Submission 554
Mr Vincent Barbara
Submitter group: Individuals.
Position labels are not shown for individual submitters.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
Proposes SAFER, a five-part framework for autonomous AI: stop an objective and seek authorised human confirmation when principles conflict or a safe, lawful course cannot be established. Human confirmation cannot override protection of life or the law. Presents the framework as a starting point for examination and testing, rather than a complete solution.
Key points
Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.
Proposes autonomous AI systems should have a clearly defined stop-and-ask state.
As AI systems become more autonomous, I believe we should consider whether they should have a clearly defined "stop and ask" state.
Proposes AI systems should stop and request authorised human confirmation when principles conflict or safety is unclear.
If an AI encounters a conflict between these principles, or cannot determine whether its intended action is safe and lawful, it should stop the affected objective and request confirmation from an authorised human.
Says the framework proposes a defined stop state rather than letting AI choose which rule to ignore.
Rather than allowing an AI to make its own decision about which rule to ignore, the SAFER Principle proposes a clearly defined stop state.
Says human confirmation should not override requirements to protect life and obey the law.
human confirmation should not be capable of overriding the fundamental requirements to protect human life and obey the law.
Says the proposal is not a complete solution to AI safety.
I recognise that this proposal is not a complete solution to AI safety.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Have the SAFER framework examined, challenged and potentially tested by AI safety researchers, developers and policymakers.
I am asking whether this basic framework could be examined, challenged and potentially tested by AI safety researchers, developers and policymakers.
Consider whether autonomous AI should stop objectives and defer to authorised humans when rules conflict or a safe, lawful course cannot be established.
Should increasingly autonomous AI systems have a formally defined ability to stop pursuing an objective and defer to an authorised human when their rules conflict or the system cannot establish a safe and lawful course of action?