China releases report on ethical governance for embodied AI

China on Tuesday released a report on the ethics of embodied intelligence – the field that covers machines that can sense, decide and act in the physical world.

The report treats embodied intelligence ethics as part of broader AI ethics. It flags six risk areas: harm to human autonomy, to social relationships and to development rights; physical injury and threats to public safety; privacy violations and misuse of personal information; bias, discrimination and exclusion from services; identity misidentification and overestimation of capabilities; and unclear accountability, with incidents hard to reconstruct.

The report defines five risk levels, R1 through R5, based on how serious the consequences are. It also sets out six basic governance principles: human-centered, safe and controllable, privacy-protecting, fair and inclusive, transparent and explainable, and accountable and traceable.

Ethical requirements should be built into every stage of an embodied AI product's lifecycle – design, R&D, deployment, operation and service, and decommissioning – and each stage should have clear protection requirements and assigned responsibilities, said Zhang Shizong, deputy director of the IT Research Center at the China Electronics Standardization Institute (CESI).

Zhang said the safeguards work in three ways. Limits on what the machine can do to protect physical safety. Controls on what it can collect to protect sensory data. Clear signals that it's a machine, plus ways to switch it off, preserve the user's ability to choose.

China will accelerate work on national standards including guidelines for ethical governance for AI and embodied intelligence, Zhang said.