TECHNOLOGY REVOLUTION

EXPLAINED
A GUIDE FOR EXECUTIVES

Ray Eitel-Porter, Managing Director,

Accenture Applied Intelligence Lead UKI
 TECHNOLOGY REVOLUTION

CONTENT

ARTIFICIAL INTELLIGENCE | 2
 TECHNOLOGY REVOLUTION

04 THE SINGLE BIGGEST TECHNOLOGY

REVOLUTION THE WORLD HAS EVER
SEEN.

10 SO WHAT IS AI ANYWAY?

34
WAIT, HAVEN’T WE BEEN HERE
BEFORE?

44 OK, I’M SOLD. BUT WHAT DO I

ACTUALLY NEED TO DO NOW?

66 HOLD ON, WE HAVEN’T TALKED ABOUT

THE RISKS INVOLVED

74 AND WHAT ABOUT THE BIGGER

PICTURE?

84 IT’S TIME TO GET SMART AND GET

STARTED

ARTIFICIAL INTELLIGENCE | 3
 TECHNOLOGY REVOLUTION

THE SINGLE BIGGEST

TECHNOLOGY
REVOLUTION
THE WORLD HAS EVER
SEEN

ARTIFICIAL INTELLIGENCE | 4
 TECHNOLOGY
TECHNOLOGYREVOLUTION
REVOLUTION

Everyone is talking about artificial intelligence

(AI). From boardrooms to factory floors,
from call centres to logistics fleets, and from
governments to venture capitalists, artificial
intelligence is suddenly the hottest topic in
town.

But is AI anything more than the latest

technology buzzword? The answer is yes.
In fact, AI might just be the single biggest
technology revolution the world has ever seen.

This guide will explain why.

ARTIFICIAL INTELLIGENCE| | 5
ARTIFICIALINTELLIGENCE 5
 TECHNOLOGY REVOLUTION
TECHNOLOGY REVOLUTION

Over the past twenty years or so, business

has faced its fair share of disruption. Indeed,
digital disruption is thought to have put more
than half of Fortune 500 companies out of
business since 2000.1 And now AI is set to
compound that disruption by shifting it up to
the next gear.

That’s because AI is what economists call

a general-purpose technology. And these
general-purpose technologies are a big
deal: think electricity and the internal
combustion engine. Their significance
lies in the fact that they cause disruption
not only through direct contribution to
society, but also through the way their
spill-over effects enable a vast range of
complementary innovations. Electricity made
possible factory electrification, telegraphic
communication, and all that followed. And
the internal combustion engine gave rise to
the automobile, the aeroplane, and modern
transportation and logistics networks. AI will
impact society on a similar scale.

6 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 6

 TECHNOLOGY
TECHNOLOGYREVOLUTION
REVOLUTION

AI has become viable today thanks to

the combinatorial effect of a series of
fast-moving technology trends (see
page 39 for the evolution of AI). Those
trends are lowering barriers to entry
across a wide range of industries. A new
wave of AI-first businesses is flowing
into the market, shifting the competitive
landscape for incumbents. These new
businesses are nimbler, unencumbered by
legacy technology systems, distribution
channels, or the need for workforce
transformations.

AI is the ultimate breakthrough

technology –

Satya Nadella, CEO, Microsoft

ARTIFICIAL INTELLIGENCE | | 7
ARTIFICIALINTELLIGENCE
 TECHNOLOGY REVOLUTION
TECHNOLOGY REVOLUTION

Private investment is booming. Venture

capital investment in AI for the first nine
months of 2017 totalled $7.6 billion2 (compared
with $5.4 billion for the whole of 2016). The
registration of AI patents is at an all-time high,
five times the figure in 2006.3 And in the US, the
number of AI start-ups has increased by twenty
times in just four years.

Governments and academic institutions are also

looking to ensure their economies maximise the
benefits of the technology. In the UK, a major
government-commissioned report estimated
AI could unlock £630 billion for the country’s
economy.4 Five leading UK universities have
also come together to create the Alan Turing
Institute, the UK’s national institute for data
science and artificial intelligence. And
a new £30 million National Innovation
Centre for Data is being established at
Newcastle University, funded by both
government and academia.

It all adds up to a fast-paced, ever-

shifting competitive, investment,
and research landscape. The
bottom line: AI is here, and every
executive must sit up and take
notice.

8 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 8

TECHNOLOGY
TECHNOLOGYREVOLUTION
REVOLUTION

By 2020, the AI
market will surpass
$40 billion5 –

Constellation Research

ARTIFICIAL INTELLIGENCE | 9
 WHAT IS AI

SO WHAT IS
AI
ANYWAY?

10 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI WHAT IS AI

So you’ve heard all about AI, and you know it’s

a big deal. But what is it exactly? Answering
that question isn’t as straightforward as it
might seem. In fact, there’s no single accepted
definition of “artificial intelligence”. And that’s
because AI as we know it isn’t really a technology
in its own right at all.

In reality, it’s a collection of different

technologies that can be brought together to
enable machines to act with what appears to be
human-like levels of intelligence.

11 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 11

 WHAT IS AI

Rather than add to the growing list of attempts to definitively describe

AI, we prefer to think of the technology as a framework of capabilities.
This is undoubtedly the best way to understand what AI is, and to get
a sense of the technologies that underlie it. Our framework is centred
around the principal things that AI enables a machine to do. There
are four:

SENSE. AI lets a machine perceive the world

around it by acquiring and processing images,
sounds, speech, text, and other data.

COMPREHEND. AI enables a machine to

understand the information it collects by
recognising patterns. Much as humans
interpret information by understanding the
patterns presented and their context, though it
does not derive true “meaning”.

ACT. AI enables a machine to take actions in

the physical or digital world based on that
comprehension.

LEARN. AI enables a machine to continuously

optimise its performance by learning from the
success or failure of those actions.

12 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI WHAT IS AI

WHAT’S IN A NAME?
COGNITIVE COMPUTING
AND AI
Cognitive computing is a term widely used by AI
practitioners. So what is it? And how is it different
from AI? Unfortunately, just as AI has no widely
accepted definition, cognitive computing can
mean different things to different people. That said,
“cognitive” in this sense can, for the most part, be
treated as referring to an AI’s perceptive capabilities - an
AI’s ability to sense and comprehend its environment.

13 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 13

 WHAT IS AI

The power of machine learning

An AI’s ability to learn is fundamental. Indeed, being able to decide
which actions are required to complete a task by analysing data, rather
than being explicitly coded to act in a pre-defined way, is arguably
what makes a system “intelligent” and differentiates AI from other
forms of automation.

And when the best AI systems are set the task of learning for
themselves, the results can be extraordinary. AlphaGo, the AI
developed by Google DeepMind, became the first computer
program to defeat a professional human player at the
highly-complex board game Go.6 AlphaGo
was taught the rules of play, and then shown
thousands of different human vs. human
games so that it could discern the winning
strategies by itself. The result: victory over
the legendary world Go champion, Lee Sedol.

But even that wasn’t the end of DeepMind’s

Go success. The company subsequently
developed a second, even more powerful,
version of AlphaGo - AlphaGo Zero7,
which taught itself winning strategies
simply by playing games against itself
– with no need to observe human players
at all. Moreover, the latest iteration of
the AI, AlphaZero, has gone even further.
AlphaZero proved it could learn chess by

14 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI

playing games against itself, surpassing human levels of skill in just

four hours. The really interesting part of this feat was that AlphaZero
wasn’t specifically designed to play chess at all. Indeed, Jonathan
Schaeffer8, professor of computer science at the University of Alberta,
and an expert in chess systems, believes this may be the very reason
it has been able to develop unconventional strategies for winning. In
this way, AlphaZero represents an important step away from narrow
AI towards general AI (for an explanation of the difference between
narrow and general AI, see page 25).

This is what we call machine learning. And the reason

it’s so powerful, as Brynjolfsson and
McAfee have observed9, is quite simple.
While we humans are fantastically skilled
at performing any number of different
activities, we don’t always know exactly
how we do what we do. So, for example, we
might find recognising another person’s face
very easy. But we don’t fully understand the
precise physiological mechanisms that let it
happen. And that makes it very difficult to
directly code the capability into a machine.

Machine learning, on the other hand,

lets a machine learn to do it all by itself.
Indeed, one of the core strengths of
machine learning is identifying patterns in
very large amounts of data.

15 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 15

 WHAT IS AI

On Wall Street today, more

than 60% of all trades are
executed by AI with little or
no real-time oversight from
humans –

Christopher Steiner,
Automate This

16 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI

A constellation of technologies
Machine learning lies at the core of AI systems. Its capability to learn from
raw data powers the visible manifestations of AI that are becoming ever
more prevalent today. So, whether it’s predictive systems that can forecast
what’s likely to happen, natural language processing that can comprehend
speech and text in close to real time, machine vision that can understand
visual inputs with extraordinary accuracy, or optimising search and
information retrieval, it’s all based on machine learning.

Predictive Systems

Search and Natural Language

Optimisation Understanding

Knowledge MACHINE Expert

Representation LEARNING Systems

Information Machine
Retrieval Vision

Figure 1 – Machine
Robotics
learning capabilities

ARTIFICIAL INTELLIGENCE | 17
 WHAT IS AI

One critical advantage machine learning has over other techniques is

its tolerance of “dirty” data. That is, data containing duplicate records,
badly parsed fields, or incomplete, incorrect, or outdated information.
These issues are a significant problem for businesses: most executives
will recognise all too well that dealing with dirty data can be the bane
of their professional lives.

Machine learning’s flexibility – it’s ability to learn and improve over

time – means dirty data can be processed with far greater accuracy.
It also means that the technology scales very well, something that
becomes ever more important in our current age of exploding data
volumes.

More than 85%

of customer
interactions will
be managed
without a human
by 202010 –
Gartner

18 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI

The different ways a machine can learn

One of the real strengths of machine learning is that there are different
types of learning algorithm it can use, including supervised, unsupervised,
and reinforcement.

SUPERVISED LEARNING. This kind of algorithm takes a labelled data

set (data that has been organised and described), deduces the salient
features that characterise each label, and learns to recognise those
features in new data. So, for example, you might show the algorithm a
large number of labelled images of cats, and it would then learn how to
recognise a cat and spot one in any number of other, completely different
pictures.

UNSUPERVISED LEARNING. This kind of algorithm requires no pre-

defined labels in the data it uses. It takes an unlabelled data set, finds
similarities and anomalies between different entries within that data set,
and categorises them into its own groupings. So, you might show the
algorithm a large number of unlabelled images containing, say, cats and
dogs, and it would sort images with similar characteristics into different
groups without knowing that one contained “cats” and the other “dogs”.

REINFORCEMENT LEARNING. This kind of algorithm works by trial and

error, using a feedback loop of “rewards” and “punishments”. So, when
the algorithm is fed a data set, it treats the environment like a game, and is
told whether it has won or lost each time it performs an action. This way,
it builds up a picture of the “moves” that result in success, and those that
don’t. DeepMind’s AlphaGo and AlphaZero (page 14) are good examples of
the power of reinforcement learning.

ARTIFICIAL INTELLIGENCE | 19
 WHAT IS AI

Figure 2 – Puppy or bagel?

20 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI WHAT IS AI

PUPPY OR BAGEL?
How hard is it to tell an animal from an item of food?
Sometimes, much more difficult than you might
think. One of the big trends storming the internet is
all about the odd similarities between certain pets
and snacks. Take the puppies and bagels in the
image opposite, for instance. At first glance, it can
be surprisingly challenging for a human to tell which
is which. Not so for an AI. Pass the images through
an image recognition API and you’ll find the AI can
distinguish the food from the pets with impressive
accuracy.11

21 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 21

 WHAT IS AI

The artificial brain

So, how does machine learning actually work? Sitting behind many of
the extraordinary advances in recent years lies a very advanced and
elegant form of computing system – one inspired by the functioning of
the animal brain itself. These systems are called neural networks, and
they underpin much of today’s cutting-edge work in AI.

A neural network comprises an interconnected set of “nodes” which

mimic the network of neurons in a biological brain. Each node
receives an input, changes its internal state, and produces an output
accordingly. That output then forms the input for other nodes, and
so on. This complex arrangement enables a very powerful form of
computing called deep learning.

Deep learning uses multiple layers of filters to learn about the

significant features of data in a data set. It’s used, for example, in both
image and speech recognition. Using a neural network, the output of
each filter provides the input for the next, where each filter operates at
a different level of abstraction. In this way, deep learning systems can
handle much larger data sets than alternative approaches.

22 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI WHAT IS AI

Facial recognition
Deep learning neural networks use layers of increasingly complex rules
to categorise complicated shapes such as faces12
Layer 1:
The computer
identifies pixels
of light and dark.
Layer 2:
The computer
learns to identify
edges and
simple shapes.

Layer 3:
The computer
learns to identify
more complex
shapes and
objects. FACES CARS

Layer 4:
The computer
learns which
Figure 3 - How a neural network
shapes and
recognises objects
objects can be
used to define a
human face.

23 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 23

 WHAT
WHAT IS
IS AI
AI

WHAT KIND OF
INTELLIGENCE ARE WE
TALKING ABOUT?
When data scientists and others talk about AI, they
often use two categorisations to clarify their meaning.
These are narrow AI vs. general AI and weak AI vs.
strong AI.

24 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI WHAT IS AI

WEAK AI STRONG AI
This describes “simulated” thinking. This describes “actual” thinking.
That is, a system which appears to That is, behaving intelligently,
behave intelligently, but doesn’t thinking as a human does, with
have any kind of consciousness a conscious, subjective mind.
about what it’s doing. For example, For example, when two humans
a chatbot might appear to hold a converse, they most likely know
natural conversation, but it has no exactly who they are, what they’re
sense of who it is or why it’s talking doing, and why.
to you.
NARROW AI GENERAL AI
This describes an AI that is limited This describes an AI which can
to a single task or a set number of be used to complete a wide
tasks. For example, the capabilities range of tasks in a wide range of
of IBM’s Deep Blue, the chess- environments. As such, it’s much
playing computer that beat world closer to human intelligence.
champion Gary Kasparov in 1997, Google DeepMind used
were limited to playing chess. It reinforcement learning to develop
wouldn’t have been able to win a an AI that learned to play a whole
game of tic-tac-toe – or even know range of different games requiring
how to play. different skills. The AI achieved
human-like levels of performance at
29 classic Atari video games using
only the on-screen pixels as its data
input.13
SUPERINTELLIGENCE

The term “superintelligence” is often used to refer to general and strong AI

at the point at which it surpasses human intelligence, if it ever does.

25 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 25

 WHAT IS AI

Instead of trying to produce

a programme to simulate the
adult mind, why not rather
try to produce one which
simulates the child’s?
If this were then
subjected to an
appropriate course
of education one
would obtain the
adult brain –

Alan Turing, 1950

26 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI WHAT IS AI

Time to train
The “learning” part of a machine learning process is, perhaps
unsurprisingly, critical to the whole concept. Much as a human brain
must learn throughout childhood to understand and process
the information it receives, so must a machine learning
algorithm or model be trained to comprehend its
environment.

When companies get the training wrong, the

results can be embarrassing – or worse.
Microsoft’s now infamous chatbot Tay 14
was an experiment in machine learning
through social media interactions. The AI
was designed to learn to hold a natural-
sounding conversation by speaking
with other Twitter users. But she had
to be quickly decommissioned when
a collection of trolls and racially
biased comments capitalised on her
lack of filters and taught her a series
of racial slurs and white-supremacist
propaganda. Microsoft’s experience
starkly highlights the need for
strong governance and controls in
deploying AI systems (for more on the
governance questions that AI raises,
see page 71).

27 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 27

 WHAT IS AI

Getting the training right takes more than just advanced maths.
Industrialised machine learning is an interdisciplinary capability.
It takes a blend of data science, engineering, and user experience
design with relevant domain knowledge. None of these capabilities
on their own will suffice.

Data Science is Big Data and

necessary but not technology alone does
sufficient not get you there

MATHS ENGINEERING
Statistics Software Engineering
Linear and Discrete Parallel and Distributed
Algebra Computing
Functional Analysis API Management
Optimisation Model Management
Algorithmic Complexity Large scale Data
Predicate Calculus Management

EXPERIENCE
Augmented User
Experience Industrialised AI is
Domain Expertise an interdisciplinary
Experience design Information Architecture capability
by itself is a lot of Visualisation
day dreaming

Figure 4 – Industrialised AI as an interdisciplinary capability

28 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI

Fitting analytics into the picture

So where does analytics fit into all this? Perhaps the question should
be: how does AI fit into analytics? If we think of analytics as the field of
analysing data to improve decision making, we can see how machine
learning, along with other statistical analyses, plugs in to the process.
After all, the goal of analytics is to derive insights from data – which is
much the same as the goal of machine learning.

Analytics, and any machine learning algorithms that support it, can
have different levels of sophistication depending on the degree of
insight required. So, at the simpler end of the scale are so-called
“descriptive analytics” which analyse historical data to understand
what happened and why. Then come “predictive analytics” which use
data to predict what will happen in the future. Finally, at the far end of
the scale, come “prescriptive analytics” which not only forecast what
will happen, but tell you what you need to do about it.

By 2020, insights-driven businesses will

take $1.2 trillion per annum from their
less-informed peers15 –
Forrester

ARTIFICIAL INTELLIGENCE | 29
 WHAT IS AI

AI, robot
When it comes to thinking about AI, robots and robotics are often front
of mind. In the public imagination, that can mean anything from Kubrick’s
HAL, to Asimov’s mechanical men, to Honda’s Asimo.

In a business context, it can mean both the automation of manufacturing

or service processes using mechanical robots – think car assembly lines
– and, increasingly, the automation of administrative or service processes
comprising both digital and manual inputs using Robotic Process
Automation (RPA).

Strictly speaking, because RPA is designed for processes that never vary,
it doesn’t require any “intelligence” at all. So, for example, if a business
process involves a person manually transferring data in a standard
form from one system (for example a piece of paper) to another, the
process can be easily automated through RPA with a form of keystroke
emulation. Accenture applied RPA to a large manufacturing client’s invoice
processing. The result: a 70% elapsed time saving; a 30% productivity
benefit; and 100% accuracy.

That said, AI techniques are now being increasingly used in the emerging
field of “cognitive RPA”. This enables a process with a degree of variation
to be automated, and thus vastly increases the scope of RPA. That can
include, for example, using machine learning to train a machine to
recognise text in an image (known as optical character recognition).

30 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI WHAT IS AI

THE RISE OF RPA

Based on a recent report by Transparency Market
Research, RPA is expected to see a compounded
annual growth rate of about 60.5% worldwide
through 2020.16

31 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 31

 WHAT IS AI

Indeed, we at Accenture often recommend RPA as an ideal starting

point for a business that wants to begin an AI journey. That’s because
successfully using RPA means first acquiring a detailed understanding of
the process to be automated. And that’s also the first step in designing a
broader and more sophisticated AI-powered automated solution. It’s also
essential in ensuring that existing, sometimes sub-optimal, processes are
not simply emulated in digital form, but are re-engineered to exploit AI to
the fullest extent possible.

An important point to note: RPA and cognitive RPA do more than simply
cut costs. They also bring new levels of consistency and speed to a
process, as well as offering 24/7 availability and the capacity to scale
the process up and down in line with demand. And it should always be
remembered that RPA replaces tasks, not people. Many organisations who
use RPA redeploy their workforces to activities that add more value to the
business – and are more interesting to boot!

We know this because we’ve done it ourselves. For all of the 17,000
jobs we’ve automated at Accenture, we’ve successfully redeployed our
colleagues in other areas of our business. Indeed, a 2017 Gallup survey17 in
the US suggested that only 13% of workers are worried about automation
eliminating their jobs. Nevertheless, the impact of RPA and AI on the
workforce is a sensitive issue which calls for careful management (for
more on the responsible use of AI, see page 68).

32 | ARTIFICIAL INTELLIGENCE
 WHAT IS AI

RPA has changed the way we structure

and allocate work, allowing us to
focus on improvement initiatives….
By automating routine tasks, we allow
our skilled employees to focus on the
more interesting and challenging parts
of their jobs, which has the dual benefit
of satisfied employees and improved
customer experience. Accenture has
been with us from the beginning, helping
us in the exploratory phases of RPA right
through to the implementation –

Kristian Kjernsmo, Managing Director,

Circle K Business Centre at Circle K Europe

ARTIFICIAL INTELLIGENCE | 33
 HAVEN’T WE BEEN HERE BEFORE?

WAIT, HAVEN’T
WE BEEN HERE
BEFORE?
34 | ARTIFICIAL INTELLIGENCE
 HAVEN’T WE HAVEN’T
BEEN HERE
WEBEFORE?
BEEN HERE BEFORE?

AI is far from a new idea, it’s true. The term “artificial

intelligence” was coined as long ago as 1956.18 And
the history of the technology’s development has
been characterised by waves of optimism followed
by disappointment and periods of inertia (these
have even been dubbed ’AI winters’). Each previous
breakthrough has only ever partly lived up to the
hype it generated, and none has managed to
kick-start the technology into the mainstream.

35 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 35

 HAVEN’T WE BEEN HERE BEFORE?

So what’s different this time?

1940-1956: 1956 1958 1961 1966 1980-

The Birth of AI 1987:
AI Boom
1956-1974: 1974-
1950 The Golden 1980:
1959 1966
Years AI
Winter

Alan Turing Samuel’s checkers

created the program used ELIZA, an
Turing Test. Machine Learning artificial
to beat human conversational
Conference held players. “therapist”
at Dartmouth created.
College where IBM’s Shoebox
the term Artificial performed
Intelligence was arithmetic by
coined. voice command.

Shakey became
the first mobile
robot “aware” of
its surroundings.

Figure 5 – The history of AI

36 | ARTIFICIAL INTELLIGENCE
 HAVEN’T WE BEEN HERE BEFORE?

1994 2000 2011 2017

1987-1994:
Second Winter
1994-Present: 1997 2004 2016 2018
Modern Age

Boom of Expert Two robotic Honda Asimo, a Google’s

Machines in cars drove long personal robot, AutoML lets
industry like distance on the is released. AI generate
the R1/XCON highway. AI.
to help sales
representatives
avoid errors IBM’s Deep Blue IBM’s Watson This
in product defeated chess beats best guide was
suggestions. champion. Jeopardy! published.

Kismet, a social Introduction

machine capable of Virtual
of expressing Agents with
emotions is Siri, Google
introduced. Now, and
the release
of IPSoft’s
Amelia.

ARTIFICIAL INTELLIGENCE | 37
 HAVEN’T WE BEEN HERE BEFORE?

The big change today is that 1. Mainframe

we’re in an unprecedented
period of technology innovation 2. Client-Server and PCs
across so many different fields.
Today’s AI applications can 3. Web 1.0 eCommerce
make use of virtually unlimited
processing power in the cloud. 4. Web 2.0, Cloud, Mobile
They can also exploit a growing
trend for custom-designing 5. Big Data,
computer chips for specific tasks, Analytics, Visualisation
especially in analytics, which is
enabling even greater levels of 6. IoT and Smart Machines
computational efficiency and
speed. Consider, for example, 7. Artificial Intelligence
the vastly increased processing
power that comes from using 8. Quantum Computing
Graphics Processing Units (GPUs)
in place of Central Processing
Units (CPUs). But Google has
taken it one step further, the
Tensor Processing Unit (TPU)
delivering 30-80 times higher
performance-per-watt that
0
contemporary CPUs and GPUs.19 36

C
m 7

:P
te 1 97
S ys
:
19 6 4 1 97 2:
SA P
19 50 Turing Test Ser ver/Host

1950 1960 1970

38 | ARTIFICIAL INTELLIGENCE
 HAVEN’T WE BEEN HERE BEFORE?

Figure 6 – The combinatorial impact

of technology
m
ea

tr
ns
ai
M
d
ou
Cl
ic
bl
Pu
2 0 1 0 : S al e s of P C Pe ak
on
az
: Am

ne

ta
4
99

da
ho
: iP

of
1
t

te s
ne

08

AI
r

20
te

by
In

tta
li c

Ze
:AW
ub

c ar
:P

.4
06
91

:4

in g
19

20

IDC

1 9 9 0 : S y s t e m /3 9 0 Quantum
d ri v
2 .0

14 ,

elf-
20
eb

0: S
:W
05

20 1
20
om
ET
N

.c
PA

ce

r
fo
2M

ue
AR

les at
Bl

a
a

9: S ,
M
9:

D e
oT
p

19 9 ig De
96

1
:B 9 :I M
19 97 1 99 20 07: IB TIME
TODAY

1980 1990 2000 2010 2020 2030

ARTIFICIAL INTELLIGENCE | 39
 HAVEN’T WE BEEN HERE BEFORE?

When you add the decreasing cost of storage20 to the mix (down
from $0.5 million a gigabyte in 1980 to 3 cents a gigabyte in 2015),
plus the exponential growth in data volumes with which we can train
AIs, together with the emergence of open source platforms and
frameworks, you’ve got a uniquely potent combination of technologies
and capabilities. It all adds up to a very powerful foundation to give AI
its critical mass for mainstream adoption.

Virtually all the leading technology giants around the world – Google,
Amazon, Facebook, Microsoft, Baidu, Alibaba, and Tencent – are
sharply focused on AI. Other entrepreneurs and investors are equally
keen. More than half of European start-ups are focused on AI, and
investments in AI businesses are typically 20 to 30% higher than those
in other businesses.21

That’s not to say everyone agrees on precisely when AI will reach its
tipping point. Nor on whether we’ll see general AI (as opposed to
narrow AI) any time soon. On the one hand, a survey of 350 experts
by the Universities of Oxford and Stanford22, concluded that there is
a 50% chance of machines outperforming humans in all tasks within
45 years. On the other hand, a quarter of the eminent AI researchers
surveyed by Etzioni in 2016 said they thought superintelligence would
never materialise at all.23

40 | ARTIFICIAL INTELLIGENCE
 HAVEN’T WE HAVEN’T
BEEN HERE
WEBEFORE?
BEEN HERE BEFORE?

Artificial intelligence would be the

ultimate version of Google. The ultimate
search engine that would understand
everything on the web. It would
understand exactly what you wanted, and
it would give you the right thing. We’re
nowhere near doing that now. However,
we can get incrementally closer to that,
and that is basically what we work on –

Larry Page, 2000

41 | ARTIFICIAL INTELLIGENCE
 HAVEN’T WE BEEN HERE BEFORE?

Your competitors are probably

already using AI today
So we don’t have general AI yet. But, with the underlying technologies
accelerating at breakneck pace, narrow AIs are already doing
remarkable things in real-world business applications.

As organisations continue to ramp up their use of AI, the complexity

of both the data and the work that it can handle will only increase. To
understand how this might play out in a business context, it can be
helpful to view the possible applications of the technology through
the following framework (here illustrated for the financial services
industry). The framework maps four different models for approaching
AI – efficiency; effectiveness; expert; innovation – against the degree
of data and work complexity involved.

In our work with our clients, we already see evidence of AI being

scaled and industrialised. Many organisations have been running pilots
over the last few years to test how AI might impact their people, their
processes, and their products. Now, we expect those organisations
to start scaling their pilots across their enterprises. As many as
three-quarters of executives say that some kind of AI will be “actively
implemented” in their organisation within three years.24

All in all, the message is clear: AI is ready. And it’s a big deal.

42 | ARTIFICIAL INTELLIGENCE
 HAVEN’T WE BEEN HERE BEFORE?

AUGMENT
EFFECTIVENESS MODEL INNOVATION MODEL
Support seamless integration Enable creativity and ideation
and collaboration
Unstructured,
Volatile,
High-Volume Account management New-product creation

Branch management Marketing campaigns

Security and Discovery of microsegments/

identity management customer clusters

DATA
COMPLEXITY
EFFICIENCY MODEL EXPERT MODEL
Provide consistent, low-cost Leverage specialised expertise
performance
Financial advising
Client/prospect discovery
Basic banking transactions Retirement planning
Structured,
Stable, Risk & regulatory compliance
Low-Volume Contact centres/Help desks
Password reset (tech support) Product management

AUTOMATE

Routine, Ad Hoc,
Predictable, WORK Unpredictable,
Rules-based COMPLEXITY Judgment-based

Figure 7– A framework for understanding AI’s potential applications

ARTIFICIAL INTELLIGENCE | 43
 WHAT DO I ACTUALLY NEED TO DO NOW?

OK,
I’M SOLD.
BUT WHAT DO I ACTUALLY NEED
TO DO NOW?

44 | ARTIFICIAL INTELLIGENCE
 WHATWHAT
DO I ACTUALLY
DO I ACTUALLY
NEEDNEED
TO DOTO
NOW?
DO NOW?

Approached in the right way, AI will be a new driver of

economic value for your business. But which way is the
right way? In such a fast-moving field, it can be hard to
see the wood for the trees. The first step is to understand
the opportunity that AI presents. By breaking it down
into three avenues, you can get a much clearer sense
of the route you should be taking. So, that means
you should be thinking about how to use AI to shift
automation up a gear, about how to augment what you do
and the way you do it, and about how AI innovation might
diffuse through your business and beyond.

45 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 45

 WHAT DO I ACTUALLY NEED TO DO NOW?

1. Automating
more
AI is the new frontier of
automation. With self-
learning autonomous
systems that mimic human
behaviour exploiting machine
learning, computer vision, and
knowledge representation and
reasoning, AI can take automation
beyond merely rules-based
predictable work, right into the
areas we currently believe need
human judgement. That opens up
a huge number of new automation
opportunities (for more on the use of AI
in industrial automation, see page 30).

46 | ARTIFICIAL INTELLIGENCE
 WHAT DO I ACTUALLY NEED TO DO NOW?

MUCH AUTOMATION … AND AI IS THE

OPPORTUNITY NEW FRONTIER OF
BUSINESS IMPACT TRANSFORMATIONAL

REMAINS… AUTOMATION

INTELLIGENT AUTOMATION

ROBOTIC PROCESS The application of AI to processes to

AUTOMATION enhance automation
Enterprise grade
FOUNDATION surface-level Often augments human actions rather
integration than replacing them entirely e.g. use of AI
Project-level ad-hoc [BluePrism, models in human decision-making
automation, e.g. Automation
scripts, macros, AI may be applied deeply in one business
Anywhere, Fusion, area/process, or for shallow automation
batch programs, Jacada] and digital
[minibots] across many processes
technology enablers
[OCR, BPM] Cognitive RPA is the addition of AI to RPA,
thereby making it more “intelligent” or
TACTICAL

versatile

PLATFORMS & AUTOMATION ORCHESTRATION

PROGRAMMED NATURE OF WORK SELF-LEARNING

STRICTLY AUTONOMOUS
CONTROLLED RULES BASED JUDGEMENT BASED UNBOUNDED
CONTAINED

Figure 8 – The intelligent automation frontier

ARTIFICIAL INTELLIGENCE | 47
 WHAT DO I ACTUALLY NEED TO DO NOW?

ROBOTS FOR RECYCLING

AMP Robotics has created a robotic system called
Cortex which uses computer vision to rapidly pick
recyclable materials from a conveyor belt of waste
products. The system is driven by an AI, called
Neuron, capable of distinguishing materials that can
be recycled from those that can’t – even if they’re
dirty or piled up with other materials – using a video
stream.25

2. Augmenting how you work

AI brings new levels of efficiency to the use of resources. In practice,
that means two things. First, augmenting human workers’ judgement.
And second, enhancing customer experience.

#1 Working smarter. When it comes to augmenting worker judgement,

machine learning is capable of extracting more meaning from very
large and highly complex data sets than a human ever could. An AI can

48 | ARTIFICIAL INTELLIGENCE
 WHAT DO I ACTUALLY NEED TO DO NOW?

thus see patterns, similarities, and anomalies, where human experts

see none. Consider cancer detection, for example. Human specialists
can recognise several hundred malignant patterns in a cancer scan,
whereas an AI can recognise thousands.

Researchers at the University of Nottingham have already created an

AI that can predict which patients are likely to have a stroke or heart
attack within ten years. The AI performed better than the standard
methods of prediction (scoring between 0.745 and 0.764 out of 1 as
against 0.728 for the standard method).26

These are hugely impressive strides forward. But they shouldn’t be

taken as a sign that human expertise will be superseded any time soon.
The very best results are still achieved when human experts work
together with AI, each bringing the best of their unique capabilities to
bear on a problem.

4 out of 5 executives (81%) agree within the

next two years, AI will work next to humans
in their organisations, as a co-worker,
collaborator and trusted advisor27 –

Accenture Technology Vision, 2018

ARTIFICIAL INTELLIGENCE | 49
 WHAT DO I ACTUALLY NEED TO DO NOW?

#2 Better experiences for customers. Using AI, and particularly the

cognitive aspects of the technology, a business can vastly improve
the way it interacts with its customers. That could mean using
digital assistants and chatbots to converse with customers 24/7
through social media and digital platforms. Or it could mean making
personalised product or service recommendations on an e-commerce
site.

When South American airline Avianca wanted to enhance the travel

experience of its 28 million passengers, for example, they quickly
settled on the idea of a chatbot assistant.28 We at Accenture helped
them create Carla, a Facebook Messenger chatbot which uses AI to
help customers manage their travel arrangements. By holding natural-
sounding conversations with Carla on a messaging platform they’re
already familiar with, Avianca’s customers have a quick and intuitive
way to check in, review itineraries and flight status, and get weather
and other updates from the airline – with no waiting on the phone for
an advisor.

Moreover, creating a chatbot or digital assistant like Carla needn’t

cost the earth or consume a business for months on end. Developing
Avianca’s chatbot took just six weeks. And almost straightaway it
acquired over 20,000 unique users who were holding something like
4,000 conversations with it every month. Carla even managed to cut
the average check-in time for Avianca’s customers in half.

50 | ARTIFICIAL INTELLIGENCE
 WHATWHAT
DO I ACTUALLY
DO I ACTUALLY
NEEDNEED
TO DOTO
NOW?
DO NOW?

HUMAN–ROBOT
JOURNALISM IS ALREADY
HAPPENING
The Press Association and Urbs Media, backed by
Google’s Digital News Initiative, are putting robots to
use in very interesting ways. Their Reporters and Data
Robots (RADAR) initiative has created software which
sifts through national data sets and inserts localised
statistics into stories written by human reporters.
The stories, which to date have covered everything
from hospital cancellations to problems with
social mobility, are then offered to local or regional
newspapers for publication. Since a pilot began at
the end of November 2017, 20 newspapers have
published articles created by the initiative, saving
their reporters time and lending more weight to their
stories.29

51 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 51

 WHAT DO I ACTUALLY NEED TO DO NOW?

CHECKING IN TO
YOUR HOTEL WITH A SELFIE
AI’s customer experience opportunities don’t begin
and end with chatbots. Consider the possibilities
of facial recognition technology for biometric
identification, for example. Singapore and San
Francisco-based GTRIIP30 have developed a
document-free mobile check-
in application which uses AI
and biometric technology to
let hotel guests check-in to
their rooms with a fingerprint
– or by simply taking a selfie.

52 | ARTIFICIAL INTELLIGENCE
 WHAT DO I ACTUALLY NEED TO DO NOW?

3. Diffusing innovation
Innovation begets innovation. The spill-over effects of a radical
new technology can cascade through entire economies, changing
everything forever, and in ways that were never foreseen. When
electricity was first industrialised, who could have imagined the
vast energy demands of today’s power-hungry world? And when the
internal combustion engine was created, who could have conceived of
the speed and scale of our interconnected global transport networks?

AI will impact society to a similar degree. Its innovations will diffuse

through businesses – and beyond to whole economies. New, as-yet
unimagined business models and opportunities will be created. From
the automation of tasks we once thought needed human intelligence,
to the ability to see patterns in vast amounts of data, to new cognitive
human–technology interfaces, this technology will have far-reaching
and radical implications for our working and personal lives.

Autonomous electric vehicles, for example, will completely upend our

current thinking about transportation. Just a few years ago, driving
was an activity thought to be so complex that it would have to remain
the preserve of human beings. Now, automated transportation is
becoming a reality. And, from automotive companies, to logistics, to
petrol sales, to global oil, the direct effects of a shift to autonomous
vehicles will be profound. But consider, too, the effects of greatly
enhanced safety on vehicle design, on insurance needs, and on

ARTIFICIAL INTELLIGENCE | 53
 WHAT DO I ACTUALLY NEED TO DO NOW?

medical and emergency response. Or what about the impact on road

network design, parking needs, car dealerships, and petrol taxes? The
sheer scale of the societal impact from an AI innovation in just one
industry is breath-taking.

One of the revolutionary aspects of AI is its simplicity of use. In

other words, humans won’t need to adapt to it, or learn a new set
of skills to use it, making it perhaps unique in the history of radical
technologies. We will be able to interact with AI through simple and
natural language, whether by voice or by text, or even through images.
The contrast with, say, learning to drive a car, or understanding how
to use a PC for the first time, couldn’t be starker. This has one very
important implication: it means the critical-mass adoption of
AI is likely to take hold even faster than previous disruptive
technologies.

The World Economic Forum

states that autonomous driving
features will help prevent 9%
of accidents by 2025 with the
potential to save 900,000 lives
in the next 10 years.31

54 | ARTIFICIAL INTELLIGENCE
 WHATWHAT
DO I ACTUALLY
DO I ACTUALLY
NEEDNEED
TO DOTO
NOW?
DO NOW?

AI FOR GOOD
Accenture has recently completed a
pilot program in the UK, which uses its
artificial intelligence platform to help
seniors manage their care and daily
lives. The technology was developed
by Accenture Liquid Studio in
London and is
tailored for older
people living
independently.32

55 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 55

 WHAT
WHATDO
DOIIACTUALLY
ACTUALLYNEED
NEEDTO
TODO
DONOW?
NOW?

THE ROBOT SURGEON

OPERATING ON LIVE
PORCINE TISSUE
In a breakthrough proof of concept for robotic
surgery, a Smart Tissue Autonomous Robot (STAR)
proved it could stitch up the small intestine of a
living pig using its own vision, tools, and intelligence.
Intestinal sutures are a particularly challenging task
for an autonomous robot, given the soft tissue’s
deformity and mobility. But, with more consistent and
leak-resistant sutures, STAR was able to outperform
human surgeons who were given the same task.33

56 | ARTIFICIAL INTELLIGENCE
 WHATWHAT
DO I ACTUALLY
DO I ACTUALLY
NEEDNEED
TO DOTO
NOW?
DO NOW?

Putting it into practice

With a clear view of the opportunities, you’re ready to put AI to use
in your organisation. First and foremost, that means developing
a strategy and a roadmap for your AI journey. That roadmap must
encompass both the re-engineering of the affected business
processes, and appropriate governance controls. Above all, it must
prioritise the high-impact AI initiatives for your business.

An important point: it’s vital to start with the business case, not the
technology. In other words, think first about what you want to do,
not what the technology could do. Only then should you add AI to
the mix and analyse where it can add value. That should include
assessing the feasibility of using AI, the effort vs. return, and the
risks involved. It should also include finding a willing business
sponsor to push adoption across your organisation. Because, after
all, getting people to change their behaviour is often the hardest
part in introducing any radical innovation.

57 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 57

 WHAT
WHATDO
DOIIACTUALLY
ACTUALLYNEED
NEEDTO
TODO
DONOW?
NOW?

WHAT’S STOPPING SOME

COMPANIES FROM USING
AI?
For all the undeniable activity, the fact remains that
most organisations are yet to begin their AI journeys.
And of those that have started, half are still in the
pilot or proof of concept stage.34 So what’s stopping
them? The reasons are varied, and not dissimilar to
those encountered with advanced analytics. For
some, it comes down to getting the right talent,
prioritising investments, and allaying concerns about
security. For others, it’s about defining compelling
business cases, strong enough leadership support,
and acquiring more general technology capabilities.

58 | ARTIFICIAL INTELLIGENCE
 WHATWHAT
DO I ACTUALLY
DO I ACTUALLY
NEEDNEED
TO DOTO
NOW?
DO NOW?

Figure 9 – What are the top three barriers to AI adoption in your organisation?35

Attracting, acquiring & developing

the right AI talent

Competing investment
priorities

Security concerns resulting

from AI adoption

Limited or no general
technology capabilities

Lack of leadership
support for AI initiatives

Unclear or no business
case for AI applications

Leaders Passives

Figure 10 – What is the level of AI adoption in your organisation? 35

No adoption 54%

Pilots 23%

Adoption 23%

59 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 59

 WHAT DO I ACTUALLY NEED TO DO NOW?

In developing an AI application, an agile “fail fast” approach is

invaluable. That means running pilots for each business problem
or opportunity to test the feasibility of your solution and assess the
technology options available to your organisation.

An appropriate operating model – with Board-level approval – is also

essential. In our experience at Accenture, we find creating a central
“hub”, or centre of excellence, to provide leadership and governance
works best. The hub can then be connected, via a series of “spokes”,
to other parts of the business (whether that’s marketing, risk, HR, or
others).

This kind of hub-and-spoke model brings some really important

benefits to your AI journey:

• It offers economies of scale in selecting the technologies and tools

you’ll need, as well as in recruitment and talent development;

• It enables a cross-organisation approach to building and

maintaining the necessary data ecosystem;

• It ensures consistency in standards, definitions, and methods, and

lets you disseminate best practices across your organisation;

• It provides a rigorous means of measuring value and prioritising

opportunities;

• It ensures the highest standards of governance are applied to all

your AI projects.

60 | ARTIFICIAL INTELLIGENCE
 WHAT DO I ACTUALLY NEED TO DO NOW?

By 2019, 40% of digital transformation

programmes and 100% of IoT
programmes will be supported by AI
capabilities36 –

IDC

61 | ARTIFICIAL INTELLIGENCE
 WHAT DO I ACTUALLY NEED TO DO NOW?

Keeping an eye on the road ahead

As you progress through an AI journey, keeping abreast of the latest
innovations and applications is essential. In such a fast-moving
landscape, things can change overnight. Knowing what the AI
leaders and the AI first-movers are doing – and understanding the
implications – will always be a hugely valuable source of business
intelligence.

So what might we expect on the road ahead?

In the short term, voice interaction will remain the hottest

consumer-facing manifestation of AI. And Amazon Alexa
and Google Home will continue their battle to be the
primary gatekeeper of the smart home, with Apple’s
HomePod as the latest entrant. Expect also to see
questions raised between platform providers and service
providers about who owns what, and who has access to
what, when it comes to customer data and relationships.

Voice is part of a more general trend for AI as the new UI. In other
words, that AI is becoming the channel of choice for customer
interactions, whether that’s over chat services, messaging, or smart
home devices. Its primacy means organisations must think very
carefully about how they use it and how it represents their brand.
Those treating it as an afterthought or add-on will quickly come
unstuck.

62 | ARTIFICIAL INTELLIGENCE
WHAT DO I ACTUALLY NEED TO DO NOW?

Within 5 years more

than half of customers
will select services
based on a business’
AI rather than its
traditional brand37 –

Accenture Technology
Vision, 2017

ARTIFICIAL INTELLIGENCE | 63
 WHAT DO I ACTUALLY NEED TO DO NOW?

One important implication of this shift is that AI teams must be

interdisciplinary, and not simply technical. Microsoft, for example,
employs a whole team, including writers and psychologists, to give its
operating system AI, Cortana, its unique personality. For more on the
interdisciplinary nature of AI development, see page 27.

Look out, too, for developments in the way brands advertise through
devices like the Amazon Echo and Google Home. Voice ads are still
very much in their infancy, and a great deal of trial and error is still to
come. Interrupting people in their homes to sell them services without
being asked will more than likely anger customers. Indeed, Amazon
updated its Alexa developer policy in April 2017 to ban all adverts
except those in music and flash briefings.38 So innovative ways to get
people coming back to a brand’s voice utility will be needed. Quid-
pro-quo bargains will likely be struck with customers – their attention
in return for something back.

A key point is this: if the state of AI technology isn’t already streets

ahead of our collective imagination, it soon will be. That makes
it very different from previous technology revolutions. What’s
possible with AI will only be limited by our willingness to experiment –
and use it.

64 | ARTIFICIAL INTELLIGENCE
75% of business executives
say AI will be actively
implemented in their
companies within three
years39 –

2017 Economist Intelligence

Unit report

65
 RISKS INVOLVED

HOLD ON,
WE HAVEN’T TALKED
ABOUT THE RISKS
INVOLVED

66 | ARTIFICIAL INTELLIGENCE
 RISKS INVOLVED RISKS INVOLVED

The huge opportunities and benefits that AI offers

don’t come risk free, of course. What kind of
innovation does? But it certainly pays to begin an AI
journey with a clear-sighted view of what the risks
might be for an organisation.

67 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 67

 RISKS INVOLVED

So, what are they? We think there are four principal risks that must
be considered up front. These relate to trust, liability, security, and
control:

TRUST
How do we demonstrate to citizens that an AI is safe to use? How do
we avoid biases, unconscious or not, being written in from the outset?
The answers to these questions lie in transparency and accountability.
Decisions taken by an AI must be open to appeal and interrogation.

LIABILITY
What happens when an AI makes an error – or even breaks the law?
Who is legally responsible? Changes to legislative and regulatory
requirements will need to be monitored carefully.

SECURITY
How do we prevent unauthorised or malicious manipulation of an AI?
Security becomes paramount, and is compounded by the increasing
use of open source code.

CONTROL
What happens when a machine takes over a process? How does a
human take it back if they need to? Careful thought is needed about
when and how control is transferred between humans and AIs. For
example, it is all very well providing a human riding in a self-driving
car with the means to take control but if they are not paying attention
100% of the time, they will be unlikely to intervene fast enough in a
critical situation.

68 | ARTIFICIAL INTELLIGENCE
 RISKS INVOLVED

72% of executives report

that their organisations
seek to gain customer
trust and confidence by
being transparent in their
AI-based decisions and
actions40 –

Accenture Technology Vision,

2018

69 | ARTIFICIAL INTELLIGENCE
 RISKS INVOLVED

Accenture Launched “Pinterest For AI Education”,

a new technology platform to train more than
180,000 of its employees globally in the latest digital
technologies in just over 20 months. It now plans
to use the interactive platform with clients to help
develop their IT workforces in critical areas such as
digital, cloud, security and artificial intelligence.
The Accenture Future Talent Platform integrates
learning services and curriculum on as-a-service and
mobile platforms to help workforces move away from
traditional training and foster a culture of continuous
learning.41

70 | ARTIFICIAL INTELLIGENCE
 RISKS INVOLVED

These issues are something we’ve given a great deal of thought to

at Accenture. Our strong recommendation is to take a “human first”
approach to AI thinking. And that means adopting a framework for what
we call “Responsible AI”. This framework recommends mitigating the risks
of using AI with four imperatives: govern; design; monitor; and reskill:

01. GOVERN
Create the right governance framework for AI to flourish. Anchor it to
your organisation’s core values, ethical guardrails, and accountability
frameworks.

02. DESIGN
Build trust into your AI from the outset by accounting for privacy,
transparency, and security from the earliest design stage.

03. MONITOR
Audit the performance of your AI against a set of key metrics. Make sure
algorithmic accountability, bias, and security metrics are included.

04. RESKILL
Democratise the understanding of AI across your organisation to break
down barriers for individuals impacted by the technology.

It’s also vital to remember that humans can be susceptible to unconscious

bias. And that has big implications when it comes to coding and training
an AI and selecting the data sets it will use. This is an area in which general
standards may be required. Collectively, we may need to aspire to a higher
degree of responsibility from our AIs than we would necessarily demand
from human colleagues.

ARTIFICIAL INTELLIGENCE | 71
 RISKS INVOLVED

The need to explain

There is one thing above all that will ensure public trust is maintained
when an organisation starts using AI: “explainability”. In other words,
being ready to explain how and why an AI came to the decision it did.
This is something that certain regulated industries are already familiar
with. Financial services institutions, for example, are required to
explain the decisions they take that affect their customers.

But there’s also a broader issue here: humans are more likely to trust
something they understand. So “explainable AI” becomes a vital part
of any AI strategy.

DARPA’S EXPLAINABLE AI
CHALLENGE
DARPA, the Defense Advanced Research Projects Agency,
has launched an explainable AI (XAI) programme with the
goal of creating a suite of machine learning techniques
which produce more explainable models. The models will be
combined with interfaces capable of translating them into
understandable and useful explanations for human users.42

72 | ARTIFICIAL INTELLIGENCE
 RISKS INVOLVED

In fact, that’s easier said than done. Machine learning is often by its
very nature a “black box” exercise. In other words, it operates in ways
that can make it very hard to explain how it arrived at the outputs it
produced. But many AI practitioners and data scientists are thinking
about this question, and new approaches offering better explanations
of the science underlying AI decisions will likely soon emerge.
In the meantime, there are some practical steps that every business
can take now to make their AI more explainable:
01. INVENTORY
Think about the decisions that are or will be taken by AI in your
organisation. Which of them would require an explanation – or create
an expectation of one? Do they relate, even indirectly, to key areas like
employment, recruitment, lending, education, healthcare, housing,
inclusion, or safety?

02. ASSESS
Consider any quantitative and qualitative models that are already
providing explanations for decisions taken by AI. How are they
performing for their intended recipients?

03. DESIGN
Revisit the design principles used for your AI. How could they
make the process of making decisions more human-centred and
understandable?

04. AUDIT
Review the data. How do you ensure your AI is using data sets that
reflect the evolving nature of your workplace?

ARTIFICIAL INTELLIGENCE | 73
 THE BIGGER PICTURE

AND WHAT
ABOUT THE
BIGGER
PICTURE?

74 | ARTIFICIAL INTELLIGENCE
 THE BIGGER PICTURE THE BIGGER PICTURE

No person is an island. And no business is either.

The actions we take can reverberate way beyond the
boundaries of a single organisation. So, given AI’s
revolutionary potential and far-reaching spill-over
effects, the broader societal implications of using it
can’t be ignored.

This means collectively addressing some important

questions. How do we make sure people have the skills
they need to thrive in an AI-driven world? How many
existing jobs will AI replace? How many new jobs will
be created? Will some people need to find income and
fulfilment from sources other than work? What new legal
frameworks are needed when AI is taking the decisions?
Might humanity even face an existential threat when AIs
become more intelligent than their creators?

75 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 75

 THE BIGGER PICTURE

The skills question

When it comes to skills, governments have for years promoted the
uptake of STEM subjects. This must be maintained and accelerated,
with the addition of new data and analytics elements into school and
training curricula. That shouldn’t just apply to STEM subjects either
– the importance of data and analytics to other disciplines must
be emphasised with equal force. Drives to increase the numbers of
apprenticeships, such as that promoted by the UK government, should
play a part too.

Industry also has a vital role in upskilling workforces for the AI age, of
course. And that will take considerable investment. But businesses
have both an economic incentive (to ensure a ready supply of talent)
and a moral obligation (to use AI responsibly) to do so. What’s more,
industry involvement becomes even more vital if the speed of AI-
driven change means our existing education and training bodies can’t
keep pace on their own.

Occupations that require some of humanity’s most intrinsic qualities

– creativity, empathy, kindness, care – may in any case be some of
the last to be affected by AI. There is little chance of an AI replicating
these core aspects of human intelligence any time soon. So there will
be many important and rewarding career paths which will remain open
to people, even if they can’t necessarily outsmart the machines.

76 | ARTIFICIAL INTELLIGENCE
 THE BIGGER PICTURE

62% of workers think AI

will have a positive impact
on their work, with most
employees falling into the
high skill/high willingness
category43 –

Accenture

77 | ARTIFICIAL INTELLIGENCE
 THE BIGGER PICTURE

The jobs question

Whether AI will ultimately have a positive or negative net impact on
employment is a big question, and not one that can be answered
in a Pocket Guide. But we at Accenture are broadly optimistic. If
organisations can use AI in a responsible and human-centred way,
and if they can focus on augmenting human intelligence as much as
replacing it, we think the impact will be positive. Approaching AI in
this way will allow human workforces to focus on the more interesting,
challenging, creative, and interpersonal parts of their jobs – and leave
the humdrum, boring, repetitive parts to the machines.

We’re not alone in our optimistic outlook. And that’s because many
people think AI will end up creating more jobs than it destroys.
Those that take this view look at the history of previous technology
revolutions and see that, in every case, there was ultimately a net
increase in overall employment. That increase
often skipped a generation, admittedly. So,
whereas an older generation of workers
might have lost out from the introduction of
a new technology, history suggests the next
generation benefitted.
In the long run, this
argument goes,
AI will have
the same net
positive impact.

78 | ARTIFICIAL INTELLIGENCE
 THE BIGGER PICTURE THE BIGGER PICTURE

THE EFFECT OF
AUTOMATION ON
MANUFACTURING
How has the automation of manufacturing processes
impacted employment? You might think the
widespread use of industrial robots means there’d
be a great deal fewer workers needed today than in
the past. Not so, according to a recent report from
Germany.

Over the past 20 years, Europe’s strongest economy

and manufacturing powerhouse has quadrupled the
number of robots used in its industries. In 1994, for
example, Germany had around 2 industrial robots
installed for every thousand workers. By 2014, this
had risen to 7.6 robots for each thousand workers
(compared to just 1.6 in the US).

79 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 79

 THE BIGGER PICTURE

In the country’s thriving automotive industry, between

60 and 100 additional robots were installed for each
thousand workers over that period.

The impact on German employment of all this

automation? Close to nil. Researchers from the
Universities of Würzburg, Mannheim, and the
Düsseldorf Heinrich-Heine University examined 20
years of employment data and found that, despite
significant growth in the use of robots, there hadn’t
been any dent in aggregate German employment:
“Once industry structures and demographics are
taken into account, we find effects close to zero”.44

80 | ARTIFICIAL INTELLIGENCE
 THE BIGGER PICTURE

Not everyone is so sanguine, it’s true. Such people argue that this time
it’s different. They point out that, whereas many previous technology
revolutions involved the mechanisation of manual labour, AI involves
the automation of far more fundamental cognitive processes. So, while
mechanical automation might replace a set of specific tasks (tractors
replacing horses in agriculture for example), cognitive automation is so
broad as to strike at the core competitive advantage of human beings:
their ability to think. In the long run, this argument goes, we should
brace ourselves for widespread job losses.

With such a broad array of views, a quick resolution to the jobs

question is unlikely. The debate will run for some time yet. In the
meantime, it is incumbent on both government and industry to make
every effort to ensure responsible and human-centred approaches are
taken to the use of AI.

In 2020 AI will create 2.3 million jobs,

whilst eliminating 1.8 million, making
2020 a pivotal year in AI-related
employment dynamics45 –

Gartner

ARTIFICIAL INTELLIGENCE | 81
 THE BIGGER PICTURE

The legal question

There is no doubt that legal and regulatory frameworks will need
to evolve as AI takes an ever-greater hold in business, industry, and
beyond. One of the first and most pressing areas will likely be the law
surrounding autonomous vehicles. Consider legal areas like personal
injury, negligence, and tort. Who, for example, will be liable if an
autonomous vehicle crashes with no driver at the wheel? The vehicle
manufacturer? The software provider? The human occupants?

And, what about the legal implications of AI design? When an

autonomous vehicle faces two equally tragic choices (saving the
life of its human occupant, say, against saving pedestrians on the
street ahead) the AI’s design will need a clearly thought-through legal
framework for deciding which option to take. This cuts right to the
core of some of the most difficult legal and moral questions there are.

Competition law may well need to adapt too. Pricing algorithms are
already widely used by online retailers, enabling much faster and
more sophisticated price adjustments. Amazon reputedly changes
the prices of some items many times each hour, equating to millions
of individual price changes every day.46 Any decision made by AI in an
open marketplace will need to be transparent and explainable.

82 | ARTIFICIAL INTELLIGENCE
 THE BIGGER PICTURE THE BIGGER PICTURE

The existential question

No-one knows for sure whether AIs will ever surpass a human’s general
intelligence. Even less, what will happen if it does. And among those
experts who think general AI is a real possibility, opinions diverge
hugely on when it might happen. We feel confident in saying that
despite AI’s truly amazing potential for humanity, general AI is a
long way from being a reality. But that doesn’t mean we shouldn’t
be thinking about its consequences and implications. So the
establishment of academic and industry-sponsored bodies to examine
the question, such as the Future of Humanity Institute at Oxford
University, is a very welcome development.

83 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 83

 IT’S TIME TO GET SMART AND GET STARTED

IT’S TIME TO
GET SMART
AND GET
STARTED

84 | ARTIFICIAL INTELLIGENCE
 IT’SIT’S
TIMETIME
TO TO
GETGET
SMART
SMART
ANDAND
GETGET
STARTED
STARTED

So that’s AI. We hope this guide has given you

a clear sense of its underlying technologies, its
game-changing capabilities, and its potentially
revolutionary implications. As it gains critical
mass, AI will change forever how we work and how
we live.

It’s a completely new factor of production, and it’s

going to drive business growth in important ways,
whether by extending automation into previously
human-only arenas, augmenting our work to
drive new levels of effective decision making, or
diffusing exponential levels of innovation through
organisations and beyond.

85 | ARTIFICIAL INTELLIGENCE ARTIFICIAL INTELLIGENCE | 85

 IT’S TIME TO GET SMART AND GET STARTED

86 | ARTIFICIAL INTELLIGENCE
 IT’S TIME TO GET SMART AND GET STARTED

Businesses big and small are already experimenting with AI solutions

to drive new growth. They’re finding ways to make their workers
smarter. They’re discovering new kinds of interactions with their
customers. And they’re starting to build AI machines that can work
faster and better than their human counterparts.

Like any radical innovation, there will be risks in starting an AI journey.

And there may be unintended consequences along the way. That’s
why taking a responsible approach to the technology is so important.
Trust, transparency, and security must be built into AI design from the
outset. And the need to explain must always be front of mind.

In the end the message is simple. AI is here. It’s real. And it’s time to sit
up, take notice, and take advantage.

ARTIFICIAL INTELLIGENCE | 87
 IT’S TIME TO GET SMART AND GET STARTED

Recommended Reading
The Second Machine Age by Erik Brynjolfsson and Andrew McAfee

Machine, Platform, Crowd by Erik Brynjolfsson and Andrew McAfee

Life 3.0 by Max Tegmark

Human + Machine: Reimagining Work in the Age of AI by Paul

Daugherty & James Wilson (forthcoming March 20th, 2018)

Homo Deus by Yuval Noah Harari

The Quest for Artificial Intelligence by Nils Nilsson

The Master Algorithm by Pedro Domingos

The Future of the Mind by Michio Kaku

I would like to thank Accenture colleagues too numerous to mention

for their help formulating the concepts described in this publication;
and Lucy Frost, Noor Sajid, Caryn Tan and Alexandra Vernon for their
assistance with research and design.

88 | ARTIFICIAL INTELLIGENCE
 IT’S TIME TO GET SMART AND GET STARTED

Sources
1N
 anterme, P. (2016, January 17). Digital Disruption Has Only Just Begun. World Economic Forum. https://
www.weforum.org/agenda/2016/01/digital-disruption-has-only-just-begun/
2 Faggella, D. (2017, August 24). Venture Investments in Artificial Intelligence – Trends in 2016 and Beyond.
TechEmergence. https://www.techemergence.com/venture-investments-in-artificial-intelligence-trends/
3H
 offman, M. (2016). Artificial Intelligence Patents Growing Exponentially. Hoffman Warnick. http://
hoffmanwarnick.com/artificial-intelligence-patents-growing-exponentially/
4 Hall, W., & Pesenti, J. (2017, October 15). Growing the Artificial Intelligence Industry in the UK. https://
www.gov.uk/government/uploads/system/uploads/attachment_data/file/652097/Growing_the_artificial_
intelligence_industry_in_the_UK.pdf
5 Wang, R. (2016, September 18). Understanding the Spectrum of Seven Artificial Intelligence Outcomes. R
“Ray” Wang. http://blog.softwareinsider.org/2016/09/18/mondays-musings-understand-spectrum-seven-
artificial-intelligence-outcomes/
6 BBC. (2017, May 25). Google AI Defeats Human Go Champion. BBC News. http://www.bbc.co.uk/news/
technology-40042581
7 Hassabis, D., & Silver, D. (2017, October 18). AlphaGo Zero: Learning from Scratch. DeepMind. https://
deepmind.com/blog/alphago-zero-learning-scratch/
8 Schaeffer, J. (2018, January 12). Techmate: How AI Rewrote the Rules of Chess. Financial Times. https://www.
ft.com/content/ea707a24-f6b7-11e7-8715-e94187b3017e
9 Brynjolfsson, E., & McAfee, A. (2017, July). The Business of Artificial Intelligence. Harvard Business Review.
https://hbr.org/cover-story/2017/07/the-business-of-artificial-intelligence
10 Gartner. (2010). Gartner Customer 360 Summit 2011. Gartner Summits. https://www.gartner.com/
imagesrv/summits/docs/na/customer-360/C360_2011_brochure_FINAL.pdf
11 C
 larifai. (2016, April 19). Animals vs. Food vs. A.I. https://crimesagainstai.com/2016/04/19/animals-vs-food-
vs-a-i/
12 Grosse, R., Lee, H., Ng, A., & Ranganath, R. (2011, October). Unsupervised Learning of Hierarchical
Representations with Convolutional Deep Belief Networks. https://www.cs.princeton.edu/~rajeshr/papers/
cacm2011-researchHighlights-convDBN.pdf
13 Antonoglou, I., Graves, A., Kavukcuoglu, K., Mnih, V., Riedmiller, M., Silver, D., & Wierstra, D. (2013, January
1). Playing Atari with Deep Reinforcement Learning. DeepMind Technologies. https://deepmind.com/
research/publications/playing-atari-deep-reinforcement-learning/
14 W
 est, J. (2016, April 2). Microsoft’s Disastrous Tay Experiment Shows the Hidden Danger of AI. Quartz.
https://qz.com/653084/microsofts-disastrous-tay-experiment-shows-the-hidden-dangers-of-ai/
15 Hopkins, B., McCormick, J., & Schadler, T. (2016, July 27). Insights-Driven Businesses Will Take $1.2 Trillion a

ARTIFICIAL INTELLIGENCE | 89
 IT’S TIME TO GET SMART AND GET STARTED

Year by 2020. Forrester. https://www.forrester.com/InsightsDriven+Businesses+Will+Take+12+Trillion+A+Year+

By+2020/-/E-PRE9365
16 T
 ransparency Market Research. (2016, November 4). Global IT Robotic Automation Market. https://www.
transparencymarketresearch.com/it-robotic-automation-market.html
17 Gallup. (2017). “Few U.S. Workers Worry About Tech Making Their Job Obsolete” p.2, August 2-6, 2017.
Retrieved from http://news.gallup. com/poll/216116/few-workers-worry-tech-making-job-obsolete. aspx?g_
source=workplace+satisfaction&g_medium=search&g_ campaign=tiles
18 C
 revier, D. (1993). AI: The Tumultuous Search for Artificial Intelligence. New York: Basic Books.
19 S
 ato, K. (2017, May 12). An In-depth Look at Google’s First Tensor Processing Unit (TPU). Google Cloud
Platform. https://cloud.google.com/blog/big-data/2017/05/an-in-depth-look-at-googles-first-tensor-
processing-unit-tpu
20 Klein, A. (2017, July 11). Hard Drive Cost Per Gigabyte. Blackblaze. https://www.backblaze.com/blog/hard-
drive-cost-per-gigabyte/
21 K
 elnar, D. (2017). The State of AI 2017: Inflection Point. MMC Ventures. https://www.mmcventures.com/
wp-content/uploads/2017/10/The-State-of-AI-2017-Inflection-Point-Summary.pdf
22 R
 evell, T. (2017, May 31). AI Will be Able to Beat us at Everything by 2060, Say Expert. New Scientist. https://
www.newscientist.com/article/2133188-ai-will-be-able-to-beat-us-at-everything-by-2060-say-experts/
23 E
 tzioni, O. (2016, September 20). No, the Experts Don’t Think Superintelligent AI is a Threat to Humanity.
MIT Technology Review. https://www.technologyreview.com/s/602410/no-the-experts-dont-think-
superintelligent-ai-is-a-threat-to-humanity/
24 B
 erlucchi, M. et.al. (n.d.). Artifical Intelligence in the Real World. The Economist. https://www.
eiuperspectives.economist.com/sites/default/files/Artificial_intelligence_in_the_real_world_1.pdf
25 R
 obots for Recycling. (2017). https://www.amprobotics.com/
26 L
 ant, K. (2017, May 5). Confirmed: AI Can Predict Heart Attacks and Strokes More Accurately Than Doctors.
Futurism. https://futurism.com/ confirmed-ai-can-predict-heart-attacks-and-strokes-more-accurately-
than-doctors/
27 Accenture Technology Vision. (2018). Redefine Your Company Based on the Company You Keep: Intelligent
Enterprise Unleashed. https://www.accenture.com/t20180208T172438Z__w__/us-en/_acnmedia/
Accenture/next-gen-7/tech-vision-2018/pdf/Accenture-TechVision-2018-Tech-Trends-Report.pdf
28 Accenture. (2018). Avianca Airlines New Chatbot is Reducing Travel Stress, One Passenger at a Time.
https://www.accenture.com/gb-en/success-avianca-airlines-new-chatbot-reducing-travel-stress
29 G
 regory, J. (2017, July 6). Press Association Wins Google Grant to Run News Service Written by Computers.
The Guardian. https://www.theguardian.com/technology/2017/jul/06/press-association-wins-google-
grant-to-run-news-service-written-by-computers
30 Lee, J. (2015, April 24). GTRIIP Provides Biometric Identification Solution for Hotel Check-In. Biometric
Update. http://www.biometricupdate.com/201504/gtriip-provides-biometric-identification-solution-for-
hotel-check-in

90 | ARTIFICIAL INTELLIGENCE
 IT’S TIME TO GET SMART AND GET STARTED

31 W
 orld Economic Forum. (n.d.). The Driverless Car Revolution. http://reports.weforum.org/digital-
transformation/the-driverless-car-revolution/
32 L
 ovett, L. (2017, December 4). Accenture’s Latest AI Platform to Help Seniors with Loneliness Wraps up
Pilot Program. MobiHealthNews. http://www.mobihealthnews.com/content/accentures-latest-ai-platform-
help-seniors-loneliness-wraps-pilot-program
33 L
 eonard, S., (2014, January 23) .Smart Tissue Anastomosis Robot (STAR): A Vision-Guided Robotics System
for Laparoscopic Suturing. http://ieeexplore.ieee.org/abstract/document/6720152/?reload=true
34 R
 ansbotham, R., Kiron, D., Gerbert, P., & Reeves, M. (2017, September 6). Reshaping Business with Artificial
Intelligence. MITSloan Management Review. https://sloanreview.mit.edu/projects/reshaping-business-
with-artificial-intelligence/
35 R
 ansbotham, R., Kiron, D., Gerbert, P., & Reeves, M. (2017, September 6). Reshaping Business with Artificial
Intelligence. MITSloan Management Review. https://sloanreview.mit.edu/projects/reshaping-business-
with-artificial-intelligence/
36 F
 itzgerald, S., and Parker, R. (2017, November 1). IDC Reveals Worldwide Digital Transformation Predictions.
International Data Corporation. https://www.idc.com/getdoc.jsp?containerId=prUS43188017
37 A
 ccenture Technology Vision. (2017). Technology for People: The Era of the Intelligent Enterprise. https://
www.accenture.com/t20170530T164033Z__w__/us-en/_acnmedia/Accenture/next-gen-4/tech-
vision-2017/pdf/Accenture-TV17-Full.pdf
38 P
 erez, S. (2017, April 20). Amazon’s New Alexa Developer Policy Now Bans ALL Ads Except in Music and
Flash Briefings. TechCrunch. https://techcrunch.com/2017/04/20/amazons-new-alexa-developer-policy-
now-bans-all-ads-except-in-music-and-flash-briefings/
39 Kelnar, D. (2017). The State of AI 2017: Inflection Point. MMC Ventures. https://www.mmcventures.com/
wp-content/uploads/2017/10/The-State-of-AI-2017-Inflection-Point-Summary.pdf
40 Accenture Technology Vision. (2018). Redefine Your Company Based on the Company You Keep: Intelligent
Enterprise Unleashed. https://www.accenture.com/t20180208T172438Z__w__/us-en/_acnmedia/
Accenture/next-gen-7/tech-vision-2018/pdf/Accenture-TechVision-2018-Tech-Trends-Report.pdf
41 Which.(2017, November 20) Accenture Launches Pinterest for Education. Which-50. https://which-50.
com/accenture-launches-pinterest-ai-education/
42 “ Gunning, G.(n.d.) Explainable Artificial Intelligence. DARPA. https://www.darpa.mil/program/explainable-
artificial-intelligence“
43 Accenture. (2017). Harnessing the Revolution, Creating the Future Workforce. https://www.accenture.
com/us-en/_acnmedia/A2F06B52B774493BBBA35EA27BCDFCE7.pdf
44 V
 ox. (2017, September 19) The rise of robots in the German labour market. https://voxeu.org/article/rise-
robots-german-labour-market
45 Gartner. (2017, December 13). https://www.gartner.com/newsroom/id/3837763
46 D’Onfro, J., (2015, January 13) The Clever Way Amazon Gets Away with Not Always offering the Lowest
Prices. http://uk.businessinsider.com/how-amazon-adjusts-its-prices-2015-1

ARTIFICIAL INTELLIGENCE | 91
 HAVEN’T WE BEEN HERE BEFORE?

ABOUT ACCENTURE
Accenture is a leading global professional services company, providing
a broad range of services and solutions in strategy, consulting, digital,
technology and operations. Combining unmatched experience and
specialised skills across more than 40 industries and all business functions—
underpinned by the world’s largest delivery network—Accenture works at
the intersection of business and technology to help clients improve their
performance and create sustainable value for their stakeholders. With more
than 435,000 people serving clients in more than 120 countries, Accenture
drives innovation to improve the way the world works and lives. Visit us at
www.accenture.com.

APPLIED INTELLIGENCE
Accenture Applied Intelligence, part of Accenture Digital, applies artificial
intelligence (AI) and human ingenuity at the core of business to help clients
become intelligent enterprises and solve their most complex business
problems. By deploying AI responsibly and combining it with our deep
industry and analytics expertise, we enable the digital transformation of
organisations, extend human capabilities, and make intelligent products and
services a reality.

Follow @AccentureAI and visit accenture.com/appliedintelligence.

Copyright © 2018 Accenture

All rights reserved.

Accenture, its logo, and High Performance Delivered are trademarks of Accenture.

92 | ARTIFICIAL INTELLIGENCE