I was running a pilot. I wrote some attacks myself and tested the monitors. They missed a change request made at 10:00 hrs that broke the stadium at 19:00 hrs in 77 of 80 decisions. No superhuman model was involved - it needed domain knowledge and an ability to see the future implications of the changes being made.
The network is like a living organism - some problems arrive later, like a cascading impact of an incorrectly planned change.
Not sure whether this will be scalable oversight?
And my view - happy to get corrected, but in my 38 years - control is a... (read more)
I was running a pilot. I wrote some attacks myself and tested the monitors. They missed a change request made at 10:00 hrs that broke the stadium at 19:00 hrs in 77 of 80 decisions. No superhuman model was involved - it needed domain knowledge and an ability to see the future implications of the changes being made.
The network is like a living organism - some problems arrive later, like a cascading impact of an incorrectly planned change.
Not sure whether this will be scalable oversight?
And my view - happy to get corrected, but in my 38 years - control is a... (read more)