Top AI Trends To
Watch In 2018
China is racing ahead in AI. Deep learning is
getting a make over. AI is coming to Cannabis
tech. We look at 13 artificial intelligence
trends reshaping industries and economies.
Artificial intelligence is changing the fundamental structure of
every industry in areas ranging from agriculture to cybersecurity
to commerce to healthcare, and more. We're also interacting
with technology in new ways, from giving voice commands to
washer-dryers to playing advanced gesture-controlled video
games.
Governments are competing to establish superior AI research,
seeing AI as a lever for greater economic influence and power.
We are also in the early stages of drastic shifts in the labor
market. The hype around machine learning may start to fade
but that's because machine learning has already penetrated
virtually every major piece of software, from calendar apps to
search engines to sales management software.
AI is heating up across every industry
Equity deals Q4'12Q4'17
AI can now out-bluff world poker champions. A humanoid robot
can do a perfect back flip and land on its feet. But despite these
advances, AI algorithms are far from perfect in basic tasks that
are easy for humans, such as understanding a scene in an image
or recognizing a conversation's context.
Meanwhile, the promise of general AI or artificial intelligence
that can quickly learn new tasks without supervision remains
uncertain. Although a handful of companies like Vicarious
Systems and Kindred have raised money to develop general AI,
there is little evidence of specifics or real traction.
AI applications today focus on very narrow tasks. But together
these narrow AI-driven tasks are reshaping businesses,
markets, and industries.
We examined our database for the metrics and trajectories of
thousands of AI companies globally to bring you 13 artificial
intelligence trends our analysts will be watching in 2018. These
range from China's ambitious plans to the emergence of capsule
networks to 6-figure salaries for AI specialists.
At CB Insights, we believe
the most complex strategic
business questions are best
answered with facts.
We are a machine intelligence company
that synthesizes, analyzes and visualizes
millions of documents to give our clients
fast, fact-based insights.
From Cisco to Citi to Castrol to IBM and hundreds of others,
we give companies the power to make better decisions, take
control of their own futureand capitalize on change.
The CB Insights platform
has the underlying data
included in this report
WHERE IS ALL THIS D ATA FRO M ?
CLICK HERE TO SIGN UP FOR FREE
Beti Cung,
CORPORATE STRATEGY, MICROSOFT
" We use CB Insights to
find emerging trends and
interesting companies
that might signal a shift in
technology or require us
to reallocate resources."
TRUSTED BY THE WO RLD ' S LEADING CO MPANIES
1New blue collar job robot babysitters
2AI for X is everywhere
3China vies with US for global AI leadership
4The future of defense turns on AI
5Cmo ests, Alexa?
6White-collar automation accelerates
7AI moves to the edge
8The emergence of 'capsule networks'
96-figure salaries in the AI talent wars
10The machine learning hype will die
11Amazon, Google, Microsoft dominate enterprise AI
12AI diagnostics gets the nod from regulators
13DIY AI is here
Table of
contents
1
New blue collar job
robot babysitters
Manufacturing jobs are notoriously vulnerable to being out-
sourced to developing countries where labor costs are cheaper.
But dropping industrial robot costs can sometimes also bring
manufacturing bases closer to site of demand.
Recently, Chinese T-shirt manufacturer Tianyuan Garments
Company signed a Memorandum of Understanding (MoU) with
the Arkansas government to employ 400 workers at $14/hr at its
new garment factory in Arkansas. Operations were scheduled to
begin by the end of 2017.
Tianyuan's factory in Little Rock, Arkansas, will use sewing robots
developed by Georgia-based startup SoftWear Automation to
manufacture apparel for Adidas.
It appears much of the heavy lifting will be done by the robots
with human workers taking over high-end jobs including robot
maintenance and operation.
This means that the number and nature of manufacturing jobs will
never equal 2008 numbers.
1
Counterintuitively,
industrial robots and
manufacturing jobs are
both on the rise in the
United States.
2
The Bureau of Labor Statistics includes different types of jobs
under its definition of the manufacturing industry. Its outlook
for quality control inspectors and assembler and fabricators, for
instance, is negative due to the impact of automation.
For context, a 2012 DARPA contract awarded to previously-men-
tioned SoftWear Automation clearly states "complete production
facilities that produce garments with zero direct labor is the
ultimate goal."
But ever-changing consumer preferences and robots that are
unable to adapt to drastic process changes stand in the way of
complete automation.
This is reflected even in Amazon's highly automated warehouses.
Amazon's collaborative warehouse robots perform much of
the heavy lifting, while workers focusing on delicate tasks like
"picking" items off shelves and slotting them into separate orders.
Robots are still less-than-perfect at gripping, picking, and
handling items in unstructured environments. Amazon already
uses over 100,000 robots in various warehouses, but at the same
time is creating thousands of new jobs for humans in its new
fulfillment centers.
3
AI for X is everywhere
Artificial intelligence is everywhere. Or more exactly, machine
learning is everywhere. Machine learning refers to the training of
algorithms on large data sets so that they learn how to identify
and generate desired patterns. Over time, the algorithms pro-
vided with the correct parameters by their human creators get
better at their tasks.
This tech can basically be used to do anything, provided there is
data to train the software on and a desired outcome in mind.
So: UK's IntelligentX wants to introduce the world's first AI-
brewed beer.
DeepFish in Russia is using neural networks to identify, well, fish.
It merges radar technology with AI to differentiate between fish
and noise in radar images.
Sweden's Hoofstep raised VC money to bring deep learning-based
behavioral analysis to horses.
Are you vegan, gluten-free, or allergic to soy? New York's Prose
wants to use AI for made-to-order hair products. It raised $7.57M
from well-known VCs including Forerunner Ventures, Lerer
Hippeau Ventures, and Maveron.
2
The 'AI for X' trend is
unstoppable. From
brewing beer to tending
to cannabis buds,
machine learning is
doing it all.
4
AI is also coming to cannabis tech. DeepGreen uses computer
vision to identify the gender and health profile of cannabis
plants. Weedguide raised $1.7M to use AI for personalized
weed recommendations.
From hobbies to revenue-generating ideas to simply taking
things too far, we expect to see more out-of-the-box "AI for X" in
2018. More broadly, the prevalence of this trend and increasingly
absurd-seeming examples reveal that machine learning is not
an exotic technology. Rather, it is one of the building blocks for
modern software and applications.
5
China vies with US for
global AI leadership
Despite a mere 9% share of deals going
to AI startups globally, China's AI startup
scene took nearly 50% of dollars going to
AI startups globally in 2017, surpassing
the United States for the first time for
share of dollars.
China is aggressively executing a thoroughly-designed vision for
AI. In some areas of AI, China is clearly beating the US.
The Chinese government is promoting a futuristic artificial
intelligence plan. It encompasses everything from smart
agriculture and intelligent logistics to military applications and
new employment opportunities growing out of AI.
Part of the resources are going to innovative China-based
startups developing AI in industries ranging from healthcare
to media.
3
6
In fact, China accounts for a mere 9% share of deals going to AI
startups globally. But China's AI startups took 48% of all dollars
going to AI startups globally in 2017, surpassing the United States
for the first time for share of dollars.
To put this proportion in perspective, in 2016, China accounted for
only 11.3% of global funding.
China dominates global AI funding
US vs. China total equity funding to startups in 2017
7
The United States still dominates globally in terms of the number
of AI startups and total equity deals. But it is gradually losing its
global deal share.
The United States is losing its global AI
deal share
Equity deal share, 20132017
Chinese companies' R&D efforts are reflected in their
patent activity.
Chinese companies seem to be overtaking their US counterparts
in AI patent applications. Based on basic keywords searches
of title and abstract, AI-related patent publications in China are
surging far ahead of patents being published in these spaces by
the US Patent and Trademark Office.
In deep learning, for example, patents published in China are 6x
what they are in the US. (Note: The patent filing process involves a
significant time-lag before the publishing of patent applications.)
AI-related patent pub-
lications explode in China
Based on keyword searches of title
and abstract, 20132017
Source: epo.org
8
Two prominent technologies fueling China's AI growth are facial
recognition and AI chips. The former advances the government's
ambitious country-wide surveillance plans, while the latter is a
direct challenge to US-made chips.
Three key players here are China-based unicorns Megvii
(dba Face++) and SenseTime, and startup CloudWalk (the
latter a recipient of a $301M grant from the Guangzhou
Municipal Government).
In 2017, around 55 cities in China were part of a plan called
Xue Liang or "sharp eyes." Footage from surveillance cameras
in public and private properties will be processed centrally to
monitor people and events. Media reports suggest that this may
eventually power China's Social Credit System: a metric to gauge
the "trustworthiness" of its citizens.
China invests heavily in facial recognition
technology
All deals, including grants, 2013 - 2017
Startup Megvii already has access to 1.3 billion face data
records on Chinese citizens and is backed by Chinese insurance
companies (Sunshine Insurance Group), government entities
(Russia-China investment group), and corporate giants (Foxconn,
Ant Financial).
9
Two investors in the company, Alibaba Group
(through Ant Financial) and Foxconn, partnered with
Hangzhou city in China in 2016 for the "City Brain"
project, using AI to analyze data from surveillance
cameras and social feeds.
Ant Financial separately uses facial recognition for
payments at Alibaba-owned retail stores.
The United States and China are also competing on
dominance in AI chip technology.
In July 2017, the Chinese government said it
planned to reach parity with the United States on
artificial intelligence by 2020 and become the world
leader by 2030. One government-backed project is
to create a chip that has 20 times the performance
and energy efficiency of NVIDIA's GPUs. Chinese
company Cambricon is pledging to make one billion
processing units in the next three years and is
developing chips specifically for deep learning.
Key Chinese tech giants like Baidu and JD are
investing in AI companies abroad, including in the US.
Recently, Baidu and JD.com backed ZestFinance,
and Tencent backed NY-based ObEN. Some start-
ups like WuXi NextCODE and Pony.ai are operating
in both countries, further blurring competitive lines.
Despite scrutiny of Chinese companies seeking
partnerships or investments in the US, there are
more Chinese investments in AI startups in the US,
than vice-versa.
Cross-border AI investments are on the rise
Equity deals, 2013 - 2017
10
4 The future of defense turns
on AI
The battlefield is moving to data centers.
As far back as in 2014, Amazon built a custom cloud computing
service for the CIA, meeting stringent compliance and regulatory
requirements for sensitive data.
In Q4'17, AWS opened these tools to other government customers
outside the intelligence community.
Amazon has also acquired two AI-cybersecurity companies
Harvest.ai and Sqrrl for securing sensitive data in the cloud.
Whether it's Amazon or a host of new startups catering to
government clients, AI promises to be the backbone of new
government-sponsored cybersecurity efforts.
In the Cold War, governments talked about their "missile gaps,"
or their disadvantages relative to rivals in terms of nuclear
warheads. Now, governments increasingly rate their gaps in terms
of cyber capabilities. As a result, the worlds of cybersecurity and
traditional defense are merging.
The AI-cybersecurity
market is getting
increasingly crowded.
Some of the startups
boast a roster of
government clients
hoping to stay one step
ahead of hackers.
11
Data breaches bring the risks into sharp relief: from the
Equifax leak of millions of citizens' social security numbers to
WannaCry ransomware to Russian meddling in elections in the
US and elsewhere.
US government organizations received one of the lowest scores
for cybersecurity in a 2017 analysis by SecurityScorecard (a
New York-based company backed by Intel Capital and Moody's
Corporation, among others). The analysis included "552 local,
state, and federal government organizations, each with more than
100 public-facing IP addresses."
Cybersecurity poses a real opportunity for the deployment
of AI algorithms, since attacks are constantly-evolving and
defenses frequently face previously-unknown types of malware.
Presumably, AI would have an edge here given its ability to
operate at scale and sift through millions of incidents to identify
anomalies, risks, and signals of future threats.
The market is now flooded with new cybersecurity companies
trying to leverage machine learning to some degree.
A total of 134 startups have raised $3.65B in equity funding in the
last 5 years. About 34 of them raised equity for the first time last
year to compete in a market still dominated by larger companies
like Cybereason, CrowdStrike, Cylance, and Tanium each with
$900M+ valuations.
12
New AI-cybersecurity startups enter market
dominated by unicorns
1st equity deals, 2013 - 2017
Even a traditional consulting firm like Accenture has been scaling
its technology in AI-cybersecurity to better serve federal govern-
ment clientele. A notable deal here is startup Endgame, which
has clients like the US Air Force. Endgame sold its government
services division to Accenture.
The intelligence community's venture arm In-Q-Tel funded
Anomali, Interset, and Cylance in 2016. UK's Darktrace claims its
system has over 3,000 worldwide deployments, including use by
government agencies. Colorado-based Logrhythm works with the
US Air force, NASA, and defense contractor Raytheon.
Other top defense contractors are investing here as well.
Lockheed Martin was an early investor Cybereason (currently
valued at over $900M). In 2017, Boeing backed Texas-based
cybersecurity startup SparkCognition through its venture
arm HorizonX.
13
Cmo ests, Alexa?
Alexa has unleashed a voice revolution.
Voice-enabled computing was all the buzz at the Consumer
Electronics Show in 2018. Hardly any IoT device was without
integration into the Amazon Echo or Google Home.
Samsung is now working on its own voice assistant, Bixby. It
wants all of its products to be internet-connected and have
intelligence from Bixby by 2020. LG made all of its appliances
in 2017 WiFi-enabled. Over 80 LG products now integrate with
Google Home.
Although Amazon had an early lead in voice computing, it has
fallen behind in terms of language support.
Amazon announced last quarter that it'll start shipping its Alexa-
powered speakers to around 80 countries. But on the downside,
it expects users around the globe to either interact in English,
German, or Japanese.
Google Home is available in English, German, French, and
Japanese. Apple's HomePod currently is available in English only,
despite its plans of launching soon in Germany and France.
5
Amazon Echo and
Google Home domi-
nate the smart home
speaker market. But
non-English speaking
markets are currently
underserved by the big
tech companies.
14
Google has a significant advantage over Amazon here. Google
Assistant for Android phones in available in English, French,
German, Italian, Japanese, Korean, Spanish, and Portuguese.
Its speech recognition capabilities used for speech-to-text
conversions and voice searches extends to 119 languages.
The Spanish smart home market is currently underserved by big
tech companies, despite being one of the most widely spoken
languages in the world after Mandarin.
In China, Alibaba reported that its Chinese-speaking Tmall Genie
(its version of Amazon Echo) has sold over 1 million units since
officially launching in July 2017.
In 2018, we will continue to see voice assistants battle it out for
dominance in non-English speaking markets.
15
White-collar automation
accelerates
A growing wave of AI-infused Expert Automation & Augmentation
Software (EAAS, pronounced /z/) platforms will usher in a new
era of AI-assisted or AI-enhanced productivity.
This AI-enhanced productivity is threatening jobs at the more
clerical end of the white-collar spectrum.
The EAAS market map below highlights some of the startups
building expert automation & augmentation software across a
number of professions and industries ranging from lawyers to
journalists to wealth managers to traders to consulting, and more.
For instance, artificial intelligence has huge potential to reduce
time and improve efficiency in legal work. On the litigation side,
natural language processing (text analytics) can summarize
thousands of pages of legal documents within minutes. This
is a task that might take a human counterpart several days to
complete. Meanwhile, it also reduces the probability of error.
As AI platforms become more efficient and commercialized, this
will impact the fee structure of external law firms that charge by
the hour.
6
White collar workers
including lawyers,
consultants, financial
advisors, journalists,
traders, and more
will face the effects
of AI as much as
blue-collar laborers.
16
Programmers are not immune. Early-stage startups are focused
on AI-based software testing, debugging, and basic frontend
development. One of the top rounds last year went to UK-based
DiffBlue, which is developing AI to automate traditional coding
tasks like bug fixing, custom code development, and translating
code from one programming language to another.
Healthcare and education are considered to be some of the
industries least at risk of automation due to the dynamic nature
of tasks. These fields also require a high level of emotional intelli-
gence. But in education, for instance, AI startups are beginning to
provide ancillary services like paper grading, language coaching,
and feedback on writing.
17
AI moves to the edge
AI is getting decentralized.
Intelligence on a device like a smartphone or a car or even
a wearable device gives it the ability to process information
locally and respond quickly to situations, instead of communicat-
ing with a central cloud or server.
For instance, an autonomous vehicle has to respond in real-time
to what's happening on the road. Decisions are time-sensitive and
latency could prove fatal.
Another case for edge AI would be training your personal AI
assistant locally on your device to recognize your unique accent
or identify faces.
2017 was marked by big tech companies making huge leaps in
edge computing.
7
2017 saw a major push
for bringing AI to the
edge, i.e. to smaller
devices and sensors
operating closer to the
periphery of computing
networks. In other
words, AI may live
within your earphones
rather than being
housed in the cloud or
in your smartphone.
18
AI on the edge reduces response times
A few examples of emerging edge AI applications
Apple released its A11 chip with a "neural engine" for iPhone 8
and X. Apple claims it can perform machine learning tasks at up
to 600B operations per second. It powers new iPhone features
like FaceID, which scans a user's face with an invisible spray of
light, without uploading or storing any user data (or their face) in
the cloud.
As the dominant processor in many data centers, Intel has had to
play catch-up with massive acquisitions. Recently Intel released
an on-device vision processing chip called Myriad X (initially
developed by Movidius, which Intel acquired in 2016).
Myriad X promises to take on-device deep learning beyond
smartphones to devices like baby monitors and drones.
Google proposed a similar concept with its "federated learning"
approach, where some of the machine learning "training" can
happen on your device. It's testing out the feature in Gboard, the
Google keyboard.
AI on the edge reduces latency. But unlike the cloud, edge has
storage and processing constraints.
More hybrid models will likely emerge with intelligent edge
devices communicating with each other and a central server.
19
The emergence of 'capsule
networks'
Neural networks have different architectures. A popular one in
deep learning today is known as convolutional neural networks.
Now a new architecture, capsule networks, has emerged and
promises to outperform convolutional neural networks (CNNs)
on multiple fronts.
CNNs, despite their success, have shortcomings that may lead
to lack of performance or even security gaps. Researchers are
looking for ways to improve AI algorithms and overcome these
drawbacks.
The example below shows a very basic illustration. A CNN
would identify individual features and mistake the second
image to be a face.
Challenges of convolution neural networks
One of the pioneering researchers in deep learning, Google's
Geoffrey Hinton, published a research paper in 2017 that intro-
duces the concept of "capsule networks," also known as CapsNet.
Deep learning has
fueled the majority
of the AI applications
today. It may now get
a makeover thanks to
capsule networks.
8
Illustration source:
jhui.github.io
20
The paper is still in the review stage, and will need to be tested in
practical scenarios. But the promise it holds has generated a lot
of buzz in the media and tech community.
Without getting into technical details, capsule networks would
allow AIs to identify general patterns with less data and be less
vulnerable to incorrect results. For example, these networks
would more easily identify that when features on a face are
rearranged it is no longer a face. This is something convolutional
networks are not good at.
Promise of capsule networks
Another issue with CNNs is that they cannot handle different
variations of input data. For instance, you have to train the
algorithm with images of the same object from different angles
or viewpoints for a CNN to identify all variations. As a result,
it would require a large volume of training data to cover all
possible variations.
Capsule networks claim to outperform CNNs here. They would
require less training data, and would take relative positions and
orientation of an object into consideration without needing to be
trained exhaustively on variations.
Hinton's paper also claims that capsule networks have been
tested against some sophisticated adversarial attacks (tampering
with images to confuse the algorithms) and was found to
outperform convolutional neural networks.
Illustration source:
jhui.github.io
21
Hackers can introduce small variations to fool a CNN.
Researchers at Google and OpenAI have demonstrated this with
several examples.
One of the more popular examples is from a 2015 paper. As can
be seen below, a small change that is not readily noticeable to the
human eye means the image results in a neural network identify-
ing a panda as a gibbon, a type of ape, with high confidence.
22
6-figure salaries in the AI
talent wars
China is hiring AI experts.
Some of the salaries listed are nearly $567-624K for a senior
machine learning researcher at BMW China, and $315-410K for
ML experts at various other companies.
The jobs are listed on Liepin, a recruitment platform that is itself a
unicorn startup from China.
According to a recent Tencent report, the estimated number of
qualified researchers currently in the field is 300,000, including
students in relevant research areas. Meanwhile, companies likely
require a million or more AI specialists for their engineering needs.
In the US, a Glassdoor search for "artificial intelligence" shows
over 32,000 jobs currently listed, with several salary ranges well
into the 6 digits.
Big tech companies are scooping up the crme-de-la-crme of AI
with competitive salary packages.
The demand for AI
talent is far outpacing
the availability of
skilled researchers.
Top AI researchers
can now mint money
in the millions.
9
23
Deepmind Technologies (acquired by Google in 2014) reported
in its financial statement last year that "staff costs and other
related costs" accounted for 104.8M. A quick LinkedIn search
puts the staff number at 415. Assuming this as team size in 2016,
and discounting other expenses, this puts the average employee
salary for the team at 252,000 (around $350,000 per annum).
Adding to this, AI researchers at big tech corporations are also
leaving to start their own companies.
Andrew Ng left Baidu to launch a $175M AI fund. The chief
technology officer of stealth AI chip startup Groq worked at the
hardware engineering unit for the Tensor Processing Unit at
Google and later at Google X.
The CTO and co-founder of chip startup Horizon Robotics was a
previous Baidu employee who led the image recognition team at
the Baidu Deep Learning Institute.
With talent bleeding to startups, the talent war becomes even
more intense.
Source: Companies House, United Kingdom
24
The machine learning hype
will die
First came big data, then the cloud, then the machine learning frenzy.
We reached peak machine learning in 2017. Incubators alone
housed more than 300 AI startups, up 3x from the previous year.
Last year investors poured in over $15.2B in funding to AI startups
across industries. It was a 141% jump in funding from 2016.
Over 1,100 new AI companies have raised their first rounds of
equity funding since 2016. To put this in perspective, that's more
than half the historic number of AI startups that have ever raised
an equity round.
Machine learning will
soon become the new
normal. And the 1,100+
new AI startups that
emerged after 2016 will
need robust business
models to stay alive.
10
25
AI sees 141% funding jump in 2017
Equity deals, 2013 2017 (excluding hardware-focused
robotics startups)
But the hype will soon die.
The normalization of machine learning will make investors picky
about the AI companies they fund.
As Frank Chen of a16z put it, "in a few years, no investors
are going to be looking for AI startups." It will be "assumed"
that startups are using the necessary AI algorithms to power
their products.
We are already seeing this happen in many industries.
Machine learning is inseparable from IIoT. We need AI to make
sense of the vast amounts of data collected from machines and
sensors, and process them in real-time. Almost all cybersecurity
companies use machine leaning to some extent today. In addition
to this, big tech companies are offering a suite of machine
learning solutions to enterprises.
Top investors are carefully gauging startups using AI. For instance,
diagnostics startup Freenome was sent 5 unlabeled blood
samples for analysis with its AI-powered algorithms before a16z
decided to back the company.
26
Amazon, Google,
Microsoft dominate
enterprise AI
As companies struggled to integrate machine learning into
their products, startups emerged to provide ML-as-a-service.
Now, as big tech companies like Google, Amazon, Salesforce,
and Microsoft improve their enterprise AI offerings, they will
make smaller companies and funding to the space obsolete.
Google released Cloud AutoML. Customers can bring their
own data to train the algorithms to suit their specific needs.
Amazon began selling AI-as-a-Service with "Amazon AI"
under its AWS banner. Amazon AI's goal is to serve big and
Investors poured in
$1.8B into enterprise
AI startups in 5
years. Now Amazon,
Microsoft, and Google
may make smaller
companies obsolete.
11
27
small-time developers that want AI without the upfront costs or
hassle. It unveiled offerings that will work like an API and allow
any developer to access Lex (the NLP inside Alexa), Amazon Polly
(speech synthesis), and Amazon Rekognition (image analysis).
In Q4'17, Amazon expanded its services to include video
recognition, audio transcriptions, and sentiment analysis. AWS
has a massive footprint. Its fourth quarter revenue alone was $5B,
up 44% quarter-to-quarter.
Microsoft is competing neck and neck with Amazon. Salesforce
and other companies like Oracle are also not far behind.
28
AI diagnostics gets the
nod from regulators
US regulators are looking at approving AI for use in
clinical settings.
The promise of AI in diagnostics is early detection and
increased accuracy.
Machine learning algorithms can compare a medical image with
those of millions of other patients, picking up on nuances that a
human eye may miss. It can do in seconds what a human would
take hours to complete.
Consumer-focused AI monitoring tools like SkinVision which
uses computer vision to monitor suspicious skin lesions are
already in use. But a new wave of healthcare AI applications will
institutionalize ML capabilities in hospitals and clinics.
This month AstraZeneca announced a partnership with Alibaba
subsidiary Ali Health to develop applications including AI-assisted
screening and diagnostics in China.
This comes on the heels of GE and Nvidia's partnership to bring
deep learning capabilities to GE's medical imaging devices, and
It will soon
be routine for
machine learning
to play a role in
medical imaging
and diagnostics.
12
29
Google DeepMind's advances in using AI to diagnose
eye diseases.
Big names Google DeepMind, IBM, GE, and Alibaba make this a
tough market for smaller startups to compete in. But this hasn't
stopped new companies from venturing into the space.
Healthcare is the hottest area of AI startup investment as our
heatmap at the beginning of the report showed. Much of this
growth is fueled by medical imaging & diagnostics companies.
The increasingly crowded healthcare AI
space
1st equity deals, 2013 - 2017
One of the first FDA approvals was for startup Arterys. Its cloud
computing platform was approved for analyzing cardiac images,
reportedly based on a series of tests for accuracy and speed of
diagnosis. It is now applying for FDA approval for AI in oncology.
Another startup, MedyMatch, is using deep learning to detect
intracranial hemorrhage from CT scans. The FDA recently gave
it a "breakthrough device designation" to expedite the process of
bringing the product to market.
The biggest bone of contention in a high-stakes industry like
healthcare is who takes responsibility for misdiagnosis by AI
software. The current wave of applications are all geared to assist
radiologists and physicians, as opposed to serving as a final
verdict on diagnosis.
30
DIY AI is here
You don't need a PhD in computer science or mathematics to
build your own AI.
Between open source software libraries, hundreds of APIs and
SDKs, and easy assembly kits from Amazon and Google, the
barrier to entry is lower than ever before.
Google launched an "AI for all ages" project called AIY (artificial
intelligence yourself).
The first product was a do-it-yourself voice recognition kit for
Raspberry Pi. From making mid-80s intercoms with intelligent
assistants inside of them, to making voice assistants sound like
characters from Dr. Who, users are knocking themselves out
creating new AI-based inventions.
Google also launched a vision kit with neural network software
programs so users can make algorithms to identify dogs and cats,
or to match emotion to facial expressions.
Amazon has launched DeepLens, a $249 deep learning-based
video camera. Amazon is offering $7,500 for the winners of
its first DeepLens hackathon, which involves building machine
learning projects.
Make your voice
assistant sound
like Dalek from Dr.
Who. Or build your
own AI camera.
13