Dr Bonnie Cheuk, AstraZeneca Digital Transformation & Global Capability Leader (Learning Culture and Learning Agility), delivered a keynote at IDC Future of Work Conference on 3 Mar 2020. She provoked the audience to go beyond the hype, and think deeper on how human and AI and data-driven Machine collaborate together.
These 3 questions were discussed:
1. How should human and machine collaborate? What skills are required?
2. Will machines replace (most) jobs?
3. Will there be new jobs to enable human-machine collaboration?
Drawing on Dervin's Sense-Making Methodology, Bonnie reminded us that human beings are not robotic machines. Human beings have feelings, experience, we are both scientists and artists, we are analytics and we are emotional.
Bonnie asked the audience how would you like to build a high performance team? Who do you want to put in the team? Do you want everyone to have the same strength, same skills? Or would you pick a team making up of players who can complement one another, and can bring out the best of one another. So in order to propose how human and machine should collaborate in the future of work, it is useful to first ask: what is the strength of human beings? What is the strength of the machine? We need to understanding how AI-driven machines learn vs how human beings learned, and play to one another's strength. And what is the strength of human? It is being human. Let the machine handle the deductive reasoning, the data-driven predictions, repetitive tasks. Let the humans do what we do well, adapting, navigating the unknown, use our human skills, promote collective sense making to make judgement, decisions. And free up the time to allow us to learn, create and innovate.
Bonnie highlighted that there are many unknowns as to how AI will be further developed, and there are ethical issues and risks that have to be addressed, and there are no precedents to follow. Collective human sense making is critical to bring out multiple perspectives from different stakeholders, to co-create AI-driven machines that human beings can trust, and to collectively address tricky ethical issues early on. Dervin’s Sense-Making Metaphor is introduced to facilitate two-way dialogue, to address power issues, and to explore common and divergent views to build common understanding of potential challenges, and co-create solutions to address them.
2. 2018 March
R&D Digital
Transformation
Leader
2018 Sept
R&D Digital
Transformation
Leader
Culture & Change
Management
Hello…. My name is Bonnie Cheuk
SOCIAL SCIENTIST RESEARCHER
PhD in Human Sense-Making
Author, social strategy in action:
driving digital transformation
Knowledge
Management
2020 Jan
AZ Global
Capability
Leader
Learning Agility &
Learning Culture
DIGITAL TRANSFORMATION
Knowledge Management, Enterprise2.0,
Digital Marketing & Comms, Innovation,
Customer Experience
WORK ACROSS INDUSTRIES
Consulting, Government,
Academia, Banking,
Pharmaceutical
/bonniecheuk
3. 3
Automation
• Explicit rules/processes
• Do better/faster
• Accurate output
• Digitalise the process
Data-Driven AI Machine Learning
• No need to understand how things work
• Give computer a set of data (as input)
• Give computer a set of results (as output)
• Ask the computer to figure out what goes on in between
• Can be a software, a robot, an algorithm
5. 5
Transforming
everything we
do for a digital
future.
Digital
technology
is…
…harnessing data to
develop better medicines,
faster, for the right patients
…accelerating clinical trials
and making them more
convenient for patients
…delivering medicines
more efficiently
…empowering patients
in a new world of
integrated health
6. Machines can read more radiology scans in one day than a
radiologist can see in a life time
15. 15
Machine Learning
Feed a million pictures of bicycle to a
computer
“Teach” the computer: this is a bicycle,
this is not a bicycle
After a while it “learns”
Show the computer any picture, and it
can tell whether it is a bicycle or not
4-year old Learning
Show 2 or 3 bicycle pictures to a
child
“Find all the other bicycles in this pile
of pictures”
Found unicycles, tricycles…
The child can generalise much
broadly than would seem possible
19. Intelligence
Reasoning
Inductive
(observe > generalise)
Deductive
(data > infer conclusion >
prediction)
Make Sense of the
World
Human skills
Empathy – Engagement –
Creativity – Perception –
Hunches – Emotions –
Relationship – Experience
Data Driven
Crunch data – complex
calculations for prediction &
reasoning – use some
measure of probability to
validate an argument
Handle Complexity
& Ambiguity
Flexible and Adaptive
Handle non-routine situation –
apply learning from one
situation into a totally different
context – navigate
unstructured physical space
Repeat in Predictable
environment
Assembly line work – Clerical
and keyboard tasks –
Navigate predictable physical
space (e.g. train run on track)
What Human Beings can do…
20. Intelligence
Reasoning
Inductive
(observe > generalise)
Deductive
(data > infer conclusion >
prediction)
Make Sense of the
World
Human skills
Empathy – Engagement –
Creativity – Perception –
Hunches – Emotions –
Relationship – Experience
Data Driven
Crunch data – complex
calculations for prediction &
reasoning – use some
measure of probability to
validate an argument
Handle Complexity
& Ambiguity
Flexible and Adaptive
Handle non-routine situation –
apply learning from one
situation into a totally different
context – navigate
unstructured physical space
Repeat in Predictable
environment
Assembly line work – Clerical
and keyboard tasks –
Navigate predictable physical
space (e.g. train run on tracks)
What Machine can do well…
21. “The future’s most valuable skills will be those that are
complementary to prediction — in other words, those
related to judgment.”
“Develop management processes that build the most
effective teams of judgement-focused humans AND
prediction-focused AI agents”
Source: What to expect from AI. MIT Sloan Management Review, July 2017
22. 22
• Human sense making – Reflecting, adapting,
navigating complexity & ambiguity
• Human judgement – Responsible decision making
(based on values, ethics, risk)
• Human (non-robotic) skills
• Perception, reasoning, creativity, innovation
• Communication, deep dialogue
• Empathy, emotional Intelligence
• Engage with customers, stakeholders, employees
• Read in between the lines
• Identify opportunities with limited data
• Human well-being, mental sharpness
Human skills provide added value in a mechanised world
24. “Oxford researchers have forecast that
machines might be able to do away half of all
U.S. jobs within two decades.”
The Atlanta: World without work (2015)
https://www.theatlantic.com/magazine/archive/2015/07/world-without-work/395294/
25. Less Automatable Activities → Highly Automatable Activities
It depends what tasks you perform in your job…
Source: Where machines can replace humans – and where they can’t yet
(McKinsey Digital Insights)
26. 26
Break down your job into specific tasks
Today
I spend my time
piecing together
information &
reworking the data
I create bespoke
spreadsheets for every
ask
Tomorrow
Machine
Machine
I share insight from
excellent financial
analysis & decision
making
I am keen to connect
with my colleagues to
help understand the
end to end process
Human
Human
The data in the
system is
never right for
us
My colleagues
respect me because
I’m the one who
knows the numbers
I am valued
because I share
the trends &
insights with my
colleagues
Finance will
succeed if we all
work better
together – my role
is just a part of
that
AI can support some tasks (e.g. data entry/processing, prediction, repetitive cognitive tasks, physical work in predictable
environment), but not all (social interaction, apply expertise, do physical work in unpredictable environment)
DoesThinks
27. 7
❑ Machine cannot eliminate jobs (but will replace tasks)
❑ Mid-level jobs will decrease, may downgrade to lower-paid jobs (lower-
paid jobs involve a lot of physical activities is harder to automate).
❑ Income gap between high and low-skills workers will increase
And… what do you do with the additional time?
Will Machines replace (most) jobs?
29. 9
AI ethics and risks: Jobs to Monitor?
Machine can do
❑ Filtering news
❑ Putting us in echo chambers
❑ Unfair matching of jobs
❑ Unfair matching of partners
We are worried
❑ What the machine might do?
We need a “pilot in the cockpit”
Human performing oversight
❑ Be part of the system operations
❑ Monitor it in case it misbehaviours
❑ Intervene and take actions
30. AI ethics and risks: Jobs to Regulate?
Regulate drugs or consumer products
❑ Safety standards
❑ Clear expectation on the function of
the product
❑ Who is liable, who is at fault if
something goes wrong
Regulate AI-driven Machines
❑ Not passive tools - some kind of
autonomy/Intentionality
❑ Adapt and learn - unable to pre-certify
machines as being ethical
Regulate AI is more like regulating human behaviours (not products)
We cannot
❑ Open people’s brain to certify they are
good people or will commit crime
We can
❑ Set clear expectation
❑ Observe behaviours in real world
❑ Hold them accountable if they violate
laws, social norms, company policies
We cannot easily regulate AI-driven Machines
❑ Machines have no experience, so we cannot
give them punishment or rewards
❑ Machines gain more autonomy as they learn
but do not gain more experience
We cannot
❑ Attribute liability to the programmer -
because the machine learn from the data
31. 1
❑ If the outcome of using AI
have an impact on other areas
❑ e.g. one stakeholder cares
about safety, another cares
about speed, costs, efficiency
❑ Can human beings make up
our minds and agree what the
machines should do for us,
before we know how human
beings should monitor it?
We need jobs to facilitate collective Sense-Making
(e.g. on moral dilemmas)
Jobs to facilitate collective Sense-Making
32. Jobs to facilitate collective Sense-Making
❑ Understand norms and expectations of different
stakeholders on AI
❑ Discuss how they perceive the expected
behaviours and anticipate surprises
❑ Be mindful of potential biases
❑ Anticipate the reaction when AI might go wrong
Need to bring
stakeholders
together to
have dialogue
33. Sense-Making Methodology Institute
http://www.sense-making.org
Sense-Making Toolkit:
Learning/Unlearning Moment
Q1: Where do I/we want to
achieve [with AI]?
Q2: What situation am I/are we facing?
What led up to this point?
Q3: What challenges, questions,
muddles do I/we have [with AI]?
How do I/we feel?
Q4: What is helping me/us?
What gets in the way?
Q7. What are the (small)
steps I/we can take to
develop [trust in AI]?
•
•
•
Q6. If I have a magic wand, I
would like ….
Collective Sense-Making to create self-awareness and trust [with AI]
Q5. Are there power issues
or constraints which prevent
me/us from speaking up?
34.
35. Jobs to explain the inner workings of complex algorithm to non-tech staff
Source: The Jobs that Artificial Intelligence Will Create, MIT Sloan Management Review, 2017
37. 37
How would human-machine collaborate when the future of AI is unknown?
❑ We know: More things are feasible using AI
❑ We don’t know: … how fast… how soon… how feasible?
38. 8
Human-Machine Collaboration
Machine
• Deductive reasoning
• Data-driven predictions
• Repetitive tasks in
predictive
environments
Human
Focus on the strategy & big picture
Human judgment focussed
Collective Sense Making
Human Skills
Lifelong Learning & Learning Agility
Work as a high performing team – play to our strength – Be Human!
39. 39
Learning Agility & Lifelong Learning Culture
Global Capability Leader, AstraZeneca
Credit to Dr Brenda Dervin for her
ongoing inspiration and guidance
@bonniecheuk/bonniecheuk http://www.sense-maker.org
Thank you. Let’s learn together!