Background and context

Customs administrations today are increasingly dependent on data and data analytics to manage operations at the border. As part of digitalisation efforts, customs are deploying data analytics solutions that feature elements of Artificial Intelligence (AI). Examples include X-ray image interpretation for automatic threat detection and natural language processing for analyzing free-text goods descriptions in customs declarations.

AI solutions hold large potential when it comes on one hand to automating time-consuming and labor-intensive customs processes and on the other hand to reaching beyond human capabilities in identifying suspicious materials, packages, containers, transport movements and operators in the context of cross-border logistics and global supply chain operations. However, like with any powerful technology, using AI calls for careful considerations of ethical, legal and technical matters to avoid the misuse or critical malfunctioning of AI-powered systems. After all, AI technologies tend to be complex, opaque and sometimes unpredictable from the human perspective.

To mitigate undesired effects of AI, the European Union is preparing a legal framework to regulate AI. The centerpiece of this framework is the upcoming “AI Act”, which lays down rules for the design and use of AI technologies.

Scope and goals

This Expert Report will analyze requirements of the proposed regulation COM(2021)206 final “Laying down harmonized rules on artificial intelligence (Artificial Intelligence Act) and amending certain Union legislative acts”. This proposal for a regulation is commonly known as the “AI Act”, and it is available for download here:

• https://digital-strategy.ec.europa.eu/en/library/proposal-regulation-laying-down-harmonised-rules-artificial-intelligence

The goal of this expert report is to review the proposed AI Act and highlight its elements that may have implications to how Customs administrations employ AI in (i) their research, development, and innovation activities and (ii) in day-to-day operations.  In particular, PEN-CP Customs partners are most interested in learning what Customs can and cannot do with AI, today and in the future.

Examples of areas where Customs administrations may benefit from AI technologies & solutions are shared in Annex 1.

Project tasks

The main tasks to be carried out are the following ones:

• Kick-off workshop with 3-4 customs experts (PEN-CP Data & risks team (= PET-1), contacts are readily available)
• Review the proposal for a regulation COM(2021) 206 final = “AI Act”
  • COM(2021)206: “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”
  • https://eur-lex.europa.eu/legal-content/EN/HIS/?uri=COM:2021:206:FIN
• Highlight the potential implication of AI Act in the context of AI-related (i) Customs Research and Innovation projects, and (ii) Customs operations
• Write the full report
• Present your work to PEN-CP Customs partners, during our monthly online meeting (= Penday)

Report structure

• Chapter 1. Upcoming AI Act in a nutshell
• Chapter 2. AI Act & requirements, limitations and constraints with Customs-relevant AI technologies and what Customs can and cannot do with AI
• Chapter 3. Conclusions and recommendations

Report length and language

Target 12-16 pages (no page limit with annexes); in English

Applicants and application process

You can apply either as a natural person or as a legal entity (registered company with a VAT ID). In case you are a team of two natural persons, please identify clearly who is the main contact person for contractual purposes. It is also important to note that the call is announced in public and is open for anyone knowledgeable on the topic to apply. We may also approach potential experts by email.

The application process consists of following three steps:

• Step 1: Applicant expresses his/her interest by sending an email to pen-cp-arttic@eurtd.com
• Step 2: Applicant receives credentials by email (from PEN-CP) enabling the proposal submission directly on the project portal
• Step 3: Applicant submits the proposal (both technical and CV -sections)

Documents to submit

Submit the following documents as part of your proposal here:

1. Technical proposal (max 2 pages)

• Describe the methodology you will use to achieve the goal for the assignment
• List the activities you plan to conduct during this assignment
• Detail the main risks and limitations you foresee with the assignment, including a brief risk mitigation plan

1. CVs of 1-2 experts who will carry out the work, highlighting their past experiences on similar topics.

 Evaluation criteria and points (max 100 points)

• Criterion 1: The quality of the methodology, project work and risk plan = max 60 points
• Criterion 2: The relevant past experience of the 1-2 experts included in the proposal = max 40 points
• Minimum threshold to be considered for selection: Criterion 1 = min 35 points; Criterion 2 = min 25 points

Evaluation and contract signing process

• Step 1: Evaluation of all applications by scoring and ranking them.
• Step 2: Selection of the applicant with the highest score and initiation of the contractual process.
• Step 3: Communication with the non-selected applications of the evaluation results.
• Step 4: Contracting with the selected applicant.
• Step 5: Possibility to contact the unsuccessful applicant (if the contract cannot be closed with the selected applicant for a reason), to suggest contract negotiation (then follow with Step 4)

Dates, contacts and consulting fee

• Call publication date: 5 October 2022
• Deadline for proposal submission: 16 November 2022
• Call results available (expected): 25 November 2022
• Project start date: Latest by 2 January 2023
• Project duration: 2-3 months (to be agreed at the time of contract signature)
• Administrative contact person: Ms. Valerie Mellier, ARTTIC; pen-cp-arttic@eurtd.com
• Consulting fee: 5.900 EUR

Note-1: The official version of this ER6 call text is available at the PEN-CP online platform page: https://pen-cp.sym.place/groups/profile/302707/pen-cp-innovation-instruments-and-innovation-events

Note-2: PEN-CP reserves rights to make updates on the dates listed above, in case no proposal reaches the minimum threshold; also, we may cancel the action, in case no competent service provider is found

Annex 1. Examples of areas where Customs administrations may benefit from AI technologies & solutions

Computer Vision and Object Identification. Customs are interested in developing automatic threat detection algorithms that can find drugs, weapons, and other contraband in X-ray images. AI-based solutions can learn to distinguish shapes, sizes, and responses to  ionizing radiation to find concealed goods inside cargo containers and vehicles of transport

Machine Learning, including Deep Learning. Modern customs follow a risk-based approach for controlling cross-border cargo flows: high-risk goods get inspected while most low-risk traffic can pass customs without intervention. The risk-based approach to targeting relies heavily on data analytics that allow customs to find anomalies in declaration and other available data, for example suspicious routings, unusual trade transactions, and inconsistent information (e.g., declared weight does not correspond to the reported type and amount of goods). Machine learning solutions have the potential to discover more sophisticated high-risk patterns, correlations, and trends in large datasets than conventional analytics or human reasoning. Machine learning could improve progressively from customs control feedback (results of cargo inspections) and continuously uncover new insights without having to be explicitly re-programmed. Over time, the machine would learn how to generate more accurate risk indicators and profiles and discard inefficient old ones that yield high rates of false positives.

Natural Language Processing (NLP). Customs could use NLP to collect unstructured and semi-unstructured data from various sources, including e-commerce sites, company websites, customs declarations, and trade documents like shipping instructions. Features of Natural Language Processing like translation, transliteration, and semantic analysis will turn unstructured information into analyzable data. NLP could be used also to predict commodity codes based on free-text goods descriptions that traders provide, this way helping customs to estimate the risk of misclassification of goods and duty fraud.

Other AI-related areas. Smart data visualizations would help customs officers to understand outcomes of risk analysis: why was a shipment selected for control? They could also use the visual analytics to cluster, sort, and connect data elements and this way get a better understanding of the overall dataset and derive new risk indicators and profiles. Algorithms with predictive powers could help customs to find faster suspicious movements of ships and cargo containers.