of import declarations registered in Siscomex, with the aim of helping the institution reduce the percentage of goods verified in import customs clearance, either by increasing the accuracy of the selections, helping customs agents decide which of them will be selected or even deciding automatically for them (1).

This is a new paradigm of control present in the agendas of all customs offices, which, supported by the use of robust technological tools, seek to balance the exercise of their functions of intervention and trade facilitation, the functions of the Customs of the 21st Century. 

This week we learned that the Argentine customs administration also decided to move forward in defining proposals for the implementation of artificial intelligence as an enhancing and complementary tool for the tasks entrusted to the customs service, creating through Resolution 16/2024 the “Innovation Committee on Artificial Intelligence (IAI)”.

The measure is the result of the “Strategic Plan 2021-2025 of the Federal Public Revenue Administration”, which defines, among other operations, the implementation of a new data exploitation model (Strategic Operation No. 13), as well as the strengthening of the customs control system (Strategic Operation No. 16) through the incorporation of new data exploitation and pattern detection tools. 

The Committee will be composed not only of the different departments of the customs administration, but also of participants in foreign trade operations and universities, which should be highlighted as a positive point, considering international guidelines.² which recommend the participation of the private sector and academia for the development of strategic and, above all, ethical AI.

We imagine, however, that the news must have provoked all kinds of reactions, which is natural when the topic of artificial intelligence is under discussion. Due to its intrinsic characteristics of transversality and interdisciplinarity, AI has occupied all social spaces, where benefits are celebrated, risks are exposed and different positions are confronted, both in relation to its intervention in a broader social sphere, as well as in more special and specific applications, as is the case of its adoption by Customs Administrations. 

We have seen at least three major approaches to addressing the issue: the first is the denialist path, which simply prohibits, represses and even criminalizes the use of AI.³, a position that seems naive today, considering all the tasks that new technology is already capable of performing. 

There is a second approach that substantially considers the economic advantages that AI can bring, in terms of providing efficiency, minimising costs and maximising results, the latter being, in addition, key concepts in foreign trade. 

The most balanced path, however, does not seem to be to consider only the economic dimension of technology in the relationships between people, companies and between these and the State, but to understand what risks AI already presents to us, try to minimize them and extract the best experiences from this new reality.

1. Risk management and artificial intelligence: the interweaving of two new customs paradigms

Before delving into the reflections on the use of AI in customs control, we must bear in mind that the risk-based operational approach is, in itself, a paradigmatic revolution, being today considered a true customs management principle⁴. There can be no talk of modern customs control without the presence of a risk management structure and process. 

This model is based on the premise of the continuous existence of risk.⁵, bringing with it two major challenges for customs administrations: (i) identifying the best way to apply knowledge related to risk management, in order to identify events and mitigate them quickly; and (ii) how to apply these tools beyond an operational level, also at the strategic management level. 

For the development and improvement of this model, two requirements are essential: (i) available and data-based information; and (ii) smart technologies. Available and data-based information is what has made possible initiatives such as the electronic release of goods, the implementation of the single window concept (single window) and non-invasive inspections. 

This massive amount of data – a phenomenon called Big Data, which will be explored further below – that customs have obtained, transmitted and exchanged with government agencies and with the administrations of other countries, are then sorted, classified and processed by sophisticated technologies equipped with artificial intelligence. 

There are various technologies available to customs administrations, many of which are developed by the institutional frameworks themselves, as is the case in Brazil. When integrated, they are capable of defining patterns through inference processes and, thus, inducing or deterring social behaviour. 

Unlike traditional computers, which operate based on algorithms pre-established by the programmer, new artificial intelligence systems have the particularity of automatically and autonomously creating the algorithms that will be used in their own operation.⁶. They are, therefore, technologies that act on information⁷and that sometimes not only influence, but replace – and automate – decision-making processes.

2. Some implications of data analysis using artificial intelligence in customs control 

One of the justifications mentioned by the AFIP for the creation of the “Innovation Committee on Artificial Intelligence” is based on the project created by the World Customs Organization in 2019 called BACUDA (“Band of Customs Data Analysts")⁸, which aims to raise awareness and train its members in data analysis.

The project team works closely with research groups from academia and research institutes to develop methodologies and disseminate good practices in data analysis, including by providing members of the Organization with algorithms specifically modeled for the detection of customs value fraud, as well as for the detection of tax misclassification. 

However, data analysis using artificial intelligence raises questions and studies that are not limited to processing speed, storage capacity and algorithmic architecture, which are typical concerns of data scientists and computer engineers. It is also necessary to understand how the use of this massive data set can impact the freedom, privacy and private life of administrators, users of customs services.

The first important feature of the Big Data The fact is that it is not information endowed with an intrinsic meaning. Data, however expressive, needs to be selected, interpreted or understood by an algorithm, which certainly depends on the model devised by the person who encodes or programs it. In addition, the massively accumulated data processed by computer systems are collected from the phenomenal world and, therefore, are only  fragments of reality, not reflecting the totality of reality. And like all fragments, they are partial, as well as precarious and fallible, in the sense that they may not correspond to the reality portrayed, no matter how numerous they are.⁹.

In a study aimed at the use of Big Data According to customs administrations, Yotaro Okazaki says that the phenomenon is characterized by the volume, speed and variety of data, which requires robust technological resources for its processing.¹⁰Volume denotes the size and scale of individually considered data sets, but can also refer to the total amount of aggregated data on the planet. Volume is considered to be speed It includes both the speed and frequency with which data can be created, updated and processed. Variety is synonymous with diversity, as data can be diverse in format, semantics, origin and transmission medium. 

But in addition to these three "V's", there is a fourth "V", which refers to the veracity of the data. Unlike known or familiar data, the Big Data It is, by its nature, generally incomplete or imperfect and, therefore, uncertain and prone to errors. Users are unlikely to suspect that the data in question was obtained from unreliable sources or already touched by unknown parties. Thus, it is important to consider that not always the costumes of data inserted into AI systems will result in a better decision, since its quality is an even more relevant factor. 

Another distinctive aspect about the Big Data It allows for greater surveillance of the lives of individuals and businesses, while making some of the legal means of protecting privacy obsolete.¹¹. In the customs environment, these same concerns exist, especially regarding the consequences of customs administrations blindly continuing to incorporate external data without customs control from the bloc with the issuance of CMC Guidelines No. 32/2008 and 33/2008.¹². 

It should be noted that a decade before the approval of the BACUDA project by the WCO, the Common Market Council had profoundly altered the logic of customs control in the bloc with the issuance of CMC Directives No. 32/2008 and 33/2008.”

CMC Directive 33/2008 established the “Standard on Customs Risk Management” under the justification that the application of risk analysis techniques offers greater facilities to foreign trade operators who have a history of compliance with customs regulations. 

Its text determines that, in order to achieve these objectives, the States Parties shall promote the information processing, which involves the use of computerized procedures that allow the analysis of large volumes of customs operations. 

To this end, the CMC defined that both information sources will be used internal  , the internships for the analysis and evaluation of customs risks. These sources include, among others, the data contained in the customs declaration, the internal database of the Customs Administrations and the information obtained from other bodies or administrations, both national and international.¹³. 

The standard still suggests that risk profiles or selectivity rules be created from a predetermined combination of risk indicators, based on the information collected, analysed and categorised. 

Finally, the Directive determines that customs control actions and their results should be monitored and periodically reviewed in order to obtain adequate feedback, preferably automatically, from the computerised risk management system, in order to improve the selectivity rules.

Considering the confidential nature of these customs risk management processes, it is clear that individuals will not have access to how the data was collected, whether there was balancing in the data set To avoid discrimination, they will also not have access (at least immediately) to the procedural iter followed by the algorithm. 

3. The problem of the absence of a formal law on data processing

One of the doctrinal objections regarding the implementation of automated decisions based on data lies in the absence of a law that expressly enables a public body to issue such decisions. Civitarese says that previously authorizing the administration to issue algorithmic decisions constitutes an inexcusable premise; a derivation of the principle of legality in its most demanding version.¹⁴

The provisions contained in the General Data Protection Regulation of the European Parliament¹⁵ They provide in their text the right of a person not to be subject to decisions issued exclusively by automated means, including the definition of profiles that produce legal effects or that have an equally significant impact on the personal sphere.

In the case of Brazil, the General Data Protection Law (LGPD) provides that the processing of personal data by the public administration must be carried out for the exclusive fulfillment of the public interest, but exempts this observance for cases of national security. 

It is true that customs issues may be security issues, but many times they are only issues of economic or fiscal interest and, in that condition, customs has the same duty to observe their prescriptions. 

In these cases, the administration must communicate the occurrence of the treatment with clear and updated information on the legal provision, the purpose, the procedures and the practices used for the execution of these activities, in easily accessible media, preferably on their electronic sites.¹⁶This is the regulated reality in the country, not necessarily concrete.

4. CONCLUSIONS

There are many challenges at the ethical, regulatory and social levels for the development and use of artificial intelligence, especially in cases involving the Public Administration. If in relations between private entities, where there is consensus and bilaterality, AI offers risks and requires limits, this condition is enhanced in typically customs relations, where legal links are marked by subordination, that is, by the power (of the Administration) and the wrist support (of the administered). 

Due to the small size of this text, it was not possible to address all the difficulties related to the implementation of AI in customs control, as well as the risks in its application, issues that we dealt with in detail in a recently published work.¹⁷For this reason, we focus only on the subject of data, the foundation for the proper functioning of these technologies. 

In order to have a minimum of confidence in relation to the adoption of AI in these processes, it is necessary that there is some type of control and regulation both on (i) the quality of the data, to know if they comply with the requirements of truthfulness, accuracy, precision, thoroughness and, above all, suitability and relevance according to the purposes that justify its use, and (ii) the quality of data processing¹⁸, to find out whether, even from quality data, the programming used for training is suitable to ensure reliable results¹⁹. 

It is therefore healthy that this process be developed in a thoughtful manner and without the external pressures normally imposed for the adoption of these new models. We cannot speak of customs efficiency and optimization of results without taking into account the ethical dimension and the capacity of artificial intelligence to impact the rights of citizens. 

Finally, as important as the aspect of legal regulation – a concern that the Argentine customs administration has shown to be observing in its strategic plan for artificial intelligence – is the training of customs agents. The presence of technological surveillance for the detection of patterns will require agents to increasingly develop critical thinking, both to know how to work using AI, but mainly to know how to refute its results. 

---

1. JAMBEIRO FILHO, Jorge Eduardo de Schoucair. Artificial Intelligence in Machine Learning Customs Selection System. RFB Innovation and Creativity Award. 14th RFB Award – 2015. ESAF Award-Winning Monograph Collection. Available in: Accessed on May 23, 2024.
2. ² OECD AI Principles overview. Available at: https://oecd.ai/en/ai-principles Access on May 23, 2024. 
3.  PEIXOTO, Fabiano Hartmann. Artificial Intelligence and Management: Ethical and Strategic Convergence. Curitiba: Otherness, 2020
4. WCO Risk Management in the Customs Context: Changing operating environments. Available in: Accessed on: 1 May 23. 
5. WCO. Customs in the 21^st Century. Enhancing Growth and Development through Trade Facilitation and Border Security. June 2008. Available in: . Accessed on: 21 May 23.
6.  LACAVA, Federico José. Automatic Administrative Act. Buenos Aires: Astrea, 2022. p. 29
7. CASTELLS, Manuel. 
8. WCO. BACUDA Project. Available at: https://bacuda.wcoomd.org/#bacuda Access on May 23, 2024.
9. SECOND, Hugo de Brito Machado. Directing and artificial intelligence: what algorithms are supposed to teach about interpretation, values ​​and justice. Indaiatuba: Editora Foco, 2023. p. 9.
10. OKAZAKI, Yotaro. Implications of big data for customs-How it can support risk management capabilities. WCO Research Paper, v. 39, 2017. Available in: . Accessed on May 39, 23.
11. For these reasons, the authors recommend that governments implement practical measures to ensure that data on identification and personal activities are only used for legitimate purposes, should be prohibited or shared by irrelevant parties, even though these measures make it difficult for government agencies or I exchange two dice. In: MAYER-SCHÖNBERGER, Viktor; CUKIER, Kenneth. Big data: A revolution that will transform how we live, work, and think. Houghton Mifflin Harcourt, 2013. 
12.  MERCOSUR. Regulations of Mercosul's decision-making bodies.Available in: . Accessed on: 08 Feb. 2024.  ¹³ MERCOSUL/CCM/DIR. No. 33/08, article 10.
13. MERCOSUL/CCM/DIR. No. 33/08, article 10.
14. CIVITARESE, Stefano Matteucci. Umano troppo umano. Administrative decisions are automated and legal principles. Public Law, v. 25, no. 1, p. 5-42, 2019. Available in: . Accessed on: 4233 Feb. 17.
15. CIVITARESE, Stefano Matteucci. Umano troppo umano. Administrative decisions are automated and legal principles. Public Law, v. 25, no. 1, p. 5-42, 2019. Available in: . Accessed on: 4233 Feb. 17.
16.  Law No. 13.709, August 14, 2018, article 23. Available in: . Accessed on: 03 Feb. 2015.
17. We deal with more details of these questões in recently published works. Reis, Raquel Segalla. Risk management in customs clearance of imports: artificial intelligence as an instrument and control agent. São Paulo, SP. NSM Editora/Caput Libris, 2024. 
18.  The doctrine already speaks in unbundling algorithms. These are types of artificial intelligence used to mitigate noise or noise in human decisions, serving as an instrument not to substitute, but to enhance human behavior. An example given by Machado Segundo would consist of an algorithm that, instead of important to the player, a prompt voting model, based on precedents, for all and any case that comes up, it has the role of apopping the decisions whose review is submitted to the court (administrative or judicial), separating those that it considers convergent as the understanding of the court, those that give it diverge, in order to demand greater attention to the latest. In: SECOND, Hugo de Brito Machado. Directing and artificial intelligence: What algorithms are supposed to teach about interpretation, values ​​and justice. Indaiatuba: Editora Foco, 2023. p. 922.
19. FRAZÃO, Ana; GOETTENAUER, Carlos. Black box eo direct face à algorithmic opacity. Barbosa, Mafalda Miranda. Digital Management and Artificial Intelligence. Indaiatuba: Editora Foco, 2021. p. 29.