European Commission publishes Guidelines for a Trustworthy Artificial Intelligence


On April 8, 2019 the European Commission has published its final version of the Ethics Guidelines for Trustworthy AI. In December, 2018 the Commission had published a draft version for public consultation, receiving over 500 reactions. The now published final version has undergone several changes, and Telefónica is proud that it could contribute to this multidisciplinary achievement. In the final version, there is strong alignment with Telefónica’s AI Principles

In this post, we will give a high-level comparison of the EC Guidelines with Telefónica’s AI Principles. The EC Guidelines are divided in three parts: the foundations of trustworthy AI expressed through four ethical principles, the realisation of trustworthy AI implying seven key requirements, and the assessment of trustworthy AI leading to an assessment list with specific questions. 


1. Foundations of trustworthy AI

The Guidelines mention the principles of Respect for human autonomy, Prevention of harm, Fairness and Explicability. Those principles are directly covered in Telefónica’s AI principles: Human-centric, Fair, and Explainable AI.


2. Realisation of trustworthy AI

The Guidelines mention seven key requirements: (1) human agency and oversight, (2) technical robustness and safety, (3) privacy and data governance, (4) transparency, (5) diversity, non-discrimination and fairness, (6) environmental and societal well-being and (7) accountability.  All those requirements are covered in Telefónica’s AI Principles or its more general Business Principles.

Moreover, we are developing and implementing a methodology for “Responsible AI by Design”, that provides an operationalization of the AI Principles in the company. It provides concrete guidance to our employees with respect to applying those requirements when they develop, buy or use systems based on AI technology. For instance, we provide technical training and tools for how to detect and remedy bias in training data sets to avoid possible unfair discrimination. We also provide training and tools to make algorithms explainable in case their outcome has an impact on people’s lives. Privacy and data governance are a key pillar for our company, and we are transparent as to what personal data we use for what purposes. Our “Fourth Platform” is built around the notion of transparency. 


3. Assessment of trustworthy AI

The third part of the Guidelines provides a non-exhaustive list of questions to ask, organised by “requirement”, to enhance the likelihood of building trustworthy AI, while minimizing the risk of creating undesired, negative consequences. Telefónica’s Methodology for Responsible AI provides a similar set of questions for each of the Principles identified, which is in broad alignment with the assessment list of the AI Guidelines.


Apart from this comparison, we are also satisfied that the Guidelines explicitly state that they are applicable beyond Europe, and encourage a global discussion on this matter; technology and business don’t stop at borders. Moreover, we welcome that the Guidelines mention the role of the business model in several of the requirements, such as Transparency and Explainability.  To understand potential, undesired consequences of AI, it is not only the technology that is relevant, but above all the business model for which the technology is put to use.

Finally, Telefónica welcomes the launch of the pilot phase with stakeholders to test the proposed Assessment of trustworthy AI and is looking forward to participate in this important exercise.