IEEE Technology and Society Magazine - June 2017 - 32

[3]. Other examples point to gender
bias such as Google's advertising
algorithm which appeared to show
higher paying jobs more to men
than women [4].

It is exceedingly rare that biased
algorithm behavior is shown to
have been intentional.
Of importance to note, however,
is that it is exceedingly rare that
biased algorithm behavior is shown
to have been intentional. Rather, as
highlighted in the May 2016 report
from the White House [5] such bias
is commonly attributable to poorly
selected data; incomplete, incorrect,
or outdated data; selection bias;
unintentional perpetuation and pro-
motion of historical biases [embed-
ded in the data]; poorly designed
matching systems; personaliza-
tion and recommendation services
that narrow instead of expand user
options; decision-making systems
that assume correlation necessarily
implies causation; or data sets that
lack information or disproportion-
ately represent certain populations.
Meanwhile, on the other side of
the Atlantic, the EU's General Data
Protection Regulation (GDPR), which
was adopted in 2016 and takes
effect in 2018, contains a recital (a
non-binding description of the law
written by its authors) stating that
"a data subject has the right to an
explanation of the decision reached
after [algorithmic] assessment" [6].
Details of how the GDPR will be
interpreted by the courts, and
thus the degree to which it provides

32

a "right to explanation" remain to
be seen.
These extensive concerns point
to a clear need for ethical de-
sign standards that can help ensure
that engineers, technolo-
gists, and the or g a n i z a -
t ion s t hey work for can
provide clarity around how
the algorithms they cre-
ate deal w ith issues of
bia s in producing and in
applying algorithms. Rec-
ognition of this need by
the industry and research
communities is clearly
shown by industry initiatives such
as the Partnership on AI [7] and
the recently launched Ethics and
Governance of AI fund [8] adminis-
tered by the Knight Foundation. It is
also shown by investment from gov-
ernment sources such as the U.K.
Engineering and Physical Science
Research Council's funding of the
UnBias project [9], an initiative I co-
lead in my role at the Horizon Digital
Economy Research Institute.
The IEEE Standards Project on
Algorithmic Bias Considerations is
designed to provide individuals or
organizations creating algorithms
with methods to provide clearly
articulated accountability and clar-
ity around how algorithms are tar-
geting, assessing, and influencing
the users and stakeholders affected
by the algorithm. Certification under
this standard will allow algorithm
creators to communicate to users
and regulatory authorities that up-
to-date best practices were used in
the design, testing, and evaluation
of the algorithm to avoid differential
impact on users that is not operation-
ally justified. The Working Group for
the Project is also open to anyone
who'd like to join by getting in touch
with IEEE-SA.

IEEE Technology and Society Magazine

When properly designed, algorith-
mic systems provide an opportunity
to help us counter existing unjusti-
fied human bias. To do so, however,
will require concerted effort. Other-
wise we run the risk that the algo-
rithms will instead entrench existing
bias. This Standards Project on Algo-
rithmic Bias Considerations is one
such effort.

Author Information
Ansgar Koene is with the Horizon
Digital Economy Research Institute,
University of Nottingham, U.K.

References
[1]  J. Angwin, J. Larson, S. Mattu and L.
Kirchner, Machine Bias, ProPublica, May 23,
2016, https://www.propublica.org/article/
machine-bias-risk-assessments-in-criminal-
sentencing.
[2]  J. Buolamwini, The Algorithmic Justice
League, "Unmasking bias," M.I.T. Media Lab, Dec.
14, 2016; https://medium.com/mit-media-lab/
the-algorithmic-justice-league-3cc4131c5148#.
senkns2it.
[3] S. Levin, "A beauty contest was judged by AI
and the robots didn't like dark skin," Guardian,
Sept. 8, 2016; https://www.theguardian.com/
technology/2016/sep/08/artificial-intelligence-
beauty-contest-doesnt-like-black-people
[4] J. Vincent, "Google's algorithms advertise
higher paying jobs to more men than women,"
The Verge, July 7, 2015; http://www.theverge
.com/2015/7/7/8905037/google-ad-
discrimination-adfisher.
[5] Big Data: Seizing Opportunities, Preserving Values, Executive Office of the Presi-
dent, May 2014; https://obamawhitehouse
.archives.gov/sites/default/files/docs/big_data_
privacy_report_may_1_2014.pdf.
[6] E. Chiel, "EU citizens might get a 'right to
explanation' about the decisions algorithms
make," Fusion, July 5, 2016; http://fusion.net/
s t o r y / 3 2 1 1 7 8 / e u ro p e a n - u n i o n - r i g h t - t o -
algorithmic-explanation/.
[7] Parnership on AI: to benefit people and
society, 2017; https://www.partnershiponai
.org/.
[8] "AI Fund FAQ,' Knight Foundation, 2017;
https://www.knightfoundation.org/aifund-faq.
[9] IEEE Standard for Algorithm Bias Consid-
erations," UNBIAS, Apr. 12, 2017; http://unbias
.wp.horizon.ac.uk/

∕

JUNE 2017
https://www.propublica.org/article/ https://www.medium.com/mit-media-lab/ https://www.theguardian.com/ http://www.theverge https://obamawhitehouse http://www.archives.gov/sites/default/files/docs/big_data_ http://www.fusion.net/ https://www.partnershiponai https://www.knightfoundation.org/aifund-faq http://unbias http://wp.horizon.ac.uk/
Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - June 2017
