Accountable Artificial Intelligence: Holding Algorithms to Account

• Published date: 01 September 2021
• Author: Madalina Busuioc
• Date: 01 September 2021
• DOI: http://doi.org/10.1111/puar.13293

Research Article

825

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,

provided the original work is properly cited.

Abstract: Artificial intelligence (AI) algorithms govern in subtle yet fundamental ways the way we live and are

transforming our societies. The promise of efficient, low-cost, or “neutral” solutions harnessing the potential of big

data has led public bodies to adopt algorithmic systems in the provision of public services. As AI algorithms have

permeated high-stakes aspects of our public existence—from hiring and education decisions to the governmental use of

enforcement powers (policing) or liberty-restricting decisions (bail and sentencing)—this necessarily raises important

accountability questions: What accountability challenges do AI algorithmic systems bring with them, and how can we

safeguard accountability in algorithmic decision-making? Drawing on a decidedly public administration perspective,

and given the current challenges that have thus far become manifest in the field, we critically reflect on and map

out in a conceptually guided manner the implications of these systems, and the limitations they pose, for public

accountability.

Evidence for Practice

• The article provides public sector practitioners with insight into the distinct accountability challenges

associated with the use of AI systems in public sector decision-making.

• It digests and explicitly links technical discussions on black-box algorithms as well as explainable AI and

interpretable models—different approaches aimed at model understandability—to public accountability

considerations relevant for public bodies.

• It provides specific policy recommendations to securing algorithmic accountability—prominent among

these, the importance of giving preference to transparent, interpretable models in the public sector over

black-box alternatives (whether in a proprietary or in a technical sense, i.e., deep learning models).

• This will become critical to administrators’ ability to maintain oversight of system functioning as well as to

their ability to discharge their account-giving duties to citizens for algorithmic decision-making.

A

proprietary algorithm widely used by US

courts to predict recidivism in both bail

and sentencing decisions was flagged by

ProPublica as biased against black defendants (Angwin

et al. 2016); natural language processing (NLP)

algorithms for textual analysis can display recurrent

gender biases (Bolukbasi et al. 2016), for instance,

associating the word “doctor” with “father” and “nurse”

with “mother”; facial recognition algorithms have

persistently been found to display much higher error

rates for minorities (Buolamwini and Gebru 2018;

Lohr 2018; Snow 2018; Medium 2019), potentially

leading to false arrests and discrimination of already

marginalized groups when used in policing (e.g.,

Garvie and Frankle 2016); algorithms used for

university admissions to predict exam grades have

recently shown serious failures, with disparate negative

effects on high-achieving students from disadvantaged

backgrounds (Broussard 2020; Katwala 2020). These

are only a few of the growing number of examples of

bias encountered in algorithmic systems used not only

in private but also public sectors.

While simultaneously algorithmic systems based

on artificial intelligence (AI) are undoubtedly

associated with tremendous technological innovation,

and are foreshadowed “to supercharge the process

of discovery” (Appenzeller 2017), the examples

above underscore the importance of oversight of

AI algorithmic decision-making. As algorithmic

systems have become increasingly ubiquitous in

the public sector, they raise important concerns

about meaningful oversight and accountability

(Bullock 2019; Diakopoulos 2014; European

Parliament Study 2019; Pasquale 2015; Yeung 2018;

Yeung and Lodge 2019; Young, Bullock, and

Lecy 2019) and the need to identify and diagnose

where the potential for accountability deficits

associated with these systems might—first and

foremost—lie.

Madalina Busuioc

Leiden University

Accountable Artificial Intelligence: Holding Algorithms to

Account

Madalina Busuioc is Associate

Professor at the Institute of Public

Administration, Leiden University, where

she leads a large European Research

Council (ERC) grant investigating public

sector reputation and its effects within

the European regulatory state. She is

also incoming Fernand Braudel Senior

Fellow at the European University Institute

(EUI, Florence), awarded for a project on

“Accountable Artificial Intelligence in the

Administrative State”.

Email: e.m.busuioc@fgga.leidenuniv.nl

Public Administration Review,

Vol. 81, Iss. 5, pp. 825–836. © 2020 The Authors.

Public Administration Review

published

by Wiley Periodicals LLC on behalf of The

American Society for Public Administration.

DOI: 10.1111/puar.13293.