A FUTURE THAT WORKS:
AI, AUTOMATION, EMPLOYMENT, AND PRODUCTIVITY
JAMES MANYIKA
Extracts From McKinsey Global Institute Research, June 2017
CONFIDENTIAL AND PROPRIETARY
Any use of this material without specific permission of McKinsey & Company is strictly prohibited
Amazing progress in AI and Automation
McKinsey & Company 2
McKinsey & Company 3
2011 2016
Humans
26% errors 5% errors 3% errors
SOURCE: Jeff Dean (Google Brain) McKinsey & Company 4
McKinsey & Company 5
McKinsey & Company 6
Why Now?
Algorithms/techniques – Neural Networks, CNNs,
1 RNNs, Deep learning, Reinforcement Learning…
Compute power – Silicon (CPUs, GPUs, Tus …);
2 Hyperscale compute capacity, cloud available …
Data – 50 exabytes (2000), 300 exabytes (2007);
3 4.4 zettabytes (2013), 44 zettabytes (2020) …
McKinsey & Company 7
Huge benefits to business,
the economy and society
McKinsey & Company 8
Machine learning has broad potential across industries and use cases
Size of bubble indicates variety Agriculture Consumer Finance Manufacturing Pharmaceuticals Telecom
of data (number of data types)
Automotive Energy Health care Media Public/social Travel, transport,
and logistics
Volume
Breadth and frequency of data
10
Lower priority
Identify Personalize
Personalize
9 fraudulent financial
Higher potential advertising
transactions products
8 Identify and
Discover new navigate roads
Personalize crops to
7 individual conditions consumer trends
Optimize pricing
Predict personalized
6 and scheduling
health outcomes
in real time
Optimize
5
merchandising strategy
Predictive
4 maintenance
(energy)
Predictive maintenance Case by case
3
(manufacturing)
2 Diagnose diseases
1 Optimize clinical trials
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9
Impact score
McKinsey & Company 9
Good for business – Drives innovation, transformation and productivity
ACCURACY OPTIMIZATION
PREDICTION
SCALABILITY
THROUGHPUT CREATION
DISCOVERY
DECISIONS
McKinsey & Company 10
Good for the economy - Automation can contribute to growth in GDP per capita
FTE automation output (United States example, 2000–65)
FTEs
Millions
450 Automation will be a
significant contributor FTEs to achieve
to the productivity boost projected GDP growth
needed to projected FTE Automation output
360 GDP per capita growth in earliest scenario
FTE Automation output
270 in latest scenario
FTEs required to maintain
current GDP per capita
180
Projected FTE
Assuming zero productivity growth,
Historical FTE based on demographic trends, the
90
projected FTEs will be less than
the FTEs required to maintain
current level of GDP per capita
0
2000 05 10 15 20 25 30 35 40 45 50 55 60 65 2070
Year
McKinsey & Company 11
What about jobs?
McKinsey & Company 12
Our approach focuses on activities and capabilities of currently demonstrated technologies
Occupations Activities (retail example) Capability requirements
Retail sales- Greet Social
1 ▪ Social and emotional sensing
people customers
▪ Social and emotional reasoning
▪ Emotional and social output
Food and Answer questions about ▪ etc
2 beverage service products and services
workers Cognitive
▪ Natural language
Clean and maintain
▪ Recognizing known patterns / categories
work areas
▪ Generating novel patterns / categories
3 Teachers
▪ Logical reasoning / problem solving
Demonstrate product
▪ Optimizing and planning
features
▪ Creativity
Health ▪ Articulating/display output
4
practitioners ▪ Coordination with multiple agents
Process sales and ▪ etc
transactions
▪ ... Physical
▪ … ▪ ... ▪ Sensory perception
▪ … ▪ … ▪ Fine motor skills/dexterity
▪ … ▪ Gross motor skills
▪ Navigation
~2,000 activities assessed ▪ Mobility
~800 occupations across all occupations ▪ etc
SOURCE: Expert interviews; McKinsey analysis McKinsey & Company 13
BASED ON
Some activities have higher technical automation potential DEMONSTRATED
Time spent on activities that can be automated by adapting currently demonstrated technology TECHNOLOGY
%
81
69
64
26
18 20
9
Time spent
in all US 7 14 16 12 17 16 18
occupations
%
Manage Expertise Interface Unpredictable Collect Process
Process Predictable
Predictable
physical data data
data physical
physical
Total wages 596 1,190 896 504 1,030 931 766
in United
States, 2014
$ billion
Most 51% of US wages
susceptible
activities $2.7 trillion in wages
McKinsey & Company 14
Some sectors have more automatable activities than others BASED ON DEMONSTRATED TECHNOLOGY
Size of bubble indicates % of Ability to automate (%)
time spent in US occupations
0 50 100
Unpredictable Collect Process Predictable Automation potential
Sectors by activity type Manage Expertise Interface physical data data physical %
Accommodation and food services 73
Most automatable
Manufacturing 60
Transportation and warehousing 60
Agriculture 57
Retail trade 53
Mining 51
Other services 49
Construction 47
In the middle
Utilities 44
Wholesale trade 44
Finance and insurance 43
Arts, entertainment, and recreation 41
Real estate 40
Administrative 39
Least automatable
Health care and social assistances 36
Information 36
Professionals 35
Management 35
Educational services 27
McKinsey & Company 15
Employee weighted overall % of activities that can be
All countries could be impacted by automation automated by adapting currently demonstrated technologies
<45 45–47 47–49 49–51 >51 No data
Automatability across economies Million FTE
Employee weighted overall % of activities that can be automated $ trillion
Remaining
countries
China
100% =
1,156M FTEs
$14.6 trillion
Japan
Big 5 in Europe
United States
India
McKinsey & Company 16
A small percentage of occupations can be fully automated by adapting current technologies, but
almost all occupations have some activities that could be automated
100
91
73
62
51
% of roles 42
34
(100% = 26
18
820 roles) 8
1
100 >90 >80 >70 >60 >50 >40 >30 >20 >10 >0%
Percent of automation potential
Example Sewing machine Stock clerks Bus drivers Fashion designers Psychiatrists
occupations operators Travel agents Nursing assistants Chief executives Legislators
Assembly line workers Dental lab technicians Web developers
While about More occupations will have portions of their tasks automated e.g.
5% 60%
of occupations could have of occupations could have
close to 100% 30%
of tasks automated, of tasks automated
SOURCE: US Bureau of Labor Statistics; McKinsey Global Institute analysis McKinsey & Company 17
Automation potential spans from high to low wage occupations BASED ON DEMONSTRATED TECHNOLOGY
Ability to technically automate
Percentage of time on activities that can be automated
by adapting currently demonstrated technology
100
File
clerks
80
Landscaping and
grounds-keeping workers
60
Chief
executives
40
20
0
0 20 40 60 80 100 120
Hourly wage
$ per hour
McKinsey & Company 18
Several factors affect the pace and extent of AI and automation
Technical Cost of Cost of labor Benefits Regulatory
feasibility and developing and related including and social
pace of and supply- and beyond factors
breakthroughs deploying demand labor
technologies dynamics substitution
McKinsey & Company 19
In summary…
McKinsey & Company 20
We’ve seen this before—but is this time different?
Distribution of labor share by sector in the United States, 1840–2010
%
90
Rest of the economy
80
70
60
50
40
30
20
10 Manufacturing
Agriculture
0
1840 50 60 70 80 90 1900 10 20 30 40 50 60 70 80 90 2000 2010
McKinsey & Company 21
So with huge benefits, some real challenges to address
Benefits Challenges
Faster innovation and business Jobs and wages
transformation Skills and training
Better performance, outcomes, Social and Dislocation and
For quality, speed
businesses economic transitions
and users Overcome human limits; Solve Distributional issues
new problems, create new
opportunities and innovations
Acceptance
Safety, utility, quality of life
Boost productivity growth, Transparency, openness
GDP growth and prosperity and competition
For
economies
Counter aging or shrinking
Other issues Biases
workforce
and society Safety, Cybersecurity
Solve “moonshot” problems
(e.g., climate) Ethics
McKinsey & Company 22
