Page 126 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

21 Building the technical foundations for embodied intelligence

in connected ICT environments

This workshop addressed embodied intelligence and its dependence on advanced ICT
infrastructures, with the goal of identifying standardization priorities. Panelists included
representatives from academia, industry, and standards bodies. The discussion followed four
themes:
i. Definitions
ii. Technical challenges and use cases
iii. Standardization gaps
iv. Pathways for collaboration

21�1 Defining embodied intelligence

The session opened by clarifying the concept of embodied intelligence. A definition from the
Chinese robotics industry community describes it as intelligent systems interacting with the
environment through physical entities such as robots, capable of environmental perception,
cognition, autonomous decision-making, and action execution.

Three main distinctions from traditional AI were outlined:

• Physical dependence: Embodied AI operates through tangible hardware with physical
constraints (e.g. robotic arms, sensors, and motors)
• Perception-action loop: Continuous real-time decision-making under physical laws,
requiring feedback and adaptive strategy.
• Irreversible consequences: Actions have direct physical effects, unlike reversible digital
outputs.

Additional interventions emphasized that embodied systems are inherently social when
operating in human environments. The form, movement, and appearance of robots influence
trust, usability, and human expectations. It was noted that robots, even when not designed for
social interaction, often elicit social responses due to their presence in human spaces. Attention
was called to the importance of sensitivity to context and awareness of the risk of embedding
stereotypes into robotic embodiments.

21�2 Technical and infrastructural challenges

Industry participants highlighted latency, connectivity, and bandwidth as critical barriers.
Teleoperation across continents was reported to create delays of up to 2-3 seconds, undermining
usability. Stable VR-based teleoperation was said to require bandwidths of around 100 Mbps.

Further challenges discussed included:

• Accessibility: Dependence on cloud infrastructure limits use in rural and underserved
regions, often where robots are most needed.
• Privacy and control: Users risk losing functionality when services close (e.g. the "Boxy"
children’s robot that stopped working when its company shut down). Concerns were
raised about where data goes, who has access, and how secure it is.
• Security: Robots and their telecom interfaces can be "data leaky" and hackable, exposing
biometric data such as eye tracking, facial expressions, and haptic signals. Risks are
significant in sensitive contexts like healthcare or elderly care.

114