A group of prominent artificial intelligence experts has put forward the Manhattan Declaration, a set of guiding principles aimed at shaping the future of AI development and deployment. The declaration was signed on September 22, following a high-level meeting at the United Nations headquarters during the 79th session of the UN General Assembly.
The Manhattan Declaration, co-sponsored by Yoshua Bengio, Founder and Scientific Director of Mila - Quebec AI Institute, and Alondra Nelson, professor at the Institute for Advanced Study, addresses the need for a shared global understanding of AI capabilities, opportunities, and risks. This initiative comes in the wake of recent developments, including recommendations from the UN Secretary-General's High-Level Advisory Body on Artificial Intelligence (HLAB-AI) and the release of the International Scientific Report on the Safety of Advanced AI.
The declaration emphasises ten key principles:
- Global scientific cooperation
- Assessment of capabilities and mitigation of AI risks
- Fostering AI as a global public good
- Inclusive participation
- Transparent research and risk assessment
- Interdisciplinary approach
- Responsible development, deployment, and use
- Support for governance initiatives
- Public engagement
- Long- and near-term perspectives
Benjamin Prud'homme, Vice President of Policy, Safety and Global Affairs at Mila, is among the 21 signatories, which include experts from diverse backgrounds and institutions worldwide.
The declaration calls for enhanced international collaboration on AI research, particularly focusing on safety, ethics, and societal impact. It acknowledges the dual nature of AI developments, recognising both their beneficial potential and significant risks, especially in the realm of foundation models.
One of the key points in the declaration states: "We recognise the pressing need to assess the capabilities and associated risks, and address the challenges posed by AI systems, especially increasingly capable ones."
The signatories stress the importance of developing AI systems that contribute to achieving the global Sustainable Development Goals, such as improved health and education. The entire lifecycle of AI systems must align with core principles that safeguard human rights, privacy, fairness, and dignity for all.