Information Science · AI Governance and Algorithmic Accountability
What must a person be able to find out
to contest an automated decision?
I’m Kerry Thornhill, an independent researcher and graduate student at the University of North Texas. I study what an affected party must be able to investigate in order to challenge an institution’s automated decision, and what the institution must preserve and disclose for that challenge to be real rather than nominal.
University of North Texas
Anuradha and Vikas Sinha College of Artificial Intelligence and Advanced Analytics
MS in Interdisciplinary Studies · In progress
Computer and Information Science and Engineering · Information Science
Specialization in AI governance and algorithmic accountability
Twenty-five years on
the institution’s side of the question.
I came to this research from the other side of the problem it describes.
I have run a food and over-the-counter pharmaceutical manufacturer under FDA oversight, a hazardous-materials motor carrier under FMCSA and PHMSA, and a Texas real estate brokerage under TREC. In each, I was the party who controlled the records, answered to the regulator, and decided what the people affected by our decisions would be told.
That experience is why I take the question seriously, and why I think most answers to it are too thin. Notice, an explanation, and an appeal form can all be present while a determination stays practically beyond challenge, because the institution holds the grounds and the affected person bears the burden.
Graduate study turned that recognition into a research program: a functional account of what meaningful contestability requires, built to be tested and revised.
The experience behind the questionsAn explanation is
only a beginning.
The Necessary Informational Conditions framework takes investigative capability, rather than transparency, explanation, or a human in the loop, as the measure of whether a decision is open to challenge.
Necessary informational conditions
Which informational operations must remain available to an affected party, and what must an institution preserve, for a material objection to be investigable?
02Institutional answerability and private power
When does a private institution’s use of algorithmic judgment generate a duty to keep that judgment open to informed challenge and consequential reconsideration?
03An empirical program
Preregistered omission and substitution tests, field measurement of institutional uptake, and a contestability toolkit tested for portability across credit, benefits, and health care.
03 / Experience with direction
A practical foundation.
A scholarly purpose.
I have led a brokerage of more than fifty agents, operated a regulated carrier across forty-eight states, and built the information systems those operations depended on. Alongside that record, a long practice in knowledge engineering: encoding expert underwriting judgment into rule-based screening systems, which is the same passage from governing standard to implemented rule that my research interrogates.
I’m pursuing a Master of Science in Interdisciplinary Studies at the University of North Texas, with research situated in Information Science and specializing in AI governance and algorithmic accountability. I intend to remain an independent researcher in this field.
Follow the trajectory04 / What the work is for
Research others can
examine and use.
The aim is a common framework for diagnosing failures of inquiry, comparing remedies, and designing procedures that others can test, extend, and refute.
Planned outputs include open research materials, documented methods, synthetic cases, and teaching modules in experimental design, qualitative coding, and open science, for researchers, students, and people who encounter AI-supported decisions.
I also direct the DG Humana Institute, a 501(c)(3) scientific and educational organization founded to publish and convene interdisciplinary scholarship on technology, dignity, and the future of work, and to serve as a home for scholars working on these questions.
Explore the research program