In the framework of the Paris Peace Forum Official Side Event on the Road to AI Summit the Forum on Information and Democracy organized a workshop on 12 November 2014 in Paris. Assembling industry representatives, researchers, civil society and a representative from an existing voluntary certification mechanism, the Forest Stewardship Council, participants explored the idea to create a voluntary certification mechanism for public interest AI.
To investigate the idea of a voluntary certification mechanism for public interest AI, the Forum organised a discussion with a board member of the Forest Stewardship Council, Julie Marsaud, as well as a representative from the AI industry, notably, Amanda Craig, Senior Director Responsible AI Public Policy, Microsoft.
Participants highlighted that the purpose of such a voluntary certification mechanism for public interest AI, as explored in the Forum’s recent research paper and developed originally in its report AI as a Public Good, is a good one responding to a lack of transparency and understanding among users of AI systems. As explained, the objective would be, among others, to tackle the information asymmetry between AI providers and users, while also creating a market incentive to develop ethical AI.
Similarly, the Forest Stewardship Council (FSC) certification, one of the inspirations for this mechanism, created incentives for responsible and sustainable forest management. As participants highlighted, one of the key difficulties for a certification mechanism for AI is the lack of knowledge about what should be standards, how different elements of AI systems work.
It is thus important to focus on the elements that would make a difference, be open to start with a more niche certification that evolves over time and also build upon existing assessments that already cover some criteria. Similarly, the FSC criteria have completely evolved since its creation more than 30 years ago. FSC also highlighted the importance of taking time and setting up a rigorous process to develop and revise assessment criteria. In general, however, it is important to create convergence with other existing initiatives and to learn from past experiences, such as in the attempts to set cybersecurity standards.
Participants also discussed the governance and financing of the mechanism, highlighting the importance of not creating a financial dependence on industry actors as this would be to the detriment of trust. It was also underlined to include on the one hand those that will have to apply the standards but also those that are impacted by the product.
The Forum on Information and Democracy will organize several future workshop sessions to further explore the idea to create a voluntary certification mechanism for public interest AI and invites all interested stakeholders to reach out.
