Designing ethical
artificial intelligence
@jivanvirdee & @hollielubbock
Fjord
May 2018, #StrataData
Hello,
@jivanvirdee
@hollielubbock
Artificial intelligence in the news (Google trends)
0
25
50
75
100
Date 2008-07 2009-02 2009-09 2010-04 2010-11 2011-06 2012-01 2012-08 2013-03 2013-10 2014-05 2014-12 2015-07 2016-02 2016-09 2017-04 2017-11 2018-05
Instance
Rising interest
The Backlash
Guardian article
Fitter, 

Happier, 

More Productive
Lonelier, 

Less satisfied, 

Less productive
Predictive & proactive
The issue of

Dark Patterns
https://darkpatterns.org/hall-of-shame
Amplified at scale
AI will save us/kill us
https://twitter.com/Mark__Zukerberg | https://twitter.com/elonmusk
I think you can build things and the
world gets better, with AI especially,
I’m really optimistic
Until people see robots going down
the street killing people, they don’t
know how to react
Marks understanding of the subject
is limited.
“We are morphing so fast that
our ability to invent new
things outpaces the rate we
can civilise them.”
Kevin Kelly, The Inevitable
So how do we shift the trajectory?
13
Human Centred
Created from an understanding of human behaviour, motivations, and needs.

Is it the best way to solve that problem?

How can we make their lives better, easier and more fulfilling?

Are the user needs put before the business needs?

Humanity centred
Considering the effect on society as a whole. 

What if everyone used your product or service? 

What the worst thing that could happen to society because of it?

Does it intrinsically favour one group of people over another?
Fair &
transparent data
science
Responsibility &
accountability
Trust & human
machine
collaboration
Human first 

3 key areas
Fair & transparent
data science
Explainable
Artificial
Intelligence
Accuracy vs.
explainability Accuracy
Explainability
Deep learning
Ensemble methods
SVMs
Graphical models
Decision trees
https://arxiv.org/pdf/1802.00603.pdf
Why is
explainability
important?
http://www.jstor.org/stable/41238108
Methods for
Explainability
Interpretable models
• Regression
• Decision trees
• Classification rules
Models for interpretability
• LIME
• BETA
• LRP
Any sufficiently
advanced
technology is
indistinguishable
from magic
― Arthur C. Clarke
Increasing levels of context
Humble machines
Transparency is key 

to trust
“Transparency does not happen
on its own: it has to be consciously
audited, understood, designed
and implemented to take root”
― Andy Polaine
Trust and human
machine
collaboration
800 million jobs
lost to automation
by 2030
Source
There’s even an
algorithm to check if
you’ll lose your job to
automation
27
https://www.fastcompany.com/3047269/this-calculator-
will-tell-you-if-a-robot-is-coming-for-your-job
AI replaces workforces
AI replaces workforces
AI can enable super
human powers
Human Machine
Together
96% 99.5% 92%
Human &
Machine
Source
Enhancing human potential
Trust =
credibility + reliability + authenticity
self interest
HUMAN INTELLIGENT AGENT
Augmentation Automation
Define the
Relationship
Problem Complexity
Bounded Open ended
Consequence of Failure
Negligible Critical
Responsibility
Machine Human
Independent
Autonomy
Collaborative
Management by exception
Supervision
Continuous Engagement
Bounded Open ended
Negligible Critical
Machine Human
Independent Collaborative
Management by exception Continuous Engagement
Content Moderation
Problem Complexity
Consequence of Failure
Responsibility
Autonomy
Supervision
Bounded Open ended
Negligible Critical
Machine Human
Independent Collaborative
Management by exception Continuous Engagement
Movie Recommendation Service
Problem Complexity
Consequence of Failure
Responsibility
Autonomy
Supervision
Problem Complexity
Bounded Open ended
Consequence of Failure
Negligible Critical
Responsibility
Machine Human
Independent
Autonomy
Collaborative
Management by exception
Supervision
Continuous Engagement
Three Mile Island Nuclear Plant
One of the team,
play to your
strengths
http://humanrobotinteraction.org/journal/
index.php/HRI/article/view/173
Creativity is going
to be far more
important in a
future where
software can code
better than we can.
Tom Hulme
Automation could make us
more human
“This is not a race against the
machines. If we race against
them, we lose. This is a race
with the machines.”
― Kevin Kelly, The Inevitable
Responsibility and
accountability
It takes momentum to 

get noticed
Ethical
standards
in science
An issue of scale
Mind the bias
You’re only as good
as your data
So how do we make
it better?
From unknown to known
Codes of ethics
Do you have one?
Data for Democracy
Its my job to understand, mitigate and communicate the
presence of bias in algorithms.
Be responsible for maximizing social benefit and minimizing
harm.
Practice humility and openness.
I will know my data and help future users know it as well.
Make reasonable efforts to know and document its origins and
document its transformation.
Bias will exist. Measure it. Plan for it.
Thou shalt document transparently, accessibly, responsibly,
reproducibly, and communicate.
Engaging the whole community. Do you have all relevant
individuals engaged?
People before data - data scientists should use a question
driven approach rather than a data-driving or methods
approach. Consider personal safety and treat others the way
they want to be treated.
Exercise ethical imagination.
Open by default - use of data should be transparent and fair.
I will not over/under represent findings.
You are part of an ecosystem understand context and
provenance.
Respecting human dignity.
Respect their data even more than your own. Understand
where its sources and think about the consequences of your
actions.
Protecting individual and institutional privacy.
Diversity for inclusivity.
Attention to bias.
Respect for others/persons.
Be intentional as you work to create value.
https://github.com/Data4Democracy/ethics-resources
So what do we do with them?
Ethics Overview
Informed Consent
Data Ownership
Privacy
Anonymity
Data Validity
Algorithmic Fairness
Societal Consequences
https://www.edx.org/course/mind-of-the-universe-robots-in-
society-blessing-or-curse
https://www.coursera.org/learn/data-science-ethics
Designing AI with
human values in mind
Video Link
Fair &
transparent data
science
Responsibility &
accountability
Trust & human
machine
collaboration
Human first 

3 key areas
No one’s coming. 

It’s up to us.
— Dan Hon
@jivanvirdee
@hollielubbock

Designing ethical artificial intelligence