CONFIDENTIAL AND PROPRIETARY

Any use of this material without specific permission of McKinsey & Company is strictly prohibited

JAMES MANYIKA

Extracts From McKinsey Global Institute Research, June 2017

A FUTURE THAT WORKS:AI, AUTOMATION, EMPLOYMENT, AND PRODUCTIVITY

2McKinsey & Company

Amazing progress in AI and Automation

3McKinsey & Company

4McKinsey & CompanySOURCE: Jeff Dean (Google Brain)

26% errors

2011 2016

3% errors

Humans

5% errors

5McKinsey & Company

6McKinsey & Company

7McKinsey & Company

Why Now?

Algorithms/techniques – Neural Networks, CNNs,

RNNs, Deep learning, Reinforcement Learning…1

Compute power – Silicon (CPUs, GPUs, Tus …);

Hyperscale compute capacity, cloud available …2

Data – 50 exabytes (2000), 300 exabytes (2007);

4.4 zettabytes (2013), 44 zettabytes (2020) …3

8McKinsey & Company

Huge benefits to business,

the economy and society

9McKinsey & Company

0 1.10.2 1.20.4 1.30.60.1

5

0.5

10

9

7

6

8

0.3 1.0

4

3

2

1

01.91.61.5 1.81.70.9 1.40.7 0.8

Optimize clinical trials

Personalize

advertising

Identify

fraudulent

transactions

Diagnose diseases

Predictive

maintenance

(energy)

VolumeBreadth and frequency of data

Impact score

Optimize pricing

and scheduling

in real time

Discover new

consumer trendsPersonalize crops to

individual conditions

Personalize

financial

products

Predict personalized

health outcomes

Identify and

navigate roads

Predictive maintenance

(manufacturing)

Optimize

merchandising strategy

Machine learning has broad potential across industries and use cases

Media

Consumer

Energy

Agriculture Manufacturing

Public/socialHealth careAutomotive

Finance

Travel, transport,

and logistics

TelecomPharmaceuticalsSize of bubble indicates variety

of data (number of data types)

Lower priority

Case by case

Higher potential

10McKinsey & Company

Good for business – Drives innovation, transformation and productivity

DISCOVERY

ACCURACY

THROUGHPUT

PREDICTION

CREATION

SCALABILITY

OPTIMIZATION

DECISIONS

11McKinsey & Company

360

270

65

450

2515 20102000 05 30

90

60454035 207050 55

180

0

FTEsMillions

Year

Projected FTE

FTEs required to maintain

current GDP per capita

Good for the economy - Automation can contribute to growth in GDP per capitaFTE automation output (United States example, 2000–65)

Historical FTEAssuming zero productivity growth,

based on demographic trends, the

projected FTEs will be less than

the FTEs required to maintain

current level of GDP per capita

FTEs to achieve

projected GDP growth

FTE Automation output

in earliest scenario

FTE Automation output

in latest scenario

Automation will be a

significant contributor

to the productivity boost

needed to projected

GDP per capita growth

12McKinsey & Company

What about jobs?

13McKinsey & Company

Our approach focuses on activities and capabilities of currently demonstrated technologies

SOURCE: Expert interviews; McKinsey analysis

Occupations

1Retail sales-

people

▪ ...

▪ …

▪ …

2

Food and

beverage service

workers

3 Teachers

4Health

practitioners

Activities (retail example)

Greet

customers

▪ ...

▪ …

▪ …

Answer questions about

products and services

Clean and maintain

work areas

Demonstrate product

features

Process sales and

transactions

~800 occupations

~2,000 activities assessed

across all occupations

Capability requirements

Social

▪ Social and emotional sensing

▪ Social and emotional reasoning

▪ Emotional and social output

▪ etc

Cognitive

▪ Natural language

▪ Recognizing known patterns / categories

▪ Generating novel patterns / categories

▪ Logical reasoning / problem solving

▪ Optimizing and planning

▪ Creativity

▪ Articulating/display output

▪ Coordination with multiple agents

▪ etc

Physical

▪ Sensory perception

▪ Fine motor skills/dexterity

▪ Gross motor skills

▪ Navigation

▪ Mobility

▪ etc

14McKinsey & Company

7 14 16 12 17 16 18

InterfaceExpertise Unpredictable

physical

Collect

data

Process

data

Predictable

physical

Manage

Some activities have higher technical automation potential

918 20

26

Time spent

in all US

occupations

%

Total wages

in United

States, 2014

$ billion

596 1,190 896 504

BASED ON

DEMONSTRATED

TECHNOLOGY Time spent on activities that can be automated by adapting currently demonstrated technology

%

1,030 931 766

6469

81

17 16 18

Predictable

physical

Collect

data

Process

data

51% of US wages

$2.7 trillion in wages

Most

susceptible

activities

15McKinsey & Company

Some sectors have more automatable activities than others

Automation potential

%Sectors by activity type

0 50 100

Ability to automate (%)Size of bubble indicates % of

time spent in US occupations

40

49

27

36

44

51

35

35

36

39

41

43

44

47

53

57

60

60

73

Mo

st

au

tom

ata

ble

Le

ast a

uto

mata

ble

In th

e m

idd

leBASED ON DEMONSTRATED TECHNOLOGY

Expertise InterfaceManageUnpredictable

physicalProcess

dataPredictable

physicalCollect

data

Accommodation and food services

Manufacturing

Agriculture

Transportation and warehousing

Retail trade

Mining

Other services

Construction

Utilities

Wholesale trade

Finance and insurance

Arts, entertainment, and recreation

Real estate

Administrative

Health care and social assistances

Information

Professionals

Management

Educational services

16McKinsey & Company

All countries could be impacted by automation

<45 45–47 47–49 49–51 >51 No data

Employee weighted overall % of activities that can be

automated by adapting currently demonstrated technologies

Automatability across economies

Employee weighted overall % of activities that can be automated

Million FTE

$ trillion

IndiaUnited States

Big 5 in Europe

Remaining

countries

Japan

China

100% =

1,156M FTEs$14.6 trillion

17McKinsey & Company

A small percentage of occupations can be fully automated by adapting current technologies, but

almost all occupations have some activities that could be automated

Example

occupations

10091

7362

5142

3426

188

1

>20>30>70 >50>80>90 >60100 >40

Percent of automation potential

% of roles

(100% =

820 roles)

>0%>10

SOURCE: US Bureau of Labor Statistics; McKinsey Global Institute analysis

Sewing machine

operators Assembly line workers

Stock clerks

Travel agents

Dental lab technicians

Bus drivers

Nursing assistants

Web developers

Fashion designers

Chief executives

Psychiatrists

Legislators

While about

5% of occupations could have

close to 100% of tasks automated,

More occupations will have portions of their tasks automated e.g.

60% of occupations could have

30% of tasks automated

18McKinsey & Company

Automation potential spans from high to low wage occupations

40

20

0

100

80

60

1200 10080604020

Hourly wage

$ per hour

Ability to technically automate

Percentage of time on activities that can be automated

by adapting currently demonstrated technology

File

clerks

Landscaping and

grounds-keeping workers

BASED ON DEMONSTRATED TECHNOLOGY

Chief

executives

19McKinsey & Company

Several factors affect the pace and extent of AI and automation

Cost of labor

and related

supply-

demand

dynamics

Benefits

including

and beyond

labor

substitution

Regulatory

and social

factors

Technical

feasibility and

pace of

breakthroughs

Cost of

developing

and

deploying

technologies

20McKinsey & Company

In summary…

21McKinsey & Company

Distribution of labor share by sector in the United States, 1840–2010

%

We’ve seen this before—but is this time different?

50 20100

1090 40

40

9020 3080 19001840 60

30

50

10

70

90

80

60

50

20

70 70 200060 80

Manufacturing

Rest of the economy

Agriculture

22McKinsey & Company

Benefits Challenges

So with huge benefits, some real challenges to address

Faster innovation and business

transformation

Better performance, outcomes,

quality, speed

Overcome human limits; Solve

new problems, create new

opportunities and innovations

Safety, utility, quality of life

For

businesses

and users

For

economies

and society

Boost productivity growth,

GDP growth and prosperity

Counter aging or shrinking

workforce

Solve “moonshot” problems

(e.g., climate)

Jobs and wages

Skills and training

Dislocation and

transitions

Distributional issues

Acceptance

Social and

economic

Transparency, openness

and competition

Biases

Safety, Cybersecurity

Ethics

Other issues