Algorithmic transparency: Difference between revisions

Revision as of 06:58, 6 February 2017

Revision as of 15:44, 10 February 2017

Line 7:

'''Algorithmic transparency''' is the capacity for the user of an [[Algorithms|algorithm]] to understand its functioning and its resulting output. It is to be opposed to the functioning of an algorithm as a [[black box]], which lacks explainability in its automated decision making.<ref>{{cite journal|last1=Diakopoulos|first1=Nicholas|title=Algorithmic Accountability: Journalistic Investigation of Com- putational Power Structures.|journal=Digital Journalism|date=2015|volume=3|issue=3|page=398-415}}</ref>

Current research around algorithmic transparency is mainly interested in the societal effects of accessing remote services running black box algorithms.<ref>{{cite web|title=Workshop on Data and Algorithmic Transparency|url=http://datworkshop.org/|accessdate=4 January 2017|date=2015}}</ref> Some approaches propose ways to gain understanding about specific remote black box algorithms, by crafting inputs, via service [[APIs]], and observing the resulting output.<ref>{{cite journal|last1=Tramèr|first1=Florian|last2=Zhang|first2=Fan|last3=Juels|first3=Ari|last4=K. Reiter|first4=Michael|last5=Ristenpart|first5=Thomas|title=Stealing Machine Learning Models via Prediction APIs|journal=USENIX Security Symposium|date=2016|url=https://www.usenix.org/system/files/conference/usenixsecurity16/sec16_paper_tramer.pdf}}</ref> <ref>{{cite journal|last1=Le Merrer|first1=Erwan|last2=Trédan|first2=Gilles|title=Uncovering Influence Cookbooks : Reverse Engineering the Topological Impact in Peer Ranking Services|journal=Computer-Supported Cooperative Work and Social Computing|date=2017|url=https://arxiv.org/abs/1608.07481}}</ref>

==See also==