40 ISE Magazine | www.iise.org/ISEmagazine
Engineering
the intelligent
enterprise
Augmented intelligence can’t match human
thinking but can optimize business processes
By Joseph Byrum
January 2019 | ISE Magazine 41
Science fiction writers say the businesses of the
future will be run by self-aware robots. After
all, these devices can make intelligent decisions
unclouded by fear or other emotions. They will
work 24/7 without the need for pay, break time,
unionizing or vacation. Such an enterprise would
outperform the fallible human competition – at least un-
til the last act of the story, when the robots inevitably go
rogue.
While last-minute twists are critical to an entertaining
plot, more sober researchers in the artificial intelligence
field explain that we are far from achieving the general
intelligence needed for a robot-run enterprise to become
reality. So the question then becomes: How do we extract
the greatest performance out of what we know today is pos-
sible with man and machine? The starting point must be to
distinguish the strengths and weaknesses of each.
The limits of human understanding
Based on the number of neurons and connections in the hu-
man brain, Northwestern University psychology professor
Paul Reber, Ph.D., estimates we have a memory capacity
roughly equivalent to 2.5 petabytes of storage, though di-
rect comparisons are obviously difficult. Whatever the exact
gure, human memory is finite, though one can always add
more storage arrays to ensure a machine will never “forget.
Mankind, on the other hand, can perform some impres-
sive mental feats. Two millennia ago, bards would travel
from village to village to recite works like Homers Iliad
from memory. It might take a few evenings to get through
the entire 15,693-line epic poem that was passed around for
generations by word-of-mouth long before books and wide-
spread literacy were common.
Multinight campre readings have gone out of style, so
a more modern bard might resemble Rajveer Meena, who
in 10 hours can recite pi to 70,000 digits. Other exemplars
of human processing ability include Scott Flansburg, who
took just 15 seconds to add a randomly selected two-digit
number to itself over and over 36 times, and Vikas Sharma,
who calculated 15 large number roots in one minute. These
demonstrations, recognized by Guinness World Records as the
height of human ability, are trivial compared to machines
able to calculate pi to 22 trillion digits.
While the most talented human cant come close to compet-
ing against machines in arithmetic and memory, dont take the
logical leap to conclude that machines are better at reasoning.
The simplest questions asked of a child can trip up a computer.
This is evident to anyone who has used the amazing tools of
text recognition, live translation and voice recognition. The
increasingly ubiquitous digital assistants are as deeply frustrat-
ing as they are impressive, with quite a way to go before they
can be considered replacements for humans.
The reason for this is straightforward. The assistants rely
upon preprogrammed responses and Wikipedia entries to
generate answers to expected questions. This makes them
more like digital parrots with a large vocabulary than intel-
ligent AI systems with a grasp of language and nuance. To say
they lack critical thinking abilities is not to deny their useful-
ness. Rather, it is a recognition of the inherent limits of the
underlying technology used in these machines.
Similarly, most children can walk before they reach the
age of 2, but only the most highly engineered robots are able
to walk on complex terrain without falling over. YouTube is
lled with videos documenting these less successful efforts.
Humans excel at judgment and creativity. We can take
ideas, mix them together and think outside the box to create
works that are truly new. Constrained by logic, machines
cannot come up with responses that arent preprogrammed.
Everything they do is, by denition, programmed. They can
only simulate spontaneity; even random number generation
must be faked.
Sure, AI has made art, movies, poetry and music. Those
are works of art in the same sense as the canvas painted by an
elephant trained to wield a brush as seen at www.elephantart.
com, or the “selfie” taken by a monkey that presses the remote
trigger for a tripod-mounted camera. The AI creates what
passes for art by sampling a range of different examples of
paintings, movies, poems and songs. It uses learning algo-
rithms to extract the various elements common to each, then
generates and recombines those elements in a “new” way us-
ing pseudo-random number generation.
This output has the appearance of creativity without the
inspiration. There is no emotional connection to the subject
matter any more than there is an understanding of the mean-
ing of the brush strokes or musical notes.
Where the machines fall short, humans excel. Likewise,
the qualities humans most lack can be supplemented by ma-
chines. The intelligent enterprise recognizes this and pairs
the most powerful aspects of machines – analysis and mem-
ory – with the most powerful aspect of humans – judgment
and creativity.
Digging deeper into how it works
Judgment and creativity arent the province of machines
because causal reasoning is not easily reduced to mathemat-
ical calculation. As the work of UCLA computer scientist
Judea Pearl has shown, mining statistics and then applying a
few calculations to a data set cannot come close to creating
an AI capable of matching wits with a human. Causal infer-
ence is the essential ingredient for achieving human-level
intelligence. Otherwise you just have systems that mimic
human speech patterns like a digital parrot unable to un-
derstand the words being said.
Pearl has done a masterful job outlining the mathemati-
S
42 ISE Magazine | www.iise.org/ISEmagazine
Engineering the intelligent enterprise
cal models of causation necessary to assist machines in help-
ing us answer the question “Why?” He describes what he
means using a classic example of causal inference in The
Book of Why that he co-authored with Dana Mackenzie:
A fire broke out after someone struck a match, and the
question is ‘What caused the fire, striking the match or the
presence of oxygen in the room?’ Note that both factors are
equally necessary, since the fire would not have occurred
absent one of them. So, from a purely logical point of view,
the two factors are equally responsible for the fire. Why,
then, do we consider lighting the match a more reasonable
explanation of the fire than the presence of oxygen?
This problem can be reduced to the form of a counterfac-
tual expression, revealing that the probability that lighting
the match caused the fire is greater than the probability the
presence of oxygen is the cause. This form of reasoning
provides an insight not available from simple statistics that
can only point us toward associations.
A visit to Spurious Correlations at www.tylervigen.com
provides plenty of examples where associative reasoning
leads to hilariously misleading results, such as the surprising
link between per capita consumption of mozzarella cheese
and the number of civil engineering doctorates awarded
each year as depicted in Figure 1.
The ability to distinguish correlations that matter from
those that dont can unlock crucial analytical capabilities.
A cognitive engine that can perform some level of causal
evaluation can excel at sorting data in terms of its impor-
tance, separating the noise, likely a coincidence, from the
signal – information that reveals significant trends. Know-
ing “why” something is the case extends the power of AI
far beyond mere imitation.
Technology approaching the qualities of human intelli-
gence still is not capable of replacing the human mind. To-
day’s cognitive engines have nowhere near the level of self-
awareness created by sci-fi writers for generations. They are
merely tools; what we can do with them is make the most
of business processes that can benefit from causal analysis.
Putting it all together: The Iron Man ‘suite
Even armed with causal reasoning, a cognitive AI system is
not particularly effective on its own. There is only so much
that can be accomplished by processing data, evaluating
the most relevant factors and simulating potential actions.
When an experienced human user takes that output and
spots data that conventional analytical methods would have
overlooked, the value of augmented intelligence becomes
clear. Such systems allow humans to act on better intel-
ligence, making choices informed by a solid understanding
of probable outcomes.
AI in the form of augmented intelligence assists human
experts in completing tasks with greater efciency. With
causal inference, an AI system can, for instance, better tar-
get marketing efforts by understanding which groups are
truly interested in a product and avoid spurious correla-
tions. Civil engineering schools wont waste efforts mar-
keting their graduate programs to mozzarella cheese lovers,
but more important uses are taking shape in the healthcare
sector.
Evariant is a company that helps large hospitals optimize
marketing efforts by predicting patients’ upcoming needs
by analyzing medical data. At first, the company hand-cod-
ed each algorithm it used, devising a new solution for each
customer. Then it realized there was a better way. It turned
FIGURE 1
A cheesy view of cause and effect
This humorous graph from Spurious Correlations at www.tylervigen.com shows that cause and effect aren’t always linked, even if the data
indicate otherwise.
Credit: Spurious Correlations
January 2019 | ISE Magazine 43
to DataRobot, a firm that offered a system to sort through
hundreds of algorithms and find the one for each particu-
lar application that would be statistically reliable and valid.
This was not a replacement for Evariants data scientists;
rather, the system took over mundane and routine coding
tasks, freeing experts to perform more effectively.
The benets of augmented intelligence extend beyond
marketing departments and data science teams. Even enter-
prise lawyers can now take advantage of systems like Klar-
ity, which reads through standard contracts to decide if its
worth the time for an attorney to review the terms or if
it’s just the usual boilerplate with no real risks. The system
draws out all the important terms of the agreement so they
can be reviewed at a glance and checked in detail when
necessary. The tool helps bypass the legal bottleneck and
speed up the approval of important deals.
Combining the analytical power of machines with the
judgment and creativity of a human is an arrangement
I compare to an “Iron Man” suit. In the movies and the
comics, Tony Stark is just an ordinary man when it comes
to physical ability. Once he dons his AI-powered suit, his
overall effectiveness grows as the suit makes suggestions
and manages the small details. The fictional example shows
us the value of pairing the humans best abilities with the
best abilities of the machine.
For the role of AI in business, it wouldnt be a physical
suit but a software suite that endows a financial analyst,
factory manager or CEO with powers that exceed those of
ordinary humans.
The enhanced judgment would deliver better optimized
performance, and a business built around such technology
would rightly be called an intelligent enterprise. Though
augmented intelligence does not make for as entertaining
a story, it does make for a profitable company. Considering
the competitive edge that augmented intelligence can pro-
vide, most future businesses will likely become intelligent
enterprises. All the rest wont be works of fiction; they’ll be
works of history.
Joseph Byrum is the chief data scientist at Principal Financial
Group. He was previously senior R&D and strategic marketing ex-
ecutive in life sciences-global product development, innovation and
delivery at Syngenta. In that role, he was chief architect of initia-
tives that won Syngenta the 2016 ANA Genius Award in Mar-
keting Analytics and 2015 Franz Edelman prize for contributions
in operations research and the management sciences. He holds an
MBA from the University of Michigan’s Stephen M. Ross School
of Business and a doctorate in genetics from Iowa State University.
Smarter AI: Machine learning
without negative data
A research team from the RIKEN Center for Advanced Intelligence
Project has developed a method for machine learning that
allows artificial intelligence to make classifications without what
is known as “negative data,” a finding that could lead to wider
application to a variety of classification tasks.
Classifying things is critical for humans seeking to detect
spam mail, fake political news or read objects or faces. When
using AI, such tasks are based on “classification technology” in
machine learning – having the computer learn via the boundary
separating positive and negative data. “Positive” data would be
photos featuring a happy face; “negative,” photos with a sad
face. Once a classification boundary is learned, the computer
can categorize the data.
Yet such technology requires both positive and negative
data for the learning process, and negative data are not always
available. In a real-life application, a retailer trying to predict
customer habits can find data on those who purchased from
them (positive) but not those who did not (negative).
Researchers succeeded in developing a method to let
computers learn a classification boundary only from positive
data and information on its confidence (positive reliability)
against classification problems of machine learning that divide
data positively and negatively.
“This discovery could expand the range of applications where
classification technology can be used,” lead author Takashi
Ishida said.