Morningstar Magazine - April/May 2018 - 48

Strategies

Artificial Intelligence Takes Off
Venture deals spike as field makes
major advancements.

2 Unsupervised Learning

In unsupervised learning, the algorithm trains with
unlabeled data and must determine the
underlying structure of the data set and how
to group it (for example, recognizing handwritten numeric digits by classifying them into
10 groups).
3 Reinforcement Learning

INVESTMENT RESEARCH: VENTURE CAPITAL

Cameron Stanfill

Rapid advancements and significant commercial
integration of applications over the past decade
have created excitement around the field of
artificial intelligence and machine learning.
The concept of artificial intelligence, or AI, has
been on scientists' minds since the 1950s, which
saw experimentations with building an artificial
brain and the Turing test 1. The most recent
breakthroughs include advancements in deep
learning-a subset of machine learning, or ML,
that is based on layered neural networks
inspired by the biology of the human brain.

insights from the data without being explicitly
programmed. We interact with AI/ML applications
every day when we talk to a voice assistant,
use facial recognition technology, and receive
movie and restaurant recommendations.
ML can be broken into the three subcategories
based on the method in which the algorithm
is trained:
1 Supervised Learning

Supervised learning is a method in which the ML
algorithm is trained on a set of labeled data
(for example, training an image recognition system
with labeled pictures to identify pictures of dogs).

Reinforcement learning starts with no training
data, which means the machine must
learn through experience and iterative trial and
error to perform its task while maximizing
long-term reward.
In the near term, we think there will continue
to be a proliferation of consumer-facing supervised
learning applications in AI/ML. These will
focus on the automation of tasks, almost to the
point of ubiquity. Over the next year, these
companies will produce the most commercially
viable AI/ML products and will represent
many acquisitions in the vertical. However, we
believe the largest AI/ML market prospects
will derive from expansion into a host of enterprise
applications and new industries.

EXHIBIT 1

Deep learning techniques are behind nearly all
the current frontier research and successful
applications of AI/ML, such as image recognition,
search engines, drug discovery, and deep
reinforcement learning. This progress was made
possible by the new research approaches to
ML problems, combined with huge advances in
computing power and the rampant
increase in data digitization and availability.

Rapid Advancement Venture capital deal counts and values in artificial intelligence
have soared over the past decade.
643

Deal Count
700

6.01

600

Deal Flow
$7.0
Bil
6.0

500

5.0
4.53

400

4.0
Background
AI is the area of computer science that focuses

on the creation of an intelligent machine that
can perceive its environment and make decisions
to maximize the chances of reaching its goal.
ML is a subfield of AI and data analysis that
is working to give computers the ability to learn
iteratively, improve predictive models, and find

3.19

3.0

2.57

200

2.0
1.52

1.0
0.0

0.07

0.20

0.24

0.30

0.44

2006 2007 2008 2009 2010

0.85

0.81

2011

2012 2013 2014 2015 2016 2017

Source: PitchBook.

1 Pioneering English computer scientist Alan Turing, who was instrumental in developing a system to break the German Enigma code during World War II, created a test to determine whether a
machine's ability to exhibit intelligent behavior was equivalent to that of a human. Turing, A.M. 1950. "Computing Machinery and Intelligence." Mind, Vol. LIX, Issue 236, pp. 433-460.

Morningstar April/May 2018

300

100
0.0

Table of Contents for the Digital Edition of Morningstar Magazine - April/May 2018