Commercialization of AI 3.0

COMMERCIALIZATION OF AI 3.0
WHITEPAPER

OVERVIEW
ARTIFICIAL INTELLIGENCE - BUZZWORD OR GAME CHANGER?
THE BASICS
Where does AI come from, where does it go?
FACTS & FIGURES
Round-up of science, applied AI and talent
COMMERCIAL VALUE
Strategies of B2B platforms and practical action points
2
3
9
18
About this whitepaper
Artificial Intelligence is trendy. Every event, every strategy meeting and every
consulting firm talks about it. This whitepaper aims to separate actual facts
and important background information from the overarching marketing buzz.
You will get a short but information-rich wrap up about: What causes the
current hype? Where are we today? What are the innovation leaders doing
with AI? And what are immediate action points to focus on by applying
artificial intelligence in your business.
WHERE CAN TECHNOLOGY TAKE
YOU?
Appanion in a nutshell
We provide high-quality insights on real-world technology-driven business
applications and help you as innovation partner with actionable strategies,
technology knowhow and business model competence.
I hope you can take away some
valuable insights and new ideas
to improve your business!
Tobias Bohnhoff
Founder

ARTIFICIAL INTELLIGENCE IS CAPABLE OF FAR MORE
THAN JUST PLAYING CHESS, JEOPARDY OR GO
Artificial Intelligence (AI) is the ability of a
computer-controlled entity to perform
cognitive tasks and react flexibly to its environment
in order to maximize the probability of achieving
a particular goal.
The system can learn from experience data,
and can mimic behaviors associated to humans,
but does therefore not necessarily use
methods that are biologically observable.
Definition is based on McKinsey, Stanford University, Wikipedia, European Commission 4
„
“
Artificial General Intelligence (AGI)
Artificial Intelligence (AI) can be
distinguished into systems designed
to solve problems in a very specific
context – the so called Narrow AI –
and a human-like Artificial General
Intelligence (AGI).
AGI is currently only a future
scenario. Some experts say it is
decades away to build such an AI,
some argue that humans will never
be able to build something at an
equal level of intelligence.
Most often, AGI scenarios outline a
dystopia for the human race due to
the uncontrollable effects this
technology potentially has.
Thought leaders in the AI space
assume that at some tipping point an
AGI would act without any
interventions of humans and that
setting up a global governance of AI
development is critical right now.
Artificial Narrow Intelligence (ANI)
Narrow AI on the other hand is
already here. It is automating or
augmenting many tasks in a business
context or in our private life.
Self-driving forklifts in warehouses,
Amazon product recommendations,
real-time trades on stock exchanges,
interacting with Siri, person
recognition in the iPhone photo
album – everything is already done
automatically, without the need of
human intervention.
It’s worth mentioning that
automation per se does not require
intelligence. This can also be done by
pre-programmed rules. However,
complex tasks require the ability to
react flexibly to their environment,
which differentiates AI applications
from the well-understood domain of
rule-based automation.

AI IS ENTERING MASS COMMERCIALIZATION - EVERY
BUSINESS IS GOING TO CHANGE
Startup perspective
Artificial Intelligence promises
disruption. It opens up the possibility
to achieve better product or service
quality at less time and cost. And if
this can be proven to the market
successfully, it will change everything.
Customers start to jump from legacy
solutions to better products in a very
short timeframe, others follow to
adopt, old solution providers die and
the new players overtake the market
lead. That is what we call disruption
and what makes AI as a space so
attractive for the many young
companies in the recent years.
Investor perspective
Platform and as-a-service business
models have been proven very
successful due to recurring revenues
and scalability. To investors, owning a
new market and earning high
margins means to be fast and to take
a lot of financial risk. As VC funding
data shows, the race for number one
in AI has just taken off. At some point,
the market will start to consolidate
and the winning companies will
emerge.
Corporate perspective
Almost all large corporations are
confronted with entirely new
processes through AI. Its value lies in
automation and augmentation of
human work. The technology not only
changes the frontend or digitizes the
communication interface, it changes
the entire way of collaborating,
planning and decision-making.
The flexibility to change a – to this
point – successful concept, is crucial
to survive. Starting early experiments,
communicating change and investing
into research and talent is the way
forward.
1) Source: CB Insights (02/2018)
2) Source: CB Insights (01/2018)
3) Source: AIIndex.org (11/2017)
4) Source: Technews.io (09/2018)
5
20152013 2014 2016 2017
1.739
4.5693.477
6.255
15.242
+72%
2020
10
20101995 20152000 2005
0
5
15
20
+12%
Venture Funding in AI startups
In million U.S. dollars, global1
AI related research publications
In thousands, global3
Number of AI press mentions
Original media coverage by tech journalists4
20152014 2016
8.772
2017 2018
42.973
14.670
90.752
125.462
+94%
Startups with „.ai“ URL suffix
Global # of startups that received financing2
5 22
51
100
225
201720162013 2014 2015
+159%
Compound annual growth rate Compound annual growth rate
Compound annual growth rateCompound annual growth rate

THE AI HISTORY IS IMPORTANT TO UNDERSTAND
WHERE WE ARE TODAY AND WHAT IS DIFFERENT
AI 1.0 – Pattern Recognition
Since the term ‚AI‘ was established in
in 1956, government agencies like
DARPA1 funded scientific research
until the mid-1970s. Successes in the
fields of search, natural language
processing and simplified models and
simulations were achieved.
The tremendous expectations raised
by AI researchers afterwards didn‘t
match the achievements for the
investors. Limited computing power,
the rather low performance of neural
networks, combinatorial explosion
and a lack of commonsense
knowledge by the systems led to
significant cuts in funding. The first AI
winter lasted until 1980.
In the 1980s, the expert system
emerged. It used logical rules, derived
from the knowledge of domain
experts. Avoiding the challenges of
commonsense logic, funding came
back until 1987. The rise of IBM‘s and
Apple‘s desktop computers made
specialized AI hardware obsolete. As
a consequence, markets went down
and trust was gone – the second AI
winter.
AI 2.0 – Deep Learning
In the mid-1990s milestones were
achieved that brought back trust.
Chess champion Garry Kasparov was
beaten by ‘Deep Blue’ in 1997.
Intelligent agents that perceive the
environment and take actions to
maximize successful outcomes were
the leading paradigm to that time.
In addition, the deep learning concept
was developed, but it took another 10
years until the availability of big data
and computational power from
graphic processing units (GPU)
created the „big bang“ of deep
learning algorithms.
1) DARPA = Defense Advanced Research Projects Agency (Part of the United States Department of Defense) 6
AIPERFORMANCE
TIME
HIGHLOW
AI 1.0
Pattern
Recognition
AI 2.0
Deep
Learning
AI 3.0
Contextual
Reasoning

AI 3.0 IS COMING - BUT IT IS NOT
EQUIVALENT TO SUPERINTELLIGENCE
AI 3.0 – Contextual Reasoning
AI 2.0 is slowly fading out as new
breakthroughs in scientific and
applied AI demonstrate the
superiority of computer intelligence
compared to human intelligence in
selected domains.
It started with chess, continued with
Jeopardy, Go and is now at the point
of beating humans in collaborative
multiplayer games like Dota.
At the same time medical diagnostics
surpasses the accuracy of human
doctors in many domains such as
cancer detection from x-ray scans.
Also, the dependency on AI-powered
systems in our daily life (private and
business) increases rapidly. Search
engines, industrial robots, real-time
trading systems and soon also
autonomous mobility and delivery
services cannot be substituted by
humans anymore.
If you would turn off the AI-engine in
the future, the world would most
likely break apart. That is why many
smart thinkers urgently warn of the
consequences of too powerful
systems. Most of them agree that AI
will be uncontrollable for humans and
the question is, if humans can
gracefully co-exist with an artificial
super intelligence.
AI 3.0 is at this point not a synonym
to super-intelligence or singularity.
Contextual reasoning describes the
fact that most cognitive processes
depend on the environment and
therefore the context. Systems that
perform tasks in silos without
incorporating outside information are
already in use but they can be easily
identified as computer applications
with limited capabilities to respond to
exceptions and changes in their
environment.
7
The new spring in AI is the most
significant development in computing
in my lifetime.
Every month, there are stunning new
applications and transformative new
techniques. But such powerful tools
also bring with them
new questions
and responsibilities.
Sergey Brin
Co-Founder of Google
„
“

THE BREEDING GROUND FOR AI IS THERE – THE
QUESTION IS HOW FAST IT WILL GROW
8
Where are we today?
Even if there are already disciplines
where narrow AI systems outperform
human capabilities, in most areas of
application that are not repetitive,
rule-based or require simple
knowledge representation, humans
still do better. AI systems rather
augment tasks to be more secure,
work faster or deliver higher quality.
Why is it different now?
It is quite unrealistic that AI will enter
another ‘winter’ very soon because
multiple critical requirements are
fulfilled by the market:
1) Financing – AI is no longer in a
scientific silo that is depending on
external financial resources. With
successfully applied and monetized AI
applications, more and more money
floods into the system to keep the fire
burning.
2) Big data – led to the deep learning
big bang – but quantity is not
everything. With sophisticated data
warehouses, the quality of data
improves rapidly and so does the
output of AI applications.
3) Computational power – increased
significantly by using graphical
chipsets, but this is just the beginning.
Performance is likely to improve by
orders of magnitude in a short
timeframe with Google’s specialized
tensor processing unit (TPU) already
in the market and quantum
computing casting its shadows ahead.
4) Global collaboration – It’s not part
of a national strategy anymore to
decide about investing in research
and hiring talent. It’s a global task
with ambitious players – may that be
countries or corporates – to become
the most successful and influential
party in this collaborative
competition.
It’s likely that machines
will be smarter than us
before the end of the century
—
not just at chess or trivia
questions but at just about
everything, from mathematics
and engineering to science
and medicine.
Gary Marcus
Professor New York University
„
“

9
FACTS &
FIGURES
LEVEL UP YOUR COFFEE TALKS

TRANSFER LEARNING SPEEDS UP LEARNING
SUCCESSES AND IMPROVES SCALABILITY
1) Source: Machine Learning Mastery (2017) – exemplary visualization 10
Research Advancement #1: Transfer Learning Performance of untrained / transfer learning models1
What does it mean?
Transfer learning is a machine learning method where a model developed and
trained for task A can be reused for task B by re-applying the already
acquired fundamental knowledge. The model for task B then already starts
learning already at a basic level of knowledge.
Why does it matter?
Re-using knowledge significantly reduces the required amount of data for
a model to learn a new task. Especially when data is hard to capture. It brings
not only faster results, it also achieves higher accuracy after the same
amount of training.
training
performance1
Applied transfer learning
Untrained model
higher
start
steeper
learning
curve
better output
AI RESEARCH

AFFORDABLE HIGH-END CHIP PERFORMANCE IS THE
KEY TO BRING AI TO THE MASS MARKET
1) Source: Baidu (2018) 11
Research Advancement #2: Chip Performance Speed benchmark for AI training with one or multiple GPUs
What does it mean?
Today, much of the processing power for training an algorithm is done with
graphics processing units (GPU), instead of central processing units (CPU). The
computational parallelism of the higher performing graphic chipsets
allows significant improvements of results as shown on the right.-
Why does it matter?
Cost efficient availability of GPU-based computational power was a game
changer in AI research. Training time of AI models decreased significantly.
Training a model with 50 GPUs is 40 times faster compared to one single GPU
unit.
1 2 3
8
15
40
56
0
5
10
15
20
25
30
35
40
45
50
55
60
2 4 50 100
# GPUs used for training
Training speedup over one GPU (x-fold)1
161 8
AI RESEARCH

RELIABLE AI-SYSTEMS REQUIRE CONTEXTUAL
ADJUSTMENTS UNDER CHANGING CONDITIONS
1) Source: ImageNet, Standford Vision Lab 12
Research Advancement #3: Contextual Learning Error rate development on ImageNET visual recognition
What does it mean?
Detecting objects in images or videos based on the sematic structure of pixels
and classification was very challenging in the field of computer vision.
Common sense logic of actions or context was missing. Due to advances in
convolutional neural network architectures and reinforcement learning,
AI systems improved a lot in the recent years.
Why does it matter?
Precise detection and classification of the environment data is key to
many applications such as autonomous driving. Moreover, contextual learning
is also relevant in natural language processing supporting smart assistants
like Siri.
0%
5%
10%
15%
20%
25%
2012
Error-rates on ImageNet1
2011 2013 2014 2015
human
level
AI RESEARCH

ADVANCED DIAGNOSIS SUPPORT HUMAN DOCTORS
AND PROGRESS ATTRACTS MORE INVESTMENT
1) Source: CB Insights 13
300
790
20162015
1.200
2017 HY 18
1.100
VC Funding in AI healthcare startups
Disclosed equity funding in million U.S. dollars (rounded)1
AI healthcare news in Q3 2018
Selected press mentions of successful AI trials or implementation projects
ARTIFICIAL
INTELLIGENCE
IN
HEALTHCARE
APPLIED AI

SMART AUTOMATION IS ON THE RISE, AI-POWERED
ROBOTS TRIGGER BILLION-DOLLAR INVESTMENTS
1) Source: International Federation of Robotics 14
Supply of industrial smart robots worldwide
In thousand units (forecasted)1
AI manufacturing news in Q3 2018
ARTIFICIAL
INTELLIGENCE
IN
MANUFACTURING
178
221
254
294
346
378
433
521
2013 2018*2014 2015 2016 2019*2017* 2020*
APPLIED AI

AUTONOMOUS WAREHOUSE-LOGISTIC ACCELERATES
THROUGH AI IN DISTRIBUTION CENTERS
1) Source: Amazon 15
Number of robots working in Amazon fulfilment centers
in global units1
AI logistics news in Q3 2018
ARTIFICIAL
INTELLIGENCE
IN
LOGISTICS
20152013 2016
30.000
120.000
2014 2017
1.400
15.000
45.000
APPLIED AI

THE IMMENSE NEED FOR AI EXPERTS DRIVES
SALARIES TO ASTRONOMICAL HEIGHTS
Size of the AI talent pool
The Global AI Talent Report states a
total number of 22,000 PhD-level AI
Researchers1. The data was derived
from LinkedIn queries and analyses
of the leading conferences. The scope
focuses on highly specialized AI
experts rather than counting entire
software development teams.
Distribution of talent has a strong
focus on native English-speaking
countries (U.S. followed by U.K.,
Canada and Australia) biased by the
research methodology. Especially
China is underrepresented, while
France, Germany and Spain
complement the significantly smaller
European counterpart.
Chinese tech giant Tencent sets a
broader focus and counts 300,000
active AI researchers and
practitioners globally. Two-thirds of
them employed in the industry and
100,000 researching in academia.
Salaries for AI professionals
According to figures from job listings
platform Indeed, average salaries for
AI programmers in the U.S. range
between $134k and $170k2. On
Glassdoor, the average annual salary
for AI listed jobs is at $111k, including
a broad range of AI related jobs in
startups or corporates.
Real domain experts like software
architects and engineers can easily
earn three to four times as much.
Especially the large tech companies
such as Google or Facebook pay from
$300k to $500k3 in base salary and
company stock a year even for
university drop-outs.
Public figures of DeepMind show
staff costs of $138 million, which
equals $345k for each of the 400
researchers. For leadership positions
in AI domains the sky is the limit. Not
only in the U.S. – this is also true for
companies in China, Korea or Europe.
1) Source. Element AI (2018) - http://www.jfgagne.ai/talent
2) Source: Medium (2018) - https://medium.com/mlmemoirs/artificial-intelligence-salaries-heading-skyward-e41b2a7bba7d
3) Source: NY Times (2018) - https://www.nytimes.com/2017/10/22/technology/artificial-intelligence-experts-salaries.html
16
Global AI talent distribution
PhD Researchers based on LinkedIn and conference anaylses1
AI TALENT

TECH COMPANIES FIGHT WAR OVER TALENTS FROM
TOP AI UNIVERSITIES
Industry vs. academia
Skilled computer and data scientists
that understand the complex
mathematical techniques to develop
artificially intelligent software and use
it to improve business are very rare.
This is why big tech companies
aggressively recruit talents from top
universities. In 2015, Uber hired 40
students out of Carnegie Mellon to
boost its self-driving car ambitions.
Even professors are increasingly
recruited to work in the industry.
Four professors from Stanford and
six from the University of Washington
were at least temporarily on a leave
in the past years.
At this point, the talent hunger of the
industry already affects the following
classes in the educational system. But
Google, Microsoft and IBM were just
the start. Every well-funded Silicon
Valley company and increasingly also
old-economy players enter the AI
talent war.
According to a Stack Overflow survey,
86.7% of developers are self-taught3
(not specified on the AI space). Given
this fact, online course offerings like
Coursera and Udacity as well as
corporate programs will most likely
continue to grow rapidly to meet the
demand.
1) Source: U.S. News (2018) - https://www.usnews.com/best-graduate-schools/top-science-schools/artificial-intelligence-rankings
2) Source: State of AI (2018) - https://www.stateof.a
3) Source: Developer Survey Results 2018 - https://insights.stackoverflow.com/survey/2018
17
1
Carnegie Mellon University
Pittsburgh, Pennsylvania (USA)
Students: 12,500
2
Massachusetts Institute of Technology
Cambridge, Massachusetts (USA)
Students: 11,319
3
Stanford University
Stanford, California (USA)
Students: 16,430
4
University of California
Berkeley, California (USA)
Students: 41,910
5
University of Washington
Seattle, Washington (USA)
Students: 46,686
Top universities for artificial intelligence research
Based on survey results of academics in the United States1
Leading employers of AI talent
Size of employed AI staff2
900
450 400
300
1.400
1.000
„There is a giant sucking sound of
academics going into industry“
Oren Etzioni
CEO of the Allen Institute for AI
AI TALENT

18
COMMERCIAL
VALUE
STARTUPS, PLATFORMS AND ACTION POINTS

19
AI STARTUP ACQUISITIONS BY THE TOP B2B
CLOUD PLATFORM VENDORS
Source: CB Insights, CrunchBase (as of October 2018)
Excluded: Chinese platform vendors such as Baidu and Alibaba
TIME
2014 2015 2016 20172013 2018

20
AI STRATEGIES OF THE LEADING
B2B DATA PLATFORMS
INFRASTRUCTURE MATURITY
LOW HIGH
FIRST MOVERS
LEADERSVISIONARIES
EARLY FOLLOWERS
STRATEGICCOMMITMENT
HIGHLOW
Google’s AI strategy is to create a
strong market position with
corresponding patents and related
technology areas as its business
relies heavily on machine learning.
With TensorFlow, Google developers
delivered the leading open-source
software library in the ML space. This,
alongside with their investment
strategy in AI start-ups, shows
Googles dedication to be and stay the
leader in AI development.
Microsoft - Since 2016, the company
keeps the pace of Google with
increasing acquisition activity. And
still, the company's entire AI efforts
are well linked with the Azure cloud. It
will be crucial to seamlessly integrate
the technology from start-ups into
the company’s portfolio.
Amazon will continue to build
around the successful Alexa project
with voice, virtual assistants and
natural language processing. AI-as-a-
service is pushed into the AWS
environment to make Amazon a
leader in this field as well.
IBM's AI strategy is lazor-focused on
its enterprise customers. In giving
them the control of their data and
insights, IBM tries to assist them
increase efficiency, lower costs or
augment human intelligence.
Simultaneously, open source projects
in the fields of AI are supported and
APIs to other vendors like Google's
TensorFlow are endorsed.
Salesforce’s Einstein AI platform has
been the central focus for the
development and marketing strategy.
CEO Benioff described the AI strategy
quite straight forward: ‘AI is the next
platform – all future applications, all
future capabilities for all companies will
be built on AI.’
SAP - Cloud surpassed the licensing
business for the first time in 2018. In
addition, acquiring Recast.AI and
creating Europe’s first corporate AI
ethics advisory panel clearly positions
AI as a central strategy building block.

21
FIVE ACTION POINTS TO CREATE SUSTAINABLE VALUE
FROM ARTIFICIAL INTELLIGENCE
TALENT ACQUISITION
▪ Data scientists, AI engineers and software architects are the scarcest resources in terms of development speed as of today
▪ Make sure to create a vibrant and attractive environment to make smart people want to work for you
▪ Start early to recruit and be competitive in talent acquisition, otherwise you can’t expect to outpace your competition in this domain
DEVELOP A COMPANY-WIDE DATA STRATEGY
▪ Data is the raw input to every AI application and determines the quality of the output
▪ Unstructured data held in silos with multiple decision-making stakeholders slows the adoption process
▪ Define and implement a coherent and flexible data architecture and set responsibilities on the executive-level
EMBRACING A TECHNOLOGY MINDSET
▪ Start to develop iteratively and find ways to apply this logic to mission critical processes as well
▪ Openly communicate and promote advances and changes through technology to avoid fear and opposition building
▪ Allow new and radical ways of thinking and encourage experiments and failure
EVALUATION OF NEW BUSINESS MODELS
▪ Think holistic! Consider how technology impacts and changes the way partners, customers, competitors and new market entrants act
▪ Be open to even license proprietary solutions to other market players in order to create new revenue streams
▪ Question current processes and evaluate external as-a-service offerings to keep focus on your core product
START TODAY
▪ Value creation of AI is strongly correlated with the experience in this domain – the sooner you start, the higher your return
▪ Facilitate ideas in your company by small-scale prototypes or proof-of-concept projects
▪ Invest time into the exploration of possibilities that are out there already
1
2
3
4
5

22
LET’S TALK ABOUT BOOSTING YOUR BUSINESS
WITH ARTIFICIAL INTELLIGENCE!
We identify suitable AI use cases and help
with target-oriented prioritization.
IDENTIFY
We create a concrete plan about
requirements, project organization and
feedback-driven iteration steps.
PLAN
We evaluate suitable solutions for your
company-specific requirements
EVALUATE
We facilitate the development and
implementation process and support you in
the selection of partners.
EXECUTE
We provide decision confidence through
process design and functional prototypes
TEST
We develop a future-proof data strategy
based on the company's long-term goals
SUSTAIN

GET IN TOUCH!
Tobias Bohnhoff
Appanion Labs GmbH ▪ Hopfenstrasse 11 ▪ 20359 Hamburg ▪ www.appanion.com
Discover more at
www.appanion.com
Disclaimer: This whitepaper is based on research and data of the previously mentioned sources. All information presented were researched and prepared by Appanion with great care. For the presented data and information Appanion cannot assume any warranty of any kind. Information and data represent in selected cases individual opinions
and may be in need of further interpretation as a basis for decisions. Advances in research may occur after the publishing of this document. Therefore, Appanion is not liable for any damage arising from the use of information and data provided in this whitepaper.

Commercialization of AI 3.0

Related slideshows

Recommended for you

Recommended for you

Recommended for you

Recommended for you

Recommended for you

More Related Content

What's hot

What's hot (20)

Similar to Commercialization of AI 3.0

Similar to Commercialization of AI 3.0 (20)

More from APPANION

More from APPANION (6)

Recently uploaded

Recently uploaded (20)

Commercialization of AI 3.0