Year: 2021
Type: review
Study design: Systematic Review
Abstract: This thesis provides a systematic literature review (SLR) of ethics-based AI auditing research. \nThe review’s main goals are to report the current status of AI auditing academic literature and \nprovide findings addressing the review objectives. The review incorporated 50 articles presenting \nethics-based AI auditing. The SLR findings indicate that the AI auditing field is still new and \nrising. Most of the studies were conference proceeding published either 2019 or 2020. Therefore, \nthere was a demand for a SLR work as the AI auditing field was wide and unorganized. \nBased on the SLR findings, fairness, transparency, non-maleficence and responsibility are the \nmost important principles for the ethics-based AI auditing. Other commonly identified principles \nwere privacy, beneficence, freedom and autonomy and trust. These principles were interpreted to \nbelong to either drivers or dimensions depending on whether something is audited directly or \nwhether achieving ethics is a desired outcome. \nThe findings also suggest that the external AI auditing leads the ethics-based AI auditing \ndiscussion. Majority of the papers dealt specifically with external AI auditing. The most important \nstakeholders were recognized to be researchers, developers and deployers, regulators, auditors, \nusers and individuals and companies. Roles of the stakeholders varied depending on whether they \nare proposed to conduct AI audits or whether they are in the position of beneficiary.
Source: UTUPub (University of Turku)
Author: Laine, Joakim
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7037868286
