Training Industry Magazine - March/April 2021 - 13

STELLA LEE, PH.D.

WHAT'S NEXT
IN TECH

MITIGATING BIAS:
THE ROLE OF AI IN D&I

Artificial intelligence (AI) has entered our
daily lives, from self-driving cars to writing
our emails and detecting fraud. Across
many organizations, AI applications are
being piloted to help automate and
streamline business processes, such as
recruitment, onboarding, training and
performance appraisals. However, AI's
role in workplace diversity and inclusion
has garnered much controversy.
On one hand, AI has the potential
to mitigate biases and provide more
equitable access to the job market.
For example, AI can be used in the
recruitment process to screen resumes
and identify the right candidates from a
large pool of applicants. Proponents of
AI promise that AI speeds up the hiring
process and helps reduce unconscious
bias by concealing candidates' names
and bypassing gender and racial
references in resumes. There are
even applications that can analyze a
candidate's facial expressions during
interviews to gain insights on their
personalities and potential fit for
the company.
On the other hand, AI is at great risk of
perpetuating inequalities and amplifying
stereotypes. Recent headlines in the
news showcase examples of how AI
technologies have learned to give
preference to male applicants and, in
one case, penalized resumes with the
word " women's " in them.
In order to ensure AI supports rather than
suppresses equity in the workplace, we
need to know where these biases stem
from and what we can do to prevent
them. Let's examine three common types
of bias that can impact AI applications:

BIAS IN DATA
For AI to work effectively, we need to feed
the system a lot of data. For example, for
a face-scanning mechanism to determine
which applicants are the " best fit, " the
system must be based on what has been
successful in the past. In the case of
executive-level professionals, this past
dataset likely consists of predominantly
white middle-age males. If we are
unaware of this bias while introducing
that dataset to the system, the system's
preference will skew toward this
demographic group. Organizations can
mitigate this issue by proactively seeking
greater representation and more varied
sample datasets.

EVEN WHEN INPUT
DATA IS UNBIASED,
ALGORITHMS CAN
PRODUCE BIASED
OUTCOMES.
ALGORITHMIC BIAS
Even when input data is unbiased,
algorithms
can
produce
biased
outcomes. For example, in learning
experience platforms (LXPs), learning
content and activities are recommended
to users based on attributes such
as prior knowledge, test scores,
demographic information, locations and
learning preferences. As learners act on
recommendations by selecting content,
these actions are added to the system
as a positive feedback loop. Over time,
these recommendations reinforce bias
coming from a certain group of users
- perhaps younger workers who use

the system more frequently. As a
result, more content is recommended
to all learners based on one group's
preference rather than the workforce's
diverse learning needs. One potential
solution to this problem is to audit
system activities regularly and increase
diversity in recommendations.
HUMAN BIAS
Believe it or not, a lot of manual work is
required to create an AI system. Human
inputs are needed to determine where
to source and select data, what data
to source and select, what is deemed
important to measure, as well as how to
interpret and present outcomes. These
decisions are made by people - and
not always by a heterogeneous group
of people with diverse perspectives. To
mitigate human bias, we need to think
carefully about our intentions when
configuring systems and be transparent
about our processes. Start with clear
documentation and communications to
reveal any assumptions and explain the
reasoning behind decision-making.
AI applications can harm our employees
and our organizations if they are not
designed, engineered and audited
properly. As L&D professionals, we should
be deliberate and responsible in shaping
how AI is used in the workplace.
Dr. Stella Lee has over 20 years of
experience in consulting, planning,
designing, implementing and measuring
learning initiatives. Her focus is on largescale learning projects including LMS
evaluation and implementation, learning
analytics, and artificial intelligent
applications. Email Stella.

T R A I N I N G I N DUSTR Y MAGAZ INE - BUILDING AN EQUITABLE W ORKP L AC E I WWW. T RAI NINGINDU S T RY . C OM/ MAGAZ I NE

| 13

https://www.trainingindustry.com/glossary/artificial-intelligence/ https://www.hbr.org/2019/10/using-ai-to-eliminate-bias-from-hiring https://www.hbr.org/2019/10/using-ai-to-eliminate-bias-from-hiring https://www.hbr.org/2019/10/using-ai-to-eliminate-bias-from-hiring https://www.hbr.org/2019/10/using-ai-to-eliminate-bias-from-hiring https://www.washingtonpost.com/technology/2019/10/22/ai-hiring-face-scanning-algorithm-increasingly-decides-whether-you-deserve-job/ https://www.washingtonpost.com/technology/2019/10/22/ai-hiring-face-scanning-algorithm-increasingly-decides-whether-you-deserve-job/ https://www.washingtonpost.com/technology/2019/10/22/ai-hiring-face-scanning-algorithm-increasingly-decides-whether-you-deserve-job/ https://www.trainingindustry.com/glossary/learning-experience-platform/ https://www.trainingindustry.com/glossary/learning-experience-platform/ https://www.reuters.com/article/us-amazon-com-jobs-automation-insight/amazon-scraps-secret-ai-recruiting-tool-that-showed-bias-against-women-idUSKCN1MK08G https://www.reuters.com/article/us-amazon-com-jobs-automation-insight/amazon-scraps-secret-ai-recruiting-tool-that-showed-bias-against-women-idUSKCN1MK08G https://www.trainingindustry.com/magazine

Training Industry Magazine - March/April 2021

Table of Contents for the Digital Edition of Training Industry Magazine - March/April 2021