Christoph commented this

Hi, interesting paper. I wanted to point out two things: A.) There is paper specifically on quantifying sustainabili...

VK
Victoria replied this
Replied 2 days ago

Hi Christoph,   Thank you for your comment! Unfortunately, I was only recently notified about it - sorry for the late reply.   Thanks for pointing out the sustainability paper - it's nice to see a version of empowerment that takes reversibility into account. Do you know how it is defined in terms of mutual information?   I'm not sure how to extend the sustainable empowerment approach to settings where the agent is not the only source of change and the task may involve irreversible actions. I expect that maximizing sustainable empowerment in such settings would result in the agent trying to prevent other agents or environment events from making the starting state unreachable. I think it would be great to find a version of empowerment that works in this setting, e.g. penalizing reduction in empowerment relative to a default outcome in some way.    I agree that it's an interesting question whether it's generally desirable for the agent to interfere with irreversible events that are not caused by humans. There are other examples of irreversible events that would be bad to interfere with, e.g. we would not want the agent to stop the ocean currents or disrupt ecosystems. Even if some irreversible process is judged to be objectively bad, I don't think it's appropriate for the side effects penalty to penalize the agent for events that it did not cause, since the aim of the penalty is to prevent the agent from causing side effects. I would not consider allowing a vase to fall over and break as a side effect of the agent's actions, so I would not penalize that.    Cheers, Victoria

Views

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test
Test description