Paper Accepted: Towards Verifiably Ethical Robot Behaviour


I am pleased to inform you that your paper:

Towards Verifiably Ethical Robot Behaviour:

Has been accepted to be presented as a talk at the AAAI-15 workshop
on AI and Ethics.

We had a large number of submissions and the acceptance rate was
under 40%.


Nicely high level comments in the review as well which should be fairly easy to incorporate into the paper.

This is our attempt (together with Alan Winfield) to convert the work he describes here into a form were we can prove stuff about it. In short the previous work developed a framework that could integrate with whatever mechanism a robot was using to decide what to do and filter options down to those that were "most ethical" - the example used was a robot that normally would not collide with humans but, in the event that a human was about to walk into a hole in the ground, the robot would collide with the human in order to prevent them doing so. We converted this into a framework where we could prove that it was doing the filtering correctly in a very abstract description of the scenario and also generate some probabilistic results about likely outcomes in more concrete versions of the scenario.

We're also experimenting with something called Open Notebook Science and so have an Open Notebook for the work in the paper - I personally think the notebook is very dull but then the point isn't to be interesting so much as transparent.