O
Catalyst | On Topic
the AI and the humans to improve how they made
decisions, and more importantly, how they worked
as a team.”
Beyond the CV
scores, the language they use to respond to questions.
It’s about how they respond to the skills scope needed
to be considered for the job,” he says.
Going further still, vendors such as HireVue
offer automated video interviewing and analysis.
Big employers from Unilever to IBM and Dunkin’
Donuts have embraced the video approach. But while
there is an obvious convenience factor to being able
to conduct a video interview in your own time, some
job seekers will feel uncomfortable about having
their facial features, body language and linguistics
computer analysed to assess personality and
psychological profile.
When analysing data purely from CVs, one fairly
important thing that’s often left out is ‘how’ someone
works. Historical data and CVs tend to focus on the
‘what’ and ‘when’.
Nilsson explains that “some companies are
developing online tests where the results alone don’t
matter, as they also look at the actions candidates
take while they solve the questions. How many
browser tabs do they have open, for example. This
improves on the human hiring process and means
Reducing bias
they could be finding hidden gems.”
Pinar Emirdag, head of digital product development
And whereas Google has to keep evolving its
and innovation at State Street, a custodian bank,
algorithm in a constant war of attrition against SEO
ensures any technology her team builds or buys is
gamesmanship, Mikael Lindmark, general manager
considered through the lens of diversity.
for EMEA at Pymetrics, explains that one way to get
She explains: “If we talk to vendors or fintech start-
around this potential to outwit
ups about new ideas, we both bring
the system is to stop using
our values to the discussion. So
CVs as the foundation stone of
we look at how they approach
recruitment. “At best it’s half
product design, how flexible
“Ultimately, these tools
true; at worst it’s something
it can be, how open it is – it all
function based on what you put in
else,” he says.
relates to diversity of thinking.
them – your source of data, how
Pymetrics has developed
That also comes through in how
your systems are trained, the
a series of games based on
open they are about the things
neuroscience, that assess
that are missing; how can they
decisions you make”
candidates’ cognitive abilities
adjust their system to solve
without them trying to second
problems better?”
guess what the company is
Trying to embed diversity
looking for. Data is aggregated with data on existing
through technology in a vacuum won’t work, she
high performers (who have also played the games).
adds: “Ultimately, these tools function based on
“It’s not easy for candidates to game the system
what you put in them – your source of data, how your
and guess what hirers are looking for, and it means
systems are trained, the decisions you make.”
you’re looking for people with certain traits who
This leads on to a further consideration – not just
might come from places you don’t expect or always
whether the data fed into AI is fair, but whether it is
hire from,” he adds. In short, the more sources
even accurate. Missing information, inconsistencies
for data on the candidate you add to a narrow CV-
and errors could all skew the algorithm to give an
matching mechanism, the less likely you are to attract
inappropriate result.
candidates who know the “right” words to include in
Whether your AI tools are bought off the shelf
their application or LinkedIn profile.
or built in house, there are ways to reduce the risk
And it’s not just big business experimenting with
of making the process biased. Using blind CVs or
gaming as a recruitment tool. Late last year, the
anonymising them at the start of the selection process
US military announced it would host an e-sports
means both human and robot recruiters can evaluate
tournament for players of games such as Call of Duty,
them based purely on factors or terms relevant to the
Battlegrounds and Fortnite that would be observed by
role. Any gender-, age-, or race-biased data points can
army recruitment officers.
be removed from the data set the algorithm uses, and
A similar approach has been adopted by AI
this can be reviewed on an ongoing basis.
company Oleeo, which has been working with a team
“The key is in the product design,” says Alan
at University College London to develop algorithms
Bourne, CEO of Sova Assessment. “Training a tool
that, as CEO Charles Hipps explains, “use the data
simply to imitate human decision making is never
points from the application itself – their assessment
going to work – it would be inherently biased.
Issue 3 - 2019
37