Editor

Debate continues in the European Union about AI legislation. The European Commission had promised comprehensive AI legislation in “the first 100 days” but that initiative stalled as did efforts to establish red lines for certain AI techniques, such as facial recognition. The Commission’s White Paper on AI now describes several policy options. Access Now and EDRi said that the Commission’s “risk-based approach” fails to safeguard fundamental rights. As they explained, “the burden of proof to demonstrate that an AI system does not violate human rights should be on the entity that develops or deploys the system” and “such proof should be established through a mandatory human rights impact assessment.”

This week the European Parliament adopted three resolutions on AI policy intended to influence legislation from the Commission. The resolution from Iban García del Blanco (S&D, ES) urged the Commission to establish legal obligations for artificial intelligence and robotics, including software, algorithms and data.  The legislative initiative of Axel Voss (EPP, DE)  would make those operating high-risk AI systems strictly liable for any resulting damage. And a third resolution on intellectual property rights, by Stéphane Séjourné (Renew Europe, FR), makes clear that AI should not have legal personality; only people may claim IP rights.

The European Parliament adopted all of these proposals in sweeping majorities, across parties and regions. But even those proposals are unlikely to meet the concerns of civil society.  As Access Now and EDRi said of the resolution on AI ethics, “They are cautious and restrained on fundamental rights, taking only tentative steps to outline the biggest threats that artificial intelligence pose to people and society, while also failing to propose a legislative framework that would address these threats or provide any substantive protections for people’s rights.”

Still ahead is the Commission proposal on AI regulation. Here again EU NGOs have made clear their position: (1) prohibit systems that infringe fundamental rights; (2) establish mandatory human rights impact assessments for all AI systems and (3) strengthen enforcement of existing law, including data protection.

Marc Rotenberg, Director

Center for AI and Digital Policy at Michael Dukakis Institute

The Center for AI and Digital Policy, founded in 2020, advises governments on technology policy

The Global Privacy Assembly, the worldwide conference of privacy commissioners, met this week to pursue a strategic plan, to address the challenges of the pandemic, and to adopt resolutions on emerging privacy issues. The Assembly is chaired by Elizabeth Denham, the UK Information Commissioner. In opening remarks, Denham spoke of the challenges of the past year and the creation of the COV-19 Task Force. She emphasized “Continued modernization. Collaboration. Community.”

The strategic planning sessions highlighted three goals: (1) advancing global privacy in the digital age, (2) maximizing the Assembly’s voice and influence, and (3) capacity building. Subsequent sessions reviewed the work of the GPA Task Force on COVID-19. Ada Chung, the Privacy Commissioner for Hong Kong, reported that data protection agencies have similar best practices for the pandemic, including privacy impact assessments, data minimization, and transparency. She said, “This indicates that GPA members worldwide share the same values and principles in personal data processing, which is encouraging.” A Resolution on Data Protection and COVID called upon those involved in contact tracing “to take a data protection and privacy by design approach when designing these measures and to recognise the pivotal role privacy and data protection play in promoting public trust.”

The Assembly adopted a significant Resolution on Accountability and AI that urged organizations deploying AI systems to implement accountability measures, including a human rights impact assessment. The Privacy Assembly also urged governments to make changes to data protection law “to make clear the legal obligations regarding accountability in the development and use of AI.” The GPA AI Accountability Resolution builds on the 2018 Declaration on Ethics and Data Protection in Artificial Intelligence as well as a recent GPA survey that identified accountability measures that  are “very important or important for either AI developers or AI users.”

The Global Privacy Assembly also adopted a Resolution on Facial Recognition Technology.  The Resolution reiterated several key principles for data protection, such as fairness and transparency, but stopped short of endorsing a formal ban which had been urged by many human rights advocates at the 2019 conference in Tirana. More than 100 organizations and 1,200 experts recommended that “countries suspend the further deployment of facial recognition technology for mass surveillance” and “establish the legal rules, technical standards, and ethical guidelines necessary to safeguard fundamental rights and comply with legal obligations before further deployment of this technology occurs.” The Assembly said it would consider the “circumstances when facial recognition technology poses the greatest risk to data protection and privacy rights,” and develop a set of principles that could be adopted at the next conference.

Civil society has traditionally had a strong presence at the international meeting of the privacy commissioners. However, there was little participation by civil society this year. The European consumer organization, BEUC, was the only organization granted observer status.

The next conference of the Global Privacy Assembly will be held in Mexico in 2021.

Marc Rotenberg, Director

Center for AI and Digital Policy at Michael Dukakis Institute

The Center for AI and Digital Policy, founded in 2020, advises governments on technology policy

The United States National Security Commission on AI met this week to vote on recommendations and an interim report to the US Congress. The Commission adopted almost all of the 80 recommendations, derived from  six work streams. The Recommendations concerned primarily AI R&D, AI and national security, training AI talent, building US AI leadership, and global cooperation on AI and also ethics. There were also recommendations (and an interesting discussion) concerning “malign” information, enabled by AI..

There were several references to “ethics” in the discussion of workforce training and the management of genomic data in the Recommendations.  “Ethics” appears in the discussion of  the AI Partnership for Defense, which seeks to “provide value-based global leadership” on adoption of AI. The Commission recommended an Agreement on AI principles and ethics to govern NATO’s development and use of AI. The Commission also proposes to operationalize the DOD Principles of AI Ethics – Responsible, Equitable, Traceable, Reliable, and Governable.  There were several references in the Draft Report to “bias” in data sets and in STEM education. There was a recommendation (#64) to “establish a Digital Coalition of democratic states and the private sector to coordinate efforts and strategy around AI and emerging technologies, beginning with a Digital Summit.” This recommendation builds on growing support for the Quadrilateral Security Dialogue, a strategic forum among the US, Australia, India, and Japan.

The words “privacy,” “fairness,” and “transparency,” appear frequently in the 263-page Draft Recommendations. However, there were no specific recommendations to operationalize these principles. This was surprising given the President’s 2019 Executive Order on AI, which emphasized American values such as privacy and civil liberties, as well as recent statements by US Chief Technology Officer Michael Kratsios, who told POLITICO earlier this week that the US stood behind AI policies that support democratic values and specifically criticized the “dystopian” social credit scores of China. There is also White House Guidance to federal agencies that underscores privacy, civil liberties, human rights, the rule of law, and respect for intellectual property. None of these goals appear in the recent recommendations of the National Security Commission on AI, and it remains unclear whether the Commission supports the application of these regulations to the defense agencies in the US federal government. This is of concern as one of the most notable early AI applications in the US was the Total Information Awareness program, coordinated by the Department of Defense, to track and profile Americans. Other governments are currently exploring the integration of data sets containing personal information to launch AI national initiatives.

In sum, the recent National Security Commission on AI Recommendations promote ethics, acknowledges bias, and proposes a strategy for allies that seeks to maintain accountability for AI systems in the defense setting. However, the report leaves open important questions about the application of national values to AI projects within the defense agencies. Also notable, the documents for the Commission are stored on the proprietary Google Drive rather than on the Commission’s website. Public documents are typically stored on a public website.

Marc Rotenberg, Director

Center for AI and Digital Policy at Michael Dukakis Institute

The Center for AI and Digital Policy, founded in 2020, advises governments on technology policy

The OECD has published the first report that attempts to assess the implementation of the OECD AI Principles among the G-20 nations. Examples of AI National Policies surveys “rationales and illustrative actions” for the 10 principles that make up the OECD/G-20 Guidelines on AI policy. The report was prepared by the G20 Digital Economy Task Force.

Key observations from the Task Force report:

• G20 countries are moving quickly to build trustworthy AI ecosystems, though most initiatives are very recent
• Many national AI strategies address multiple G20 AI Principles simultaneously, which the OECD contends reinforce the strong complementarity of the Principles
• So far, few national policies emphasize Principles of robustness, security and safety, and accountability,
• Many national policies emphasize R&D, fostering a digital ecosystem, human capacity, and international cooperation

The Task Force also found that “there is potential for steering public research towards socially oriented applications and issues, and for leveraging R&D activities to make progress on issues such as accountability, explainability, fairness and transparency.” The Task Force emphasized that there “is currently a critical window for G20 members to continue their leadership on AI policy issues and to promote implementation of the G20 AI Principles. Development, diffusion and use of AI technologies are still at a relatively early level of maturity across many countries and firms, and policy-making on AI is in an active experimental phase.”

In the world of international policy frameworks, assessing implementation of policy goals remains one of the most difficult tasks. Reporting is often uneven, there is considerable trust placed in country self-reporting, and there may be disputes as to the correct metrics and the standards for progress. Nonetheless, the G20 Digital Economy Task Force should be credited for this substantial survey of the G-20 efforts to implement the OECD/G-20 AI Principles.

The G-20 adopted the OECD AI Principles at the 2019 Ministerial meeting in Osaka. Work on the OECD/G-20 AI Principles continued at the Ministerial meeting in Ridyah. The 2021 G-20 Ministerial will be held in Rome. The next G-20 meeting provides a critical opportunity to asses further progress on the AI Principles, as well as to see whether the G-20 countries have begun to implement the Principles of robustness, security and safety, and accountability,

Marc Rotenberg, Director

Center for AI and Digital Policy at Michael Dukakis Institute

The Center for AI and Digital Policy, founded in 2020, advises governments on technology policy