Towards a more trustworthy AI: relevance, analysis and prevention of discriminatory biases in the EU context

Tools that operate with Artificial Intelligence (AI) are becoming more and more numerous and are increasingly integrated in our daily lives, allowing the development and improvement of certain tasks, but, at the same time, raising significant ethical concerns.

For this reason, in their latest article, our ethics researcher Pablo Cerezo Martínez, our societal impact researcher Alejandro Nicolás-Sánchez and our CEO Francisco J. Castro-Toledo based on existing literature and different European documents from the last 8 years to date, develop an investigation on the importance of addressing the ethical challenges in the development of AI tools, with special emphasis on the relevance of addressing the issue of discriminatory biases in decision making by AI.

This research highlights the importance of carrying out analyses of potential risks, such as biases in data collection and their subsequent impact on algorithms, or the application of continuous monitoring throughout the life cycle of the tool, adopting approaches that are in line with the Ethics By Design and, on the other hand, pointing out different insufficiencies that are currently found in different documents when we try to provide common definitions and frameworks on the issue of algorithmic biases.

The methodology of the study is based on the review of 21 documents issued by European bodies between 2016 and 2024, including the European Union Agency for Fundamental Rights (FRA), the European Commission or the Council of Europe among others. The inclusion criteria were that the documents were accessible, published by European public institutions and that they covered technical, social, ethical and legislative aspects of AI development, with special emphasis on algorithmic bias.

Finally, the results of the study present a quantitative analysis of the recommendations found in the documents, categorising 152 bias minimisation measures into design, governance and organisational strategies. The absence of a unified conceptual framework at the European level is highlighted, which affects the clarity and effectiveness of bias mitigation measures. However, key areas for future research are identified, such as the continued evaluation of a unified European conceptual framework, global comparative studies and the development of advanced technologies to mitigate bias in AI.

This analysis provides a comprehensive overview of European efforts to address bias in AI and lays the groundwork for future research and development in this critical area.

Read the full text here: https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2024.1393259/full