Beijing’s AI Principles

(link; English translation)

These notes on the very new set of principles from China’s Beijing Academy of Artificial Intelligence (BAAI) are based on the original Chinese version. As far as I can tell, there is no significant difference between the original and the English translation, except for some mistranslations (somewhat ironic given the proficiency of machine translations). Note to self: The fact that there is an official English translation suggests that these principles are meant for an international audience. The brief informal presentation (blog) and the general language used could also imply a lack of extensive effort in developing these principles.

The principles are split into three main categories - R&D, Use and Governance. Main themes include:

  • Developing AI for the good of humanity
  • Developing AI for the good of the environment
  • Accountability
  • Informed consent
  • Data security
  • Robustness
  • Explainability
  • Diversity and inclusivity
  • Fairness and bias
  • Human-AI alignment
  • Openess
  • Job displacement
  • Education and training
  • Multidisciplinary integration
  • Adaptive governance and legislation

R&D

  • Do good - this refers to the good of both humanity and the environment
  • For the service of humanity - this refers specifically to an alignment to human ethics and values, including human rights such as privacy and freedom
  • Responsible - AI practitioners should minimize the ethical, legal and social risks related to the technology
  • Ethical - the development of AI should lead to the building of trust, including the following components: fairness and reducing bias; transparency, explainability, predictability and “auditability”.
  • Diverse and Inclusive
  • Open - promote open systems and prevent the monopolization of data and platforms, as well as to share the benefits of AI amongst different regions and demographics

Use

  • Using AI proficiently and wisely
  • Awareness and consent - stakeholders should be aware of and consent to the impacts of AI systems
  • Education and training - stakeholders in AI systems should receive appropriate training in mental, emotional and technical areas

Governance

  • Optimize employment
  • Harmony and cooperation - this refers to cooperation between different fields and organizations
  • Adaptation and moderation
  • Detailed and practical
  • Long-term planning