Mapping AI Ethics: from principles into practice

Abstract

The objective of this research proposal is to advance in closing the gap between practice and research in the area of AI/MACHINE ETHICS through a mapping of state of the art for Ethics in AI at different levels: macro, mezzo and micro: - "Macro" (regional, national and significant corporation guidelines), which affect the "purpose", type of applications that might be allowed to use (or ban) in public spaces, for example, where privacy and security issues are paramount. - "Meso" (applications/use cases,) Scenarios where there is a clear human-machine interaction: business decision making, recruitment, even autonomous cars, how is that hand-over of moral control between human-machine process managed? Is the person readily prepared for "assuming" control or taking the decision? Is the info for the decision "biased"? Can we expect a better decision? Which ethical aspects/framework have been considered? - "Micro" (algorithmic implementations). Which are the current implementations of ethical framework/theories in machine learning or alternative AI models? Where are the current limits? In terms of methodology, a literature review on the latest developments of AI/machine ethics at the three levels has been carried over.

The objective of this research proposal is to advance in closing the gap between practice and research in the area of AI/MACHINE ETHICS through a mapping of state of the art for Ethics in AI at different levels: macro, mezzo and micro: - "Macro" (regional, national and significant corporation guidelines), which affect the "purpose", type of applications that might be allowed to use (or ban) in public spaces, for example, where privacy and security issues are paramount. - "Meso" (applications/use cases,) Scenarios where there is a clear human-machine interaction: business decision making, recruitment, even autonomous cars, how is that hand-over of moral control between human-machine process managed? Is the person readily prepared for "assuming" control or taking the decision? Is the info for the decision "biased"? Can we expect a better decision? Which ethical aspects/framework have been considered? - "Micro" (algorithmic implementations). Which are the current implementations of ethical framework/theories in machine learning or alternative AI models? Where are the current limits? In terms of methodology, a literature review on the latest developments of AI/machine ethics at the three levels has been carried over.