How should embodiment and deployment scale affect obligations?
A system's permissions and combined deployment determine what it can put at risk. Embodiment changes the obligations attached to consequential systems while ordinary product safety retains its own scope.
SIRDA editorial hypothesis
For systems within catastrophic-risk scope, physical access and coordinated deployment scale can determine additional evaluation and operating obligations. Assessment can examine shared control and correlated failure across the combined system. This hypothesis fails if dividing a dangerous deployment into smaller units evades review or if embodiment alone draws routine machines into the mandate.
0 objections ยท View challenges
Challenges to the editorial hypothesis
No published objections yet.
The argument
Consider the same model in three arrangements: a research sandbox, a laboratory with physical controls, and a fleet of machines receiving shared instructions. These are thought experiments about deployment, with different permissions, exposures, and opportunities for interruption. The relevant assessment describes the entire arrangement.
The mandate attaches to the catastrophic capabilities and risk of that arrangement. A conventional factory arm or household appliance remains an ordinary product-regulation question. An aggregate system can enter review because its combined operation meets the defined threshold, even where its components considered separately appear modest.
An assessment needs to record which machines the system can reach, who authorizes physical actions, how a shared failure could propagate, and whether operators can interrupt it. Deployment changes reopen the relevant portion of the assessment. This keeps an approval tied to the conditions that justified it.
How to test it
Compare a sandbox with a coordinated physical deployment, then split the latter across subsidiaries and jurisdictions. Test whether the framework tracks combined control and consequential changes while keeping routine machines outside its scope.
Related Ecosystem
- Ada Lovelace Institute
- AI Incident Database
- AI Now Institute
- AI Verify Foundation
- Alan Turing Institute
- Algorithmic Justice League
- All Tech Is Human
- Aza Raskin
- Boston Dynamics
- Center for Human-Compatible AI
- Center for Humane Technology
- Figure AI
- Humane Intelligence
- Institute for Security and Technology
- ITU AI for Good
- Johns Hopkins Center for Health Security
- Leverhulme Centre for the Future of Intelligence
- Microsoft Research
- MIIT Technical Committee on Humanoid Robots and Embodied Intelligence Standardization
- Mila - Quebec AI Institute
- NTI | bio (Nuclear Threat Initiative)
- Oxford Institute for Ethics in AI
- Partnership on AI
- Physical Intelligence
- Stanford Institute for Human-Centered AI
- Tesla (Optimus)
- Thinking Machines Lab
- Transluce
- Tristan Harris
- UL Standards & Engagement
- Unitree Robotics
- Vector Institute
- Your Undivided Attention
Related questions
Related failure conditions
Discussed in
Contributions
No published contributions yet.