Abstract

In recent years there has been much talk about the advancements in artificial intelligence (AI). Large strides have been made particularly in the area of machines learning algorithms and their application. This research investigates the premise that our AI algorithms have flaws, may be biased, or perhaps are simply incomplete. We seek to start a conversation around the need for an ethical audit framework for algorithmic development and deployment. We suggest that algorithmic applications go through what O’Neil (2016) calls an “algorithm audit”, specifically an ethical algorithm audit. This ethical AI auditing mechanism would provide some external validation of “doing no harm” and call out potential biases or flaws, based on industry or globally accepted best practice auditing procedures.

Share

COinS
 

Towards a Framework for Ethical Audits of AI Algorithms

In recent years there has been much talk about the advancements in artificial intelligence (AI). Large strides have been made particularly in the area of machines learning algorithms and their application. This research investigates the premise that our AI algorithms have flaws, may be biased, or perhaps are simply incomplete. We seek to start a conversation around the need for an ethical audit framework for algorithmic development and deployment. We suggest that algorithmic applications go through what O’Neil (2016) calls an “algorithm audit”, specifically an ethical algorithm audit. This ethical AI auditing mechanism would provide some external validation of “doing no harm” and call out potential biases or flaws, based on industry or globally accepted best practice auditing procedures.