Periodic Reporting for period 3 - SHERPA (Shaping the ethical dimensions of smart information systems (SIS) – a European perspective)
Reporting period: 2020-05-01 to 2021-10-31
Smart information systems (SIS), the combination of artificial intelligence (AI) and big data analytics, are key technologies that have the potential to significantly affect most aspects of modern societies. They attract significant public and private investment and are the basis of much public debate. These systems promise benefits across many application areas. At the same time, they elicit concerns about their ethical, human rights and broader social consequences. The SHERPA project sought to contribute to a better understanding of these technologies, the way in which their ethical and human rights consequences materialise in practice and what can be done to address these. The team worked with a broad range of stakeholders to develop recommendations for policy and other stakeholders to ensure that AI innovation ecosystems are conducive to human flourishing. The objective of the project was to make a practical difference to how European societies deal with these technologies by providing a better evidence base, proposing practical and implementable recommendations and working to ensure that these recommendations are considered in European decision-making structures.
The project’s initial phase aimed at gaining a better understanding of ethics and human rights implications of SIS. 10 case studies application areas that employ SIS were collected and analysed. A further five areas were investigated using the newly developed methodology of policy scenarios. In parallel, the project analysed the human rights impacts of SIS and technical aspects of security weaknesses of SIS, which can undermine their trustworthiness. These investigations were accompanied by the creation of a stakeholder board comprising approximately 30 individuals representing a range of stakeholder groups. The stakeholder engagement in the project was further supported by a stakeholder list including more than 1,000 individuals. In addition, a Delphi study of experts was undertaken to validate key project findings. The consortium contributed to standardisation activities and explored threats to and defense approaches for SIS, focusing in particular on the class of training data poisoning attacks. The project undertook successful dissemination and communication activities, including a website (reaching 153,393 views), Twitter, LinkedIn and Facebook (with 668 followers on Twitter, 354 connections and 191 followers on LinkedIn and 57 followers on Facebook), videos (including 11 animations, interviews with policymakers, webinar recordings), as well as communication tools, such as webinars, regular newsletters, and printed materials (flyers, infographics and posters). Our artist, Tijmen Schep, received a .eu Web Award for "How Normal Am I"?, an interactive documentary about face recognition that has now been viewed 700.000 times. The project closely collaborated with other EU projects, SIENNA and PANELFIT and its successor project TECHETHOS which led to the high-profile shared final event of the three projects which was co-hosted by the Panel for the Future of Science and Technology of the European Parliament (STOA). SHERPA was accepted as a Partnering Project for the Human Brain Project (HBP) which facilitated the creation of synergies between SHERPA and the HBP, including the launch of the HBP’s Opinion on AI. SHERPA made a significant contribution to the academic debate, as evidenced by the publication of 26 journal papers (as well as 5 more pending acceptance/publication) and two open access books (one to be published in 2022). The award-winning artistic representation of the project attracted broad attention and contributed to the positive reception of the project. The aim of the project of providing actionable advice to policy and decision makers was achieved through the advocacy work of the project. A total number of 21 advocacy activities/events took place, encompassing all the key institutions involved in regulating artificial intelligence at the European level. They included 10 interviews published on Youtube, and conducted mainly with Members of the European Parliament, notably from the STOA and the Artificial Intelligence in the Digital Age committee (AIDA). We also held 9 bilateral meetings with various stakeholders, from national research bodies, member states, the European Parliament, the European Economic and Social Committee (EESC) and the European Commission (EC). Within the EC, we discussed the project with Directors, Advisors and Heads of Units from DG Connect and DG Education and Culture (EAC). With those interviews and meetings, we reached 22 policy makers. Before the pandemic, two events were held in person with 39 attendees in total.
SHERPA went beyond the state of the art in several ways. It made an important academic contribution to the AI ethics debate by demonstrating the value of adopting the concept of innovation ecosystems to analyse the impacts of AI. Based on the empirical and conceptual research and stakeholder interaction, SHERPA could argue that the promotion of human flourishing will benefit from the use of the ecosystems approach. This conceptual basis has allowed SHERPA to develop a set of recommendations based on three main principles: the clear delimitation of the ecosystem in question, ecosystem governance and the knowledge base of the AI ecosystem. These three principles gave rise to nine individual recommendations related to: conceptual clarity, a regulatory framework, an EU Agency for AI, the creation of the role of an AI Ethics Officer, development of ethics by design approaches, AI impact assessments, educational principles, inclusion into standardisation, and specific technical guidance for security of AI. The project was well-timed which allowed it to feed its findings and recommendations into various high-level policy processes. Most notably the development of the SHERPA recommendations coincided with the work of the AI High Level Expert Group with whom SHERPA worked closely, and the subsequent proposal for a European Regulation on AI. As part of the advocacy work, a number of exchanges with individuals involved in the drafting and review of the Regulation took place. Policy briefs prepared for the regulatory options and EU AI agency were also shared with policy makers at the European level. The impact of SHERPA is expected to be long-lasting. SHERPA has been cited in several calls of the Horizon Europe (HE) Framework Programme and will influence future research. It is a constitutive pillar of the TechEthos project which has already started and which contains consortium members. The ethics by design approach which was developed across SHERPA and SIENNA will form part of the ethics recommendations for AI projects in HE. These materials have been piloted in a number of contexts, including as delivered in training sessions for researchers and employees within the EC H2020 funding programme, and as offered to those who intend to bid for HE funding in October 2021. Ethic by Design will also be developed as a commercially available training opportunity through the ORBIT company, a spin-out company of De Montfort University and the University of Oxford. SHERPA’s security-focused work was mentioned in the “Input to the Horizon Europe Programme 2021-2027 Priorities for the definition of a Strategic Research and Innovation Agenda in Cybersecurity” document prepared for the EC by the European Cyber Security Organisation (ECSO).