BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Targeted Disclosure to Support Auditing and Accountability for Aut
 omated Decision-making - Joshua Kroll
DTSTART:20171101T110000Z
DTEND:20171101T120000Z
UID:TALK94699@talks.cam.ac.uk
CONTACT:Adrian Weller
DESCRIPTION:Audits of automated decision-making systems promise to identif
 y problems with the correctness\, fairness\, and accountability of those s
 ystems. However\, audits alone cannot provide sufficient information to ve
 rify properties of interest. Audits are a type of black-box testing\, the 
 least powerful testing scenario available for computer systems\; even inte
 rnal audits may not have sufficient information to conclude whether or not
  a statement about a computer system is true. All auditing relies on at le
 ast some disclosure about the internals of the system under examination. T
 he evidence necessary to establish properties of interest for any particul
 ar system will depend strongly on the context of that system and its deplo
 yment. Disclosures need not rise to the level of full transparency\, but m
 ust only constitute evidence that a system satisfies properties of interes
 t. Further\, such evidence must be robust\, convincing\, and verifiable. I
 t must also tolerate underspecification of the task for which a system was
  designed and avoid lending credence to incorrect solutions or low-confide
 nce guesses. The purpose of evidence is to establish properties of interes
 t in the context of a particular system and its deployment\; explanations\
 , interpretations\, or justifications alone are not evidence of correctnes
 s\, robustness\, or any other property\, and should not be treated as such
 .\n\nThis talk describes necessary evidence and disclosures for effective 
 auditing and outlines practical steps and a research agenda in targeted\, 
 partial disclosure to facilitate accountability. It focuses on the require
 ments for understanding and governing software systems\, especially machin
 e-learning systems. Specifically\, it questions the value of human-interpr
 etable machine learning systems in fulfilling these requirements. Finally\
 , it outlines open research questions in the area of building human-govern
 able data-driven systems.
LOCATION:CBL Seminar Room
END:VEVENT
END:VCALENDAR