Deliverables

This document constitutes Deliverable 2.1, derived from Task 2.1 and forming an integral part of EMPOWER’s Work Package 2. Task T2.1 focuses on establishing a link between ethical principles and legal standards in order to develop a robust framework for the appropriate development and use of AI tools within EMPOWER. This task aims to achieve several objectives, starting with the alignment of ethical and legal standards, and extending to the broader objectives of ensuring that the use of AI respects both ethical considerations and legal requirements.

A major focus of this document is an in-depth examination of the ethical dimensions of AI. This analysis underlines that for AI systems to comply with ethical guidelines, their trustworthiness is essential. Based on this understanding, the document is divided into three main sections:

Lawful AI discusses the detailed legal requirements surrounding AI systems, assessing current legal frameworks to ensure that AI functions are in alignment with prevailing laws.

Ethical AI extends the discourse beyond strict legality to encompass broader moral considerations. This section addresses the societal implications of AI, underscoring the importance of AI solutions that, in addition to legal compliance, also align with broader societal values.

Robust AI outlines technical requirements for designing and implementing reliable AI systems, focusing on their consistent and error-free operations.

As AI becomes an increasingly integral part of different sectors, comprehensive guidance is essential. This deliverable, which reflects the Consortium’s commitment to ethical and legal considerations, is intended to serve as a fundamental guide for all EMPOWER activities. The primary objective is to ensure that activities are grounded in a framework of AI ethics, providing a roadmap against which all subsequent phases of EMPOWER can be aligned to ensure both ethical and legal compliance.

This document constitutes deliverable D.2.2 of the EMPOWER project and contains a draft Ethical and Privacy Impact Assessment (EDPIA) of the use of AI in the pilots.

The purpose of the Ethics and Data Protection Impact Assessments (EDPIA) is to perform an impact assessment of the processing activities carried out in each use case selected by the tool providers and LEAs (6 in total). The assessment was carried out on the basis of the information provided by the tool providers, through a questionnaire specifically designed for this purpose (a copy of the questionnaires completed by each provider is included in Annex A of this document). The tool providers’ answers were evaluated by assigning a weight to each of the controls within the questionnaires in order to perform an impact assessment, taking into account the characteristics of each tool and the security and risk mitigation measures implemented.

For each use case identified for carrying out the pilots (detailed in deliverable D.3.1), a specific assessment was carried out. Each EDPIA contains specific risk analyses for each of the tools involved in the use case under consideration.

The selected use cases tool providers involved are:

1. Edge Surveillance Video-ESV; tool provider HERTA;
2. Person re-identification; tool provider VICOM;
3. Video-summarisation tool; tool provider VICOM;
4. Logo identification; tool provider CERTH;
5. Calligraphus; tool provider VOI;
6. Transkit; tool provider TREBE;
7. Clustering tool; tool provider CERTH;
8. Federated learning tool; tool provider CERTH.

Based on the information collected with the questionnaire completed by the tool provider, the logo identification tool (#4) does not process any personal data. According to the applicable legislation, it is not necessary to perform an EDPIA in such cases. This also applies to use case #8, which relies on the same tool.

The EDPIAs have been performed in compliance with Article 35 et seq. of GDPR, Article 27 of LED and A29 AWP29 (now European Data Protection Board also “WP29”) in its Guidelines (WP 248 rev.01).

This document details the analysis and assessment of the legal, privacy, security, and ethical requirements during the design stage of the use cases.

In particular, the Ethics and Data Protection Impact Assessment is one of the main tasks of WP2. It aims to ensure that activities performed throughout the project’s lifecycle adhere to the highest of European Union’s ethical, privacy, and data protection standards.

Therefore, this document:

• Describes the activities carried out and monitoring actions implemented throughout the project to verify that the use cases comply with EU’s ethical guidelines and data protection laws (GDPR and LED).
• Identifies and describes the ethical issues associated with the use of the tools in the specific contexts of the selected use cases.
• Includes the Data Protection Impact Assessments.
• Performs a critical self-assessment of how privacy enabling technologies have been applied in the use cases and recommends improvements.

This document constitutes deliverable D.2.3 of the EMPOWER project and contains a self-assessment questionnaire that serves as a pivotal component in evaluating the trustworthiness of AI tools that are being refined and pilot tested within the scope of the EMPOWER project. The results of this assessment will be used to analyse the trustworthiness of the AI tools in the EMPOWER project and will be reported in D.2.4 as per the deliverable description in the Grant Agreement.

Structured to facilitate a deep dive into the core aspects of AI trustworthiness, the questionnaire is divided into several key sections. Through a series of meticulously designed questions, the questionnaire aims to elicit insightful responses that shed light on the ethical, technical, and operational dimensions of the AI tools in question.

The primary goal of this self-assessment exercise is to gather essential, qualitative, and quantitative information that will provide a solid foundation for assessing the trustworthiness of each AI tool under consideration within the EMPOWER project. This encompasses evaluating the tools’ adherence to ethical standards, their capacity for fair and unbiased decision-making, their resilience against adversarial attacks or data corruption, and their compliance with privacy and AI regulations. Furthermore, while this self-assessment exercise is tailored to the immediate objectives of the EMPOWER project, its structured approach and comprehensive coverage of ethical, technical, and operational considerations provide valuable insights that could be useful for assessing the trustworthiness of AI tools among a wider group of stakeholders beyond the scope of the project.

This document provides a detailed description of EMPOWER project’s orientation towards communication, dissemination and exploitation. This deliverable situates the EMPOWER project in the ecosystem of EC funded projects in this domain and outlines a bespoke approach to communicating with law enforcement stakeholders and other relevant organisations. The EMPOWER project has as its focus the adoption of AI based investigative tools for those LEA personnel engaged in the fight against child sexual exploitation. This is reflected in our communication, dissemination and exploitation strategies. EMPOWER seeks to leverage its relationship with innovation brokers in order to improve uptake of toolsets created in a number of recently completed and ongoing EC funded projects in this field. Furthermore, as part of the dissemination / exploitation piece, EMPOWER will produce training material to improve capacity of LEAs operating in this area.