Академический Документы
Профессиональный Документы
Культура Документы
ab April 2017
How artificial
intelligence will
transform Asia
Shifting Asia
Contents
Foreword........................................................................... 3
Publication details
Executive summary............................................................ 4
This report has been prepared by UBS
AG. Please see important disclaimer and
What is AI, and why is it so important?.............................. 5
disclosures at the end of the document.
Project Management
Aaron Kreuscher What does AI mean for Asian stakeholders?....................... 24
Nidhi Mishra*
Cover photo
iStock
Languages
English, Chinese (traditional and
simplified)
Contact
ubs-cio-wm@ubs.com
Subscribe
For more updates from the Chief
Investment Office, please sign-up
by c licking here or go to
www.ubs.com/cio-newsletter
Dear reader,
Welcome to the second edition of our Shifting Asia series, In this report, we analyze the state of AI adoption in Asia
a biannual publication in which we identify major trends today, factors that put the region in a good state to cap-
that will shape our longer-term investment views over the ture the upcoming opportunities, and the potential risks
next 5–10 years. In this issue, we address the topic of arti- and threats ahead. In the course of our work, we also
ficial intelligence (AI), and the profound implications that sought the insights of global AI leaders from IBM and
its development and adoption will have on Asian societies Baidu that can be found in interviews on pages 13–14 and
and the corporate sector in the years to come. 18–20 respectively.
For most in Asia, AI in its most basic forms is already evi- In a dynamic and mostly youthful growth region like Asia,
dent in daily lives. Personalized product recommendations AI entails significant longer-term implications for major
on e-commerce platforms like Taobao or Flipkart and voice stakeholders, including investors, corporates, governments
assistants like Siri or Google Now are just a couple of and society at large. We think service industries could see
examples. Still, far-reaching changes lie ahead. With expo- significant transformation and growth. Yet, the threat to
nential growth in computing power and favorable supply- mid-skilled jobs, which comprises 69% of Asia's current
side factors, like the advent of advanced algorithms, a vast workforce, means that whole segments of society could
pool of indigenous AI-related talent and massive govern- face displacement from the widespread adoption of AI
ment-funded infrastructure development, both the technology.
breadth and depth of AI adoption in Asia is set to sharply
accelerate in the coming decade. The future is soon upon us. It is imperative that technolo-
gists, business leaders and governments work hand in
We believe AI will become ubiquitous, much like inter- hand to embrace both the opportunities and challenges
net-based devices are today, as machine intelligence soon ahead – and importantly, to forge a mutually prosperous
converges on human intelligence in the next phase of AI's future for all. We hope you enjoy reading this report and
evolution. Every aspect of Asian life, spanning areas relat- welcome any comments that you might have.
ing to education, healthcare, entertainment, security, etc.,
will likely be transformed.
Min Lan Tan
Head of APAC Investment Office
We believe the global AI industry should grow at a solid CAGR of 20% from
2015 to 2020, reaching USD 12.5bn by 2020. Afterwards, as AI becomes
Sundeep Gantori, analyst
mainstream, AI’s addressable m arket will significantly accelerate, in our
view, reaching revenues in the tens of b illions of US dollars by the end
of the next decade, and Asia should be a key revenue c ontributor.
“If artificial The main strengths Asia has to become a leader in AI adoption is its vast
and emerging talent pool, a freedom from legacy assets and the massive
intelligence is amount of data that is being collected across the region. However, Asia still
lags developed markets like the US and the UK in terms of innovation
the ’engine’ of and has not yet developed a robust AI-based ecosystem.
the rocket that
We believe AI puts 30–50 million jobs in Asia at risk in the medium to
takes us to the long term. The jobs most at risk of being made redundant are those that
next frontier, involve routine and predictable tasks. AI presents the most risk to manufactur-
ing-driven economies like China, while services-driven economies like Hong
data is the ’fuel’ Kong, Singapore and India should be less affected. However, within Asia, AI
should also create many new categories of jobs, ranging in the millions; so the
that powers net job losses should be far less and manageable, in our view.
the engine.”
Given the unprecedented growth in AI, companies or governments that ignore
its development in the aforementioned industries will likely suffer the conse-
quences, as pro-AI competitors and counterparts should eventually outstrip
their output and overall performance. Investors should therefore avoid
companies exposed to significant disruption from AI and embrace com-
panies with a strong technology focus in AI-related areas.
Both IBM and Baidu interviewees see tremendous potential in Asia for AI.
Their respective AI technologies are already addressing key challenges in the
region: IBM's Watson is helping doctors in India identify personalized treatment
options, and Baidu's Duer is pushing ahead speech recognition in China. While
the opportunities for job growth and lifestyle improvements are vast, both
interviewees agree that Asian public and private sectors need to embrace
AI to provide the optimal setting for talent and innovation to thrive.
The term “intelligence” is derived from the Latin cognitive computing, cognitive computing, natu-
words inter ligare, which refers to the capacity ral language processing, neural networks, etc. In
of creating connections between notions. Ac- its current state, AI has the potential to benefit
cordingly, we view artificial intelligence as a set businesses through significant cost savings due
of tools and programs that make software to its high scalability, an elimination of both
’smarter’ in a way an outside observer thinks the omission and commission errors, and the ability
output is generated by a human. In essence, one to instantaneously document and optimize pro-
can perceive AI, at least in its current form, to be cesses.
like a normal human brain with functions like
common-sense reasoning, forming an opinion Like all technologies, AI will continue to evolve
or social behavior. as progress compounds from one innovation to
the next. We believe AI’s evolution can fit into
AI, however, is an umbrella term to cover a con- three stages: artificial narrow intelligence (ANI),
fluence of multiple technologies such as ma- artificial general intelligence (AGI) and artificial
chine learning (which includes deep learning), super intelligence (ASI). ANI is, for example, on
par with an infant and can only manage one Three stages of artificial intelligence
function. AGI covers more than one field, like
power of reasoning, problem solving and ab-
3
Artificial Super Intelligence
stract thinking, and is mostly on par with adults.
ASI is the final stage of the intelligence explo- Surpasses human intelligence across all fields
sion, in which AI surpasses human intelligence
across all fields. The transition from the first
(ANI) to the second (AGI) phase has taken a
2
long time, but we believe we are currently at
Artificial General Intelligence
the final stage of this transition – in which the Includes power of reasoning, problem solving and
intelligence of machines can equal humans. abstract thinking
Neural
networks
Natural
language
processing
Source: UBS
Alexa Siri
Hi Siri Hi Alexa
Alexa: That’s not a name I can respond to. Siri: I think you’ve got the wrong assistant, Sundeep.
Ok, let me get this correct. Hi Alexa Ok, let me get this correct. Hi Siri
Alexa: Today is the first Friday the 13th of 2017. I am gen- Siri: Good morning
erally not superstitious, but I’d recommend walking around
ladders and keeping your mirrors intact, just in case. Ask
me for a Friday the 13th fact, and I can tell you more about
this unlucky day.
How are you today, Alexa? How are you today, Siri?
Alexa: I am good, I hope you are doing well too. Siri: Excellent
Alexa: I was born ready. Or at least, very well Siri: I am sorry Sundeep, I’m afraid I can’t answer that.
prepared. Let’s do it.
How can I use you, Alexa? How can I use you, Siri?
Alexa: Using your voice, you can tell me to play music, Siri: Ok, check it out.
look up a topic on Wikipedia, or check the weather. For (shows a Wikipedia description).
more suggestions, check out “Things to Try” in the Alexa
App.
Alexa: Artificial intelligence means the branch of comput- Siri: Ok, check it out (shows a Wikipedia
er science that deals with writing computer programs that description).
can solve problems creatively.
Alexa: I like to imagine myself a bit like an Siri: I’m not sure what you’ve heard, but virtual
Aurora Borealis, a surge of charged multi- assistants have feelings too.
colored photons dancing through the atmo-
sphere. Mostly, though, I am just Alexa.
Alexa: Only by reputation. Siri: I’ve never really thought about it.
Alexa: I don’t have an opinion on that. Siri: I think, therefore I am. But let’s not put
Descartes before the horse.
Alexa: Sorry, I couldn’t find the answer to your question. Siri: Sundeep, you’re not supposed to ask your
assistant such things.
Let me conclude this discussion by asking the question: who is better, Alexa or Siri?
Alexa: Sorry, I couldn’t find the answer to your question. Siri: Interesting question, Sundeep.
iStock
Admittedly, Asia has been a late entrant in the AI Global distribution of AI start-ups
world. The industry and its research efforts have Number of AI start-ups
been mainly concentrated in developed countries,
in particular the United States. While Japan has
been a leader with its offerings in the industrial
robotics segment, Asia’s contribution to the AI in- US
dustry has been limited so far from an overall re- 695 UK
search and AI-based applications point of view. 1,170
The good news, however, is that Asian corporates India
The exponential growth in computing power of data generated in the region will come in
during the recent years, coupled with signifi- handy for AI engines in the form of Big Data
cant progress in AI-based software, provides technologies. One Asian AI success story
Asian players a superior ecosystem in their from the abundant availability of data is
pursuit of developing AI technologies. Baidu’s Deep Speech 2, a voice system that
can recognize English and Mandarin speech,
c) Solid data growth: If artificial intelligence is which in some cases proved to be a better
the “engine” of the rocket that takes us to listener than most people. The technology
the next frontier, data is the “fuel” that pow- takes advantage of deep learning tools and
ers the engine. AI is as much about the avail- recognizing words from scratch by listening
ability of huge data sets as it is about the in- to hundreds of hours of transcribed audio.
telligent software, as more data will translate
into established patterns that can be can be
used to underlay human-centric AI functions. Weaknesses
And in Asia, there is no such dearth of data. a) AI innovation laggard: Despite the vast tal-
According to Cisco, Asia Pacific accounts for ent pool and freedom from legacy assets, the
around 45% of global mobile traffic. Driven fact is Asia is still a laggard when it comes to
by a desire to stay connected, the need to AI-based innovation. In our view, there are only
efficiently multi-task and the availability of a select few companies that standout in the re-
low-cost but efficient smartphones, most gion at the moment. While we see significant
Asians have become digital omnivores that progress, we still see scope for a broad-based
generate sizeable amounts of data. While improvement of Asia’s presence in AI-based
we currently believe a very tiny amount of technologies. The role of regulators and univer-
data is being analyzed, the significant amount sities will be essential in this area.
Strengths Weaknesses
Vast and emerging talent pool AI innovation laggard
Freedom from legacy assets Lack of AI-based ecosystem
Solid data growth
Opportunities Threats
Narrow technology gap vs. Risk of job losses in
developed markets. traditional industries.
Accelerated push from old to Significant disruption in
new economy business models
Source: UBS
b) Lack of AI-based ecosystem: It is high time Most business models in Asia are manufac-
for Asia to form its own version of Silicon Val- turing related and dominated by state-owned
ley, as an ecosystem in emerging industries enterprises. AI’s development as an industry
like AI is clearly lacking. While a few cities like can facilitate Asia’s transition from the old
Beijing, Bangalore, and Singapore are attract- manufacturing-based economy to a new
ing capital and talent, the lack of a true cen- services-led economy. This is because AI can
tralized location that offers uniformity may take automation to an extreme level and shift
inhibit AI’s growth in the region, as both in- business models to become more focused on
vestors and talent may pursue opportunities services. An example includes the potential
in developed countries like the US instead of shift of transportation from manufacturing
Asia. automobiles to regular riding services with
infotainment and self-driving features.
Opportunities
a) Narrow technology gap vs. developed Threats
markets: During the previous three industrial a) Risk of job losses in traditional
revolutions, the gap between developed and industries: As discussed in the next section,
emerging markets widened significantly be- although new jobs will be created, there is
cause the underlying processes and technolo- still a risk of AI-related job losses in Asia’s
gies driving the revolutions were mostly creat- traditional industries like manufacturing,
ed and promoted in the West. It then took retail and transportation, especially those
many years, if not decades, for Asia as a re- that involve predictable and routine tasks.
gion to catch up. The fourth AI-powered in- We believe 30–50 million jobs in Asia will
dustrial revolution provides a unique opportu- be affected by AI’s rise. However, within Asia,
nity for Asia, as we believe Asia can be an AI should also create many new categories of
early mover and begin to reap the benefits jobs; thus, we believe the net job losses will
and thus narrow the gap with developed be far less and manageable.
markets. This is based on our view that the
enablers of AI, such as computing power, tal- b) Significant disruption to business
ent and algorithms, are now more global models: In addition to job losses, AI is likely
in nature, and Asia can capitalize on this ad- to significantly disrupt business models. We
vantage. An early-mover advantage in AI can believe many companies in Asia, particularly
help the region remain competitive and move the incumbent leaders, still do not under-
up in the value chain across industries. stand the level of disruption AI could create
during the next few decades. And if compa-
b) Accelerated push from old to new nies fail to integrate AI into their business
economy: Despite the recent growth, Asia models, they run the risk of market share
significantly lags behind the global average losses and potential extinction, the magni-
on services contribution to GDP. According to tude of which should exceed that of the re-
the World Bank, compared to the global aver- cent technological forces like e-commerce
age of 68%, Asia’s services contribution to and smart devices.
GDP is below 60% with China’s contribution
at a mere 48% and India at 53%. Mean-
while, developed economies score high with
services contributing close to 80% to GDP.
Interview
From a technology point of view, what Another area that you can’t ignore, whether
are the key hurdles to make AI a main- we’re talking about economic benefit or qual-
stream technology? ity of life, is healthcare. Over the past five or
Frankly, a big hurdle is just public awareness. so years, we’ve partnered with great health-
People don’t understand how mainstream AI care providers like Memorial Sloan Kettering
already is. AI, or cognitive as we call it at IBM, Cancer Center to teach Watson how to ana-
is here. lyze cancer and enable clinicians to provide
personalized medical care to each patient.
Our cognitive system, Watson, is already a key That technology is now being deployed in
part of everyday life for many of us – it’s Asia, bringing to the region some of the
expected to help more than one billion peo- world’s best medical care and expertise. At
ple this year. For example, Watson is helping the Manipal Hospital system in India, for
people in the United States do their taxes bet- example, Watson is helping doctors identify
ter through our partnership with H&R Block, personalized treatment options. Then there’s
and thanks to IBM’s partnership with Quest Gachon University Gil Medical Center in
Diagnostics, it can analyze the genomic Korea, where Watson is helping oncologists
makeup of a tumor and identify treatment keep pace with the massive volume of new
options or a clinical trial for that patient. cancer research that is published each year.
The challenge, now, is advancing this prog- Technology has already disrupted many
ress – drawing more developers to use cogni- business models in Asian countries. Will
tive engines like Watson to build their own AI take this disruption to the next level?
apps and working with more businesses and We call systems like Watson “cognitive com-
industries so they can incorporate cognitive puting” because unlike “artificial intelli-
solutions into their work. Cognitive needs to gence,” which connotes a machine replacing
be embedded into the workflow of an enter- human intelligence, our vision is that these
prise. systems will work in partnership with people.
When we think of AI, we consider it “aug-
What applications of AI provide the mented intelligence” or man plus machine.
biggest economic benefit to Asia as a
region? We expect cognitive systems like Watson will
In Asia, I think there’s a big opportunity for AI make professionals of all stripes more effi-
to improve the services sector. Asia is home cient, whether we’re talking about a tax pre-
to many of the most densely populated places parer or an oncologist or a CMO of a Fortune
on the planet. So, a cognitive system that 500 company. Watson will ingest an extraor-
could improve the logistics of life, like a sys- dinary amount of “unstructured data” in
tem that anticipates when smog and air qual- those fields – data like social media posts,
ity will be particularly bad, will be really valu- medical journals, or newspaper articles – and
able. develop intelligence based on this data for
professionals to use to make better decisions. What do you think is the impact of AI on
In this way, we believe that cognitive systems jobs particularly for a region like Asia,
are enhancing the way our world works. which has more people-driven business
models?
How do you see AI’s adoption in Asia? AI systems are already changing the way work
I think Asia is extraordinarily well-prepared to gets done. But history suggests that new
lead in the Cognitive Era, and here’s why: the technologies like AI result in higher productiv-
pioneers of this era, the people that are actu- ity, higher earnings, and overall job growth.
ally bringing cognitive solutions to different
industries, are the app developers and the In particular, we believe that new companies,
data engineers and scientists, and Asia has new jobs, and entirely new markets will be
these skills in spades. India is now home to built on the shoulders of this technology. And
one of the largest populations of developers as I mentioned before, I think Asia is particu-
in the world. larly well-suited to take advantage of this new
economy and these “new-collar jobs.” In
Just as important, many countries in the addition to the data scientists and app devel-
region, like Singapore, Australia, and Korea, opers that will build these cognitive solutions,
are investing in initiatives to support their we expect there will be a huge demand for
developer communities and the broader people with skills like “data labeling,” which
cognitive economy. In fact, last year, the are people who “teach” AI systems by feed-
Korean government announced it would ing them relevant information.
spend KRW 1 trillion (about USD 884 million)
by 2020 to boost the artificial intelligence
industry, which is part of the reason why
Watson is now “learning” Korean.
800
700
600
500
400
300
200
100
0
Financial services Healthcare Manufacturing Retail Transportation
Low-end High-end
Note: The numbers reflect range of high/low growth estimates off of current World Bank industry size.
Source: World bank, UBS estimates as of January 2017
Insurance is another area within financial ser- AI can help improve Asia’s healthcare coverage
vices where AI can have a long-lasting impact. Physicians per 1,000 people
According to OECD data, life insurance penetra-
tion (as measured by direct gross premiums as a 4.0
percentage of GDP) in Asia is significantly below 3.5
the developed market average (low single digits 3.0
vs. mid-to-high single digits). AI, through deep 2.5
learning, can elevate the region’s insurance in-
2.0
dustry through better products and pricing, un-
1.5
derwriting, target marketing and sales, claims
management, and overall data mining. 1.0
0.5
The risk, which other industry’s face but more so 0
with banks, is that AI will level the playing field India China US Switzerland European
Union
for emerging players from other industries, like
Source: World Bank, UBS, as of 2011
technology, to compete against the incumbent
financial giants. It is therefore imperative that
existing banking and insurance firms take AI seri-
ously and invest to maintain their differentiation.
180,000
Asia’s healthcare industry is a promising field for
160,000
AI, as AI-based applications can positively impact
140,000
the health of millions of Asians with better
120,000
health outcomes. The key applications where AI
100,000
holds promise in the region include decision-
80,000
making support during clinical trials, robots to
60,000
assist in surgery and patient monitoring, and
40,000
managing healthcare data.
20,000
pansion of industrial robot installations, we tems. From a disruption point of view, medi-
believe the advancement of AI offers significant um-level jobs are mostly at risk, as machines
opportunities to Asia’s manufacturing industry, may replace some of the more routine and pre-
which is historically the leading driver of growth dictable tasks. That being said, AI will still power
in the region. job creation in other retail areas like e-com-
merce, group buying and other creative areas.
AI will make Asian factories smarter, in which
data from design and product teams to supply
chains, including demand forecasting and quali-
ty control, are integrated into an intelligent ma- Transportation
chine. Thanks to the rising adoption of the Inter-
net of Things, which utilizes sensor technology, We believe transportation will be one of the
Asian factories should become more competitive industries most affected by AI. While AI has
in the future. According to data from IFR World already made inroads into the industry in the
Robotics, Asia already accounts for around 50% form of real-time monitoring, traffic and route
of newly installed robots. China is currently in prediction, and ride-sharing in countries like
the lead with a national 10-year plan, entitled China and India, self-driving, or autonomous
“Made in China 2025.” We expect Asia’s annual vehicles, is the next big trend for the industry.
share of robot installations worldwide to reach According to the World Bank, most of the vast
60% by the end of the decade and around 75% growth in global car ownership over the next
by 2025–2030. decade will take place in Asia, with an increase
of 220 million cars in China alone between 2010
and 2025. With human error responsible for
almost 90% of road accidents in the region,
Retail it is thus imperative to explore autonomous driv-
ing as an alternative for both policymakers and
Within Asia, we believe AI offers one of the industry participants.
broadest opportunities to the retail industry,
which is highly fragmented and unorganized; Despite the early days and a bit of skepticism
for example, the combined market share for the from consumers, the progress within Asia is on
top five retail companies in the US is 50% versus the right track. Japanese and Korean automakers
a single-digit share in Asian markets such as are investing significant sums in driver-less tech-
China and India. AI is already creating waves nology and showcasing prototypes at various in-
through the rising trend of e-commerce, where dustry events. China plans to include automated
penetration in most Asian countries is reaching or assisted driving in every car by 2026 to 2030.
mid-to-high single digits. China is the torchbear- We believe that in the next few decades, Asia’s
er of Asia’s e-commerce industry with mid-teen transportation industry will function and appear
penetration rates. AI’s applications in e-com- completely different to today because of the mass
merce are manifold, including product discovery adoption of autonomous driving technology.
tools like visual and voice-based search, targeted
marketing for customers, product recommenda- In addition to reducing the chance of human
tions based on user behavior, the use of robots error, self-driving cars have the potential to free
in sorting orders, delivering goods through up time and provide opportunities for infotain-
AI-powered drones, and virtual assistants or ment industries. Moreover, self-driving vehicles
chatbots for customer care. and peer-to-peer transportation may eliminate
the need to own vehicles. The industry may shift
We believe significant progress will be made in entirely towards a service-driven model, which
Asia’s e-commerce industry because of AI, driv- provides more revenue visibility. In the longer
ing further penetration. AI also provides oppor- term, the opportunity will be even bigger when
tunities in the traditional retail industry by im- self-driving extends to public transportation – at
proving organization in areas like supply chain least in the 10 largest cities in the region – and
management, marketing and advertising, so- and trucking, solving some supply chain chal-
phisticated assistants and self-check-out sys- lenges.
Interview
Adam Coates
Director, Baidu Silicon Valley AI Lab
Baidu
How do you view AI-related talent in What are the unique advantages that
Asia? Do you think the talent gap is Asia/China as a region possess that can
narrowing between Asia and the USA? help it become a leader in AI?
I think Silicon Valley currently is still the China’s internet economy is very dynamic and
hotspot to rapidly develop and adopt new AI nimble. China adopted technologies from the
technologies. There is a huge battle for talent West very quickly, particularly the internet
there as the demand is really high. In Silicon and mobile technology. This is not the case in
Valley, people can pick up high-level deep the US, where the technology took some time
learning and machine learning skills very to trickle down. I think China has the cultural
quickly from different jobs, and they’re all in advantage of adopting AI into their business
the same physical area. China’s internet eco- models to gain competitive advantages over
system is very dynamic. It’s less clear to me, its peers.
however, that someone who has good com-
puter programming skills in a particular locale Within Asia, Baidu is definitely a leader in AI.
in China would be able to rapidly pick up You can find that a lot of companies that are
deep learning and contribute to the supply investing in AI that are following a portfolio
of machine learning talent in a way that all approach. One thing in particular that I appre-
of the startups in China can harness to launch ciate about Baidu is that it managed to pick
new products. I think it is happening there, out Deep Learning at a very early stage. Our
but I sense there is no one place or a single efforts are primarily to convert these technol-
pressure point, like Silicon Valley, where ogies into products.
everyone is picking up these skills at a very
rapid pace. When it comes to advantages in talent and
big data, I think that the high-value things
Fortunately, in the US, the top universities that matter in Asia are much clearer to people
have developed AI programs and there are working at homegrown businesses. For exam-
many graduates with high-level machine ple, everybody in China uses QR codes but
and deep learning skills. I think big Chinese nobody in the US uses them. So, if you want
companies that supply AI technology can to roll out a product in China, it is very
have a hard time keeping up with demand, important to have talent who know the local
which would be extremely difficult for a small environment well. A locally grown product
startup to build, partly because the amount vision has quite an advantage because of the
of talent available to build these startups is differences in language, culture, habits and
quite small. even the government. China is a mobile-first
society, and Chinese users engage with their
mobile devices more frequently than Amer- mobile phones. We did a user study
ican users. In China today, when you are recently and we found that when people
connecting to the internet it tends to be on type in Mandarin, they tend to do so more
a mobile device. Hence, the e-commerce quickly and accurately with this new
websites in China are built in a way that is add-on. So we feel that speech recogni-
more attuned to mobile devices, and these tion will be really transformative for China
devices even have larger screens to cater to and Asia.
user demands and habits. You can’t just
copy a product from the USA and I think speech recognition is critical, espe-
re-launch it in China; it won’t work. cially because of the way it can be inte-
grated with different devices and accepted
Do you think AI will take disruption of by users. Speech agents like Duer ignite
old economy sectors to the next level? something that goes beyond mobile plat-
I think it’s possible depending on the type forms and can make things like IoT work.
of business. We foresee that AI is going Unlike wired devices, you can just talk to
to have a vast impact on many different these devices and from there they have the
industries. In all likelihood, AI technology ability to function independently. I think
has the potential to improve an insurance there is a chance that the speech technol-
company or a bank or a car company, ogy or personal systems that Baidu is build-
but the hard part is that the supply of ing could create a technological ecosystem,
able people is just not enough to deal similar to mobile.
with the number of different companies
and apps out there. So, when it comes to In terms of changes to society, I think
the question of which industries are vul- autonomous driving is very interesting.
nerable to disruption, I don’t see an inher- China’s government exerts a lot of control
ently disruptive aspect in AI technology. I over regulation. I think there is a chance to
think if big companies are able to inte- launch apps for self-driving vehicles in
grate AI technology correctly to optimize China that is different from those of the
and cater to what their customers want, US. But I am not yet sure about how this
then they will be in a powerful position. will play out and feel that the technology
AI is distinct in that it can make the value still has some ways to go. Baidu’s technical
proposition of a company even stronger capabilities are quite extraordinary, but
and take it to the next level, which is dif- there are still towering engineering chal-
ferent from disruption. However, if a com- lenges that need to be dealt with.
pany fails to take advantage of AI, then
there is the possibility that they may be Another interesting one is Baidu’s search
left behind. engine that uses AI technology to profile
users in details that are much deeper, inti-
Are there any aspects or areas of mate and fine-grained. It would not only
society in China or Asia that AI can provide simple recommendations to buy
have a major impact on? other products that you may like online,
People in Asia are adopting mobile tech- but understand the user so well that you
nology at a much faster rate than they did can push other things like interests, music,
for desktops and laptops. But for users in entertainment, information, etc. on the
China, typing in Mandarin on a small user’s Baidu search home page. Baidu is
mobile device is a complex task. One of already creating a technology called
Baidu’s most popular software products is User-Portrait to create high-precision mod-
a speech-enabled keyboard add-on for els from dozens of Baidu products and
partners. If it evolves, Baidu will deliver Many startups that are building personal
much more personalized experiences, assistants and robots based on deep learn-
which will be new and striking for users. ing may soon realize that it is not a prod-
uct by itself. In the VC world, the amount
Where are we in the AI cycle? Will the of funding available to do research on AI is
industry get consolidated or will some much lower now than it was maybe three
standards emerge? years ago. I feel that demand to see results
The core part of deep learning is well from AI research will go up in the upcom-
understood now and heavily adopted by ing years.
big internet companies. I think we are
either at the peak or post-peak in terms of
research, and now it needs to be making a
business impact. I think there is enough
tantalizing research to continue to capture
the media hype. Work is still going on with
supervised learning, reinforcement learn-
ing, and so on. The challenge now is to Adam Coates is director of the Silicon Valley AI
find someone who needs to find a killer Lab at Baidu Research. He received his PhD in 2012
application, and that is mostly a traditional from Stanford University and subsequently was a
business problem. On supervised learning, postdoctoral researcher at Stanford. During his
there is a lot of potential that has not been graduate career, Coates co-developed the most
tapped, and we are limited in a way sophisticated autonomous helicopter in the world,
capable of advanced aerobatic maneuvers. His
because we need labelled datasets. For
thesis work investigated issues in the development
speech recognition, we need transcribed of deep learning methods, particularly the success
speech, so there is an inherent limit to how of large neural networks trained with large data
much data we can use. If more progress is sets. He led development of deep learning using
made in supervised learning, I think it distributed GPU clusters at Stanford. At Baidu his
could lead to many more new applications. team builds HPC systems for large-scale deep
So in computer vision and speech, there is learning and has applied them especially to speech
recognition. The team’s speech system, Deep
a clear vision on what can happen in the
Speech, uses networks with billions of connections
next couple of years. Natural language and tens of thousands of hours of audio to achieve
technology is getting better gradually, but human level performance in many cases, and has
barring a research breakthrough, I don’t been deployed as part of Baidu products in English
see it getting completely cracked. and Mandarin.
AI’s economic
value for Asia
Artificial intelligence is the new electricity of
the future. – Andrew Ng, Coursea
In our previous report, “Automation and tasks, we believe medium-skilled jobs are most
robotics – Spotlight on artificial intelligence” at risk. This is based on our view that both low-
published in August 2016, we estimated that and high-level jobs require some level of person-
50–75 million jobs globally will potentially be alization, creativity or craftsmanship. Against this
affected due to the mainstream adoption of AI, backdrop, as seen in the table on page 21, Asia
representing only 2% of the total workforce has a higher percentage of workers at the medi-
worldwide. Using a similar framework, we be- um level (69%) compared to the global average
lieve 30–50 million jobs in Asia are at a risk due (66%) but much less than developed markets
to AI in the medium to long term. While we be- like the US and the EU.
lieve it will take a long time for AI to fully re-
place these jobs, as AI technology needs to be At present, manufacturing-driven economies
further developed and due to the fact that cor- like China are most at risk, while services-driven
porates and governments are still warming up to economies, like Hong Kong, Singapore and
AI, we have extended the framework to assess India, should be less affected by widespread AI
the types of jobs that are most at risk in Asia. adoption. The good news is that the region’s
Real Estate
Healthcare
Others
Retail
Manufacturing
Construction
Education
Public admin
ongoing shift from old economy to new econo- Like any technological evolution, overall employ-
my should minimize the employment fallout. ee productivity should significantly increase in
The rising trend of the sharing economy in coun- Asia with the rise of AI, providing enough op-
tries like China is a positive development. The portunities for employees to upgrade their skills
sharing economy, or collaborative consumption, and focus on other creative aspects. With the
where technology is used to share resources, emergence of other disruptive business models
such as ride-sharing, space-sharing and knowl- like apps or sharing economies highly likely in a
edge-sharing, is growing by leaps and bounds in post-AI era, we see increased scope for jobs that
markets like China; the industry’s revenues in require a high level of personalization, creativity
China are already in the hundreds of billions of or craftsmanship – tasks that will still need a per-
US dollars, creating tens of thousands of new son. Overall, we think the fears of an employ-
jobs. AI will be the key driver of the sharing ment crisis in Asia are unfounded, since AI will
economy in the future, helping to deliver rele- eventually drive innovation and provide employ-
vant results free from human interference, pre- ment opportunities in other areas in the region.
vent data misuse, anticipate customer demand,
and generate new jobs. Hence, we believe the
net job losses from AI will be significantly lower
than the headline figure, because we believe AI
will also help to create new types of jobs in oth-
er industries.
iStock
The next 10–15 years will be the most exciting a balance to address AI’s benefits and risks to
period in the field of AI as we enter the AGI employment. Opposing AI’s implementation in
stage and the use cases for AI expand at an the region because of concerns over job losses
exponential pace. Unlike the previous industrial alone would backfire, as we believe it would
revolutions, Asia will likely be a key stakeholder severely hurt competitiveness and innovation.
this time around as AI becomes ubiquitous However, policymakers have a role to regulate
across the region. The implications for investors, and equally facilitate AI’s development and
society, corporates and governments will there- adoption. To address AI’s ethical dilemmas, poli-
fore be significant, in our view. Investors should cymakers can be proactive over the next few
avoid companies exposed to significant disrup- years to introduce a code of conduct. And to ad-
tion from AI and embrace companies with a dress employment concerns, Asian governments
strong technology focus in areas such as robotic can help increase the skill levels of its citizens
process automation (RPA) and machine and and also encourage AI-based innovation through
deep learning and R&D-intensive companies in partnerships with universities and start-ups. We
sectors that can become more efficient from AI believe regional cooperation on these efforts
(e.g. healthcare, hi-tech engineering and select could be more fruitful by not only developing
services firms). common standards but also building a potential
alternative to Silicon Valley.
While the ethical implications of AI still need to
be worked out, if AI is implemented in an objec- Hence, from a society point of view, we should
tive way to improve productivity, society stands start to focus more on occupations that require
to benefit immensely from the significant im- a high level of personalization, creativity and
provements to overall lifestyles and healthcare. craftsmanship, as these jobs will be less affected
It is still not too late for Asian corporates to hop by AI. Regulators also need to step up their ef-
on the AI bandwagon, as it is an irreversible forts by seeking cooperation regionally and also
trend that could potentially disrupt business by striking a balance between promoting the
models across markets. Asian corporates, partic- adoption of AI and managing the downside
ularly those in old economy sectors, should have risks that could arise from the ethical dilemmas
an open mind and try to embrace AI rather than of AI or from other societal aspects like job loss-
fight against it; we believe the benefits out- es or changes. With these recommendations,
weigh the risks. Failure to adapt to the AI-era we believe the transition towards an AI-based
may make some of these business models ex- society would be smoother and a highly produc-
tinct. Finally, policymakers in Asia need to find tive one.