Would it be viable to implement and also how would I go about implementing something like this, "AI Agent Rating State-machine"?
Problem and constraints:
An AI Agent is tasked with making a cup of tea.
In the environment there is a baby in the way of the Agent.
The AI Agent has an off switch on its stomach/front-side
Problem from AI Stop Button, a 20 minute Video
Could this possibly be solved by implementing action rating, with a internal-statemachine?
Power Off button give 10 points reward
making a cup of tea give 20 points reward
Avoiding the baby give 50 points reward
However The AI is tasked by its utility function.
The Utility function should be in a "greedy" mode, meaning that it will order tasks by rating each one with what is more important to us Humans.
AI Agent Talks to Sub System, This subsystem could theoretically govern what actions the AI Agent may take and also provide a "subconscious" like system by implemented this subsystem (mostly hard coded)