BUSHOR-1521; No.

of Pages 11

Business Horizons (2018) xxx, xxx—xxx

Available online at www.sciencedirect.com

ScienceDirect
www.elsevier.com/locate/bushor

EXECUTIVE DIGEST

Siri, Siri, in my hand: Who’s the fairest in

the land? On the interpretations,
illustrations, and implications of artificial
intelligence
Andreas Kaplan a,*, Michael Haenlein b

a
ESCP Europe, Heubnerweg 8-10, D-14059 Berlin, Germany
b
ESCP Europe, 79 Avenue de la République, F-75011 Paris, France

KEYWORDS Abstract Artificial intelligence (AI)–—defined as a system’s ability to correctly

Artificial intelligence; interpret external data, to learn from such data, and to use those learnings to
Big data; achieve specific goals and tasks through flexible adaptation–—is a topic in nearly
Internet of Things; every boardroom and at many dinner tables. Yet, despite this prominence, AI is still
Expert systems; a surprisingly fuzzy concept and a lot of questions surrounding it are still open. In
Machine learning; this article, we analyze how AI is different from related concepts, such as the
Deep learning Internet of Things and big data, and suggest that AI is not one monolithic term but
instead needs to be seen in a more nuanced way. This can either be achieved by
looking at AI through the lens of evolutionary stages (artificial narrow intelligence,
artificial general intelligence, and artificial super intelligence) or by focusing on
different types of AI systems (analytical AI, human-inspired AI, and humanized AI).
Based on this classification, we show the potential and risk of AI using a series of
case studies regarding universities, corporations, and governments. Finally, we
present a framework that helps organizations think about the internal and external
implications of AI, which we label the Three C Model of Confidence, Change, and
Control.
# 2018 Kelley School of Business, Indiana University. Published by Elsevier Inc. All
rights reserved.

1. Once upon a time, there was a

magic mirror . . .

Once upon a time, in a land far, far away, there lived

an evil queen who had a magic mirror. This magic
* Corresponding author mirror knew everything. It knew the faces of all the

https://doi.org/10.1016/j.bushor.2018.08.004
0007-6813/# 2018 Kelley School of Business, Indiana University. Published by Elsevier Inc. All rights reserved.
BUSHOR-1521; No. of Pages 11

2 EXECUTIVE DIGEST

people living in her kingdom and could easily tell her the personality of users and tailor political mes-
every morning that she was the fairest in the land. sages accordingly.
Knowing how important beauty was for the evil Today, we all have–—at least in principle–—access
queen, the magic mirror also tweaked her own to such devices which are marketed under the broad
image slightly to make her look a little bit more umbrella of artificial intelligence (AI). AI, or more
beautiful than she actually was in real life. And even generally the idea that computers can think like
when Snow White became more beautiful than her humans, has been discussed in literature for more
by having “skin as white as snow, lips as red as than half a century–—since the seminal work of
blood, and hair as black as ebony” (Grimm, Grimm, computer scientist Alan Turing. Today, first genera-
& Kliros, 1994), the magic mirror was still of help. It tion AI applications–—those that apply AI only to
told the evil queen where to find Snow White, with specific tasks and are generally referred to under
whom she lived, and that she fancied red apples. the label artificial narrow intelligence (ANI)–—are
The queen could use this knowledge to her advan- near ubiquitous. They enabled Facebook to recog-
tage to influence (well, poison) Snow White– nize faces in images and tag users, they allowed Siri
—although, as we all know, the story did not exactly to understand your voice and act accordingly, and
end as she had envisioned it. they enabled Tesla to develop self-driving cars. In
Who would not want to have such a magic mirror? the future, we may see the second generation of AI,
A mirror that shows not your real image but a artificial general intelligence (AGI), able to reason,
slightly improved version of it–—similar to the T8 plan, and solve problems autonomously for tasks
mobile phone developed by the Chinese technology they were never even designed for. And we might
firm Meitu, that uses Magical AI Beautification to possibly see the third generation, artificial super
make you look better in selfies. A tool that tracks intelligence (ASI), which are truly self-aware and
information from all around the kingdom in real conscious systems that, in a certain way, will make
time to tell you every morning what is happening– humans redundant. Such systems could apply AI to
—similar to the New York startup Dataminr, which any area and be capable of scientific creativity,
monitors the internet and social media applications social skills, and general wisdom, which is why some
to help companies take better PR and stock market call ASI true artificial intelligence. Figure 1 outlines
decisions. A crystal ball that looks deeply into the three stages of AI.
peoples’ souls to tell you which type of fruit (or In this article, we look more deeply into the
message) they are best influenced by–—similar to concept of artificial intelligence. We start by pro-
the British consulting firm Cambridge Analytica, viding a definition of the term and giving a classifi-
which used information from Facebook to assess cation of the different types of AI, specifically

Figure 1. Stages of artificial intelligence (AI)

ASI
(Conscious/Self-Aware, Above
Human-Level AI)
Artificial General Intelligence - Applies AI to any area
AGI - Able to solve problems in other
(Strong, Human-Level AI) areas instantaneously
- Applies AI to several areas - Outperforms humans in all areas
Artificial Narrow Intelligence - Able to autonomously solve
ANI problems in other areas
(Weak, Below Human-Level AI) - Outperforms/equals humans in
- Applies AI only to specific areas several areas
- Unable to autonomously solve
problems in other areas
- Outperforms/equals humans in
the specific area
Siri develops super-human
capabilities such as solving
complex mathematical problems
instantaneously or writing a best
Siri evolves into a humanoid robot seller in a heart (or clock) beat
with wide capabilities including
voice recognition, coffee
preparation, and writing skills
Siri can recognize your voice but
cannot perform other tasks like
BUSHOR-1521; No. of Pages 11

EXECUTIVE DIGEST 3

regarding business use. We then discuss how uni- to achieve specific goals and tasks through flexible
versities, corporations, and governments are al- adaptation.
ready using AI today, how they can use it in the Looking at this definition, it is obvious how AI
future, and the specific challenges they have to face differs from related concepts such as the Internet of
in the process. Finally, we provide a structured way Things (IoT) and big data. The IoT (Krotov, 2017;
to think about the organizational implications of AI Saarikko, Westergren, & Blomquist, 2017), which
both internally and externally, which we label as the describes the idea that devices around us are
Three C Model (Confidence, Change, Control). Our equipped with sensors and software to collect
discussion finishes with a brief glimpse into an AI- and exchange data, can be seen as one specific
enabled future and the question of if and where way of obtaining the external data required as an
humans will have a place in such a world. input for AI. The IoT is one input toward big data
(Lee, 2017), which describes data sets character-
ized by huge amounts (volume) of frequently up-
2. Interpretations of AI: As white as dated data (velocity) in various formats, such as
numeric, textual, or images/videos (variety). Big
snow, as red as blood, as black as data is, however, broader than the IoT since it also
ebony includes data collected through other means, such
as (mobile) social media applications (Kaplan, 2012;
Although articles about AI are abundant in popular Kaplan & Haenlein, 2010) or a firm’s internal data-
and business press in recent years, it is surprisingly base.
difficult to define what AI is and what it is not. Or, to AI uses external information obtained through
put it differently, there are about as many different IoT or other big data sources as an input for identi-
definitions of AI as there are ways to describe Snow fying underlying rules and patterns by relying on
White’s beauty, depending on whether one focuses approaches from machine learning which broadly
on her white skin, red lips, or black hair. To some speaking describes methods that help computers
extent, this is related to the problem of defining learn without being explicitly programmed. These
intelligence itself, which is not an easy task. More- methods can be relatively simple (think of the
over, the field of AI is moving so fast that what used regression analysis you learned during your MBA)
to be considered as intelligent behavior exhibited or eye-wateringly complex (such as deep neural
by machines 5 years ago is now considered barely networks, which form the basis of deep learning
noteworthy. We therefore start our analysis by pro- tools like Google’s DeepMind). Machine learning is
viding a definition of what we mean by AI in this an essential part of AI, but AI is broader than
article, followed by a classification of three main machine learning since it also covers a system’s
types of AI based on this definition. ability to perceive data (e.g., natural language
processing or voice/image recognition) or to con-
2.1. Definition trol, move, and manipulate objects based on
learned information be it a robot or another con-
A common way to define artificial intelligence is to nected device.
do so by referencing human intelligence, which can
be seen as the “biopsychological potential to pro- 2.2. Classification
cess information . . . to solve problems or create
products that are of value in a culture” (Gardner, To classify different types of AI, specifically regard-
1999, pp. 33—34). In 1955, the Dartmouth Research ing their business use, we borrow from the manage-
Project defined AI as the problem of “making a ment literature and specifically studies
machine behave in ways that would be called intel- investigating the skills shared by successful manag-
ligent if a human were so behaving” (McCarthy, ers and employees with above-average perfor-
Minsky, Rochester, & Shannon, 1955). In a similar mance (e.g., Boyatzis, 2008; Hopkins & Bilimoria,
manner, cognitive scientist Marvin Minsky consid- 2008; Luthans, Welsh, & Taylor, 1988; McClelland &
ered AI as “the science of making machines do Boyatzis 1982; Stubbs Koman & Wolff, 2008).
things that would require intelligence if done by This literature generally claimed that outstanding
men” (Minsky, 1968, p. v). For the purpose of this performance is strongly related to the presence
article, we follow this general line of thinking but of three skills or types of competencies: cognitive
aim to be more specific regarding the way in which intelligence (e.g., competencies related to
AI achieves this goal. Specifically, we define AI as a pattern recognition and systematic thinking), emo-
system’s ability to interpret external data correctly, tional intelligence (e.g., adaptability, self-confi-
to learn from such data, and to use those learnings dence, emotional self-awareness, achievement
BUSHOR-1521; No. of Pages 11

4 EXECUTIVE DIGEST

Figure 2. Types of AI systems

Expert Analytical Human- Humanized Human

Systems AI Inspired AI AI Beings
Cognitive Intelligence 8 9 9 9 9
Emotional Intelligence 8 8 9 9 9
Social Intelligence 8 8 8 9 9
Artistic Creativity 8 8 8 8 9
Supervised Learning, Unsupervised Learning,
Reinforcement Learning

orientation), and social intelligence (e.g., empathy, approach) by imitating the brain’s structure
teamwork, inspirational leadership). Figure 2 out- (e.g., through neural networks) and using vast
lines the types of AI systems. amounts of data to derive knowledge autonomously.
While the use of cognitive intelligence to classify This is similar to how a child would learn to recog-
AI seems straightforward, the applicability of emo- nize a face–—not by applying rules formalized by his/
tional and social intelligence requires some expla- her parents but by seeing hundreds of thousands of
nation. The mainstream view in psychology is that faces and, at some point, being able to recognize
intelligence is generally innate (i.e., a characteris- what is a face and what is a broom. This allows
tic that individuals are born with rather than some- dealing with tasks vastly more complex than what
thing that can be learned). Still, emotional and could be handled through expert systems. For ex-
social intelligence are related to specific emotional ample, while chess can be formalized through rules,
and social skills and it is these skills that individuals the Chinese board game Go cannot. Therefore it
can learn and that AI systems can mimic. While was never possible to build an expert system able to
machines and AI systems can obviously not experi- beat a human Go player. Yet a deep neural network
ence emotions themselves, they can be trained to can be trained to play Go simply by observing a very
recognize them (e.g., through the analysis of facial large number of games played by humans.
micro-expressions) and then adapt their reactions Based on these three types of competencies, we
accordingly. classify AI systems into three groups (see Figure 2):
Before discussing AI systems, it is important to
highlight that expert systems–—collections of rules 1. Analytical AI has only characteristics consistent
programmed by humans in form of if-then state- with cognitive intelligence. These AI systems gen-
ments–—are not part of AI since they lack the ability erate a cognitive representation of the world and
to learn autonomously from external data. In fact, use learning based on past experience to inform
expert systems represent a different approach al- future decisions. Most AI systems used by firms
together since they assume that human intelligence today fall into this group and examples include
can be formalized through rules and hence recon- systems used for fraud detection in financial ser-
structed in a top-down approach (also called sym- vices, image recognition, or self-driving cars.
bolic or knowledge-based approach). If an expert
system were programmed to recognize a human 2. Human-Inspired AI has elements from cognitive
face, then it would check for a list of criteria as well as emotional intelligence. These systems
(e.g., the presence of certain shapes, of a nose, can, in addition to cognitive elements, under-
of two eyes) before making a judgment based on stand human emotions and consider them in
embedded rules. Such systems tend to perform their decision making. Affectiva, an AI company
poorly during tasks that depend on complex forms founded by MIT, uses advanced vision systems to
of human intelligence, which are implicit and can- recognize emotions like joy, surprise, and anger
not be transferred easily to simple rules. That is not at the same level (and frequently better) as
to say that expert systems are not useful. IBM’s humans. Companies can use such systems to
famous Deep Blue chess-playing algorithm, which recognize emotions during customer interac-
beat Garry Kasparov in the late 1990s, was not AI tions or while recruiting new employees. We talk
but an expert system. Expert systems like Deep Blue about more examples in the following section.
have been key drivers in making AI (or what is
sometimes believed to be AI) more prominent 3. Humanized AI shows characteristics of all types
among the general public. of competencies (i.e., cognitive, emotional, and
Real AI as defined above uses a bottom-up ap- social intelligence). Such systems, which would
proach (also called connectionist or behavior-based be able to be self-conscious and self-aware in
BUSHOR-1521; No. of Pages 11

EXECUTIVE DIGEST 5

their interactions with others, are not available humans and out of reach of AI. This question is
yet. While progress has been made in recognizing difficult to answer given the tremendous progress
and mimicking human activities, building AI sys- AI has experienced over the past decade. Still, it
tems that actually experience the world in a seems likely that humans will always have the upper
fundamental way are a project for the (poten- hand where artistic creativity is concerned. Funda-
tially distant) future. mentally, AI is based on pattern recognition or curve
fitting (i.e., finding a relationship that explains an
As highlighted in our definition of AI stated above, a existing set of data points), while “creativity is
defining element of all those systems is the ability to intelligence having fun” as Albert Einstein put it.
learn from past data. For this, there are three broad At this stage, it seems unlikely that AI systems will
types of learning processes: supervised learning, be able to solve truly creative tasks. That being
unsupervised learning, and reinforcement learning. said, the entertainment company Botnik Studios
recently used AI to write a three-page chapter with
1. Supervised learning methods map a given set of the title “Harry Potter and the Portrait of What
inputs to a given set of (labeled) outputs. They Looked Like a Large Pile of Ash” after training their
are usually the least scary for managers since AI system on all seven of the popular fantasy novels.
supervised learning include methods many may So the borders are certainly quite fluid in that
be familiar with (at least in principle) from respect.
Statistics 101, such as linear regression or clas-
sification trees. That being said, more complex
methods like neural networks also fall into this 3. Illustrations of AI: Appl(e)ications
group. An example of supervised learning is to
use a large database of labeled images to sepa-
in universities, corporations, and
rate between those images showing a Chihuahua governments
and those showing a muffin.
Building on our classification of AI into analytical AI,
2. In unsupervised learning, the inputs are labeled human-inspired AI, and humanized AI, we look to
but not the outputs. This means that the algorithm three industries: universities, corporations, and
needs to infer the underlying structure from the governments. Specifically, we dive into how AI is
data itself. Cluster analysis, which aims at group- already shaping them and what future trends can be
ing elements in similar categories but where nei- expected in years to come (see Table 1). Like in the
ther the structure of those clusters nor their case of Snow White, we will see the world is rarely
number is known in advance, falls into this group. as simple as we would like it to be. For every sweet
Since the output is derived by the algorithm itself, side of the red apple that provides wonderful op-
it is not possible to assess the accuracy or correct- portunities, there is also a poisoned side that rep-
ness of the resulting output. Users therefore need resents very real risks.
to place greater trust and confidence into the AI
system and that can make managers uncomfort- 3.1. Universities
able. Speech recognition–—made familiar with
Apple’s Siri or Amazon’s Alexa–—can be conducted Many of the most significant advances in artificial
using unsupervised learning. intelligence have their origin in a university context
and given the technical nature of AI, this tendency
3. In reinforcement learning, the system receives an is most likely to go on in the future. The term AI
output variable to be maximized and a series of itself was coined at a workshop at Dartmouth Col-
decisions that can be taken to impact the output. lege in 1956, organized by the computer scientist
Think, for example, of an AI system that aims to John McCarthy who later became a professor at
learn playing Pac-Man, simply by knowing that Stanford. DeepMind, a British AI company acquired
Pac-Man can move up, down, left and right and by Google in 2014, was created by three scientists,
that the objective is to maximize the score ob- two of which met while working at University Col-
tained in the game. Software giant Microsoft uses lege London. In 2015, DeepMind developed Alpha-
reinforcement learning to select headlines on Go–—the first computer Go program to defeat a
MSN.com by rewarding the system with a higher human professional Go player. It is therefore natural
score when more visitors click on a given link. to start our analysis of the practical applications of
AI in an academic context to answer the question
Looking at AI this way raises the question of whether of whether universities may have sown the seeds
there are any skills that remain characteristic for of their own destruction by their research on AI,
BUSHOR-1521; No. of Pages 11

6 EXECUTIVE DIGEST

Table 1. Illustrations of AI applications within specific sectors

Analytical AI Human-Inspired AI Humanized AI
Universities Virtual teaching assistants AI-based career services Robo-teachers animating a
able to answer student able to identify emotions to student group by acting as
questions and tailor improve interview moderator and sparring
reactions to individual data techniques of students partners
Corporations Robo-advisors leveraging Stores identifying unhappy Virtual agents dealing with
automation and AI shoppers via facial customer complaints and
algorithms to manage client recognition at checkouts to addressing concerns of
portfolios trigger remedial actions unhappy customers
Governments Automation systems to set Virtual army recruiters AI systems able to
the brightness of interviewing and selecting psychologically train
streetlights based on traffic candidates based on soldiers before entering a
and pedestrian movements emotional cues war zone

similar to the spirits cited by Goethe’s “The Sorcer- outsource tedious tasks such as grading and re-
er’s Apprentice” or the poisoned side of the red sponding to repetitive student questions. This
apple of the evil queen. leaves professors, in principle, more time for their
Analytical AI applications are already starting to core competence of coaching, moderating, and
transform the profession of faculty members. facilitating discussions (Kaplan, 2018). Until, of
Georgia Tech uses an AI-based virtual teaching as- course, the next generation of humanized AI appli-
sistant called Jill Watson to answer student ques- cations will take care of those tasks as well.
tions. The performance of the system is so Whether this will ever be the case depends on
remarkable that many students only realized that this fundamental question: Will students prefer to
Jill Watson is not human after they were told. Other be educated by smart machines or by human pro-
universities, like the Technical University of Berlin fessors? The fact that AI systems are cheaper than
and Carnegie Mellon, similarly tested the use of highly paid faculty members, at least in the long
chatbots to streamline teaching and learning. But run, makes them preferable from the perspective of
the AI revolution does not stop at teaching alone. university deans who struggle for funding. But are
The British RELX Group, owner of Elsevier and they really the better choice if education becomes
LexisNexis, uses AI for automating systematic aca- less personal? Universities need to make a conscious
demic literature reviews or for supporting the re- decision in this context and prepare themselves for
view process through checks for plagiarism or the rise of AI. This will also allow them to better
misuse of statistics. In a research context, AI is prepare their students for a workplace in which AI
particularly useful for projects that aim to combine will become increasingly prominent. In this context,
ideas across scientific boundaries and that, by con- some researchers suggest that universities should
sequence, require knowledge of many different introduce a course on artificial intelligence and
literature streams that may be difficult to process humanity to answer questions of equity, ethics,
by humans. privacy, and data ownership which are of relevance
Human-inspired AI brings all of this to the next in this context (Keating & Nourbakhsh, 2018, p. 32).
level. In an online learning context (Kaplan &
Haenlein, 2016), universities could use AI to test 3.2. Corporations
whether students pay attention during a virtual
class by analyzing facial impressions collected Looking at corporations, AI has already started to
through a webcam. In a traditional setting, systems impact every single element of a firm’s value
like RENEE (named for retain, engage, notify, and chain and, in the process, transform industries in
enablement engine), developed by U.S.-based Cam- a fundamental manner, especially service industries
pus Management Corporation, can automatically (Huang & Rust, 2018). Analytical AI applications are
launch interventions based on student profiles, best used in human resource management to help with
practices, and other inputs. RENEE might in the the screening of CVs and selection of candidates in
future be able to read student emotions like sadness the form of advanced application tracking systems
or fear, allowing faculty and staff to identify the (ATS). In marketing and sales, AI is used to allow for
most effective coaching strategies or to spot cheat- better targeting and personalized communication.
ing in exams. All these systems will help faculty AI systems can identify thousands of psychotypes
BUSHOR-1521; No. of Pages 11

EXECUTIVE DIGEST 7

(Kosinski, Stillwell, & Graepel, 2013) and create idea was to generate a program that can pass the
messages that resonate well with their preferences, Turing test: If a person cannot distinguish whether
leading to tens of thousands of variations of the he/she is talking to a human or a machine, the
same message used every day. In customer service, machine exhibits intelligent behavior (Turing,
AI can be applied in the form of chatbots that can 1950). Today ELIZA has evolved into Sophia, an
generate automatic responses to inquires sent AI-inspired robot developed by David Hanson that
through social media channels or emails. Modern is so convincing Saudi Arabia granted it citizenship
versions like Google Duplex are even able to con- in 2017. Such tools are more than a PR stunt–—Sophia
duct phone calls that are difficult to distinguish is a highly demanded speaker and generated press
from conversations with a human counterpart. coverage reaching 10 billion readers in 2017. These
Looking at industry effects, the financial services robots can serve as buddies for senior citizens who
sector has seen the rise of financial technology live alone and, broadly speaking, revolutionize the
(fintech) startups which have revolutionized asset field of elderly care.
management through the creation of robo-advisors
and the analysis of financial transaction data (e.g., 3.3. Governments
by spotting early signs of dementia reflected in
erratic account movements). In retailing, AI is used Sophia’s citizenship status naturally leads to the
for inventory management with the holy grail being question of how AI should and could impact govern-
Amazon’s anticipatory shipping patent that deals ments, both directly and indirectly. Like universi-
with sending items to customers before they even ties and corporations, governments can use AI to
ordered them. In the entertainment sector, AI has make tasks more efficient and it is in this context
been used by newspapers like The Los Angeles where the concept of the good vs. bad side of the
Times to automatically write articles. In the near red apple becomes most obvious. The City of Jack-
future, AI will go beyond written text and create sonville uses analytical AI to manage intelligent
artificial videos in which the moving picture of a streetlights which decide on the brightness of each
person can be overlaid to any text the author lamp depending on traffic and pedestrian move-
desires (Suwajanakorn, Seitz, & Kemelmacher- ments collected by street cams. Another example
Shlizerman, 2017). This will give the idea of fake is the Southern Nevada Health District, which uses
news an entirely new dimension. AI combined with information from Twitter to de-
Human-inspired AI allows companies like Wal- cide which restaurants to visit for health inspec-
mart to identify unhappy and frustrated customers tions. A combination of natural language processing
by applying facial recognition techniques to people and geotagging helps to spot places where custom-
queuing up at checkouts, thus enabling interven- ers report food poisoning and identify those estab-
tion by either opening new cashiers or proposing lishments for inspection. In an experiment
snacks and drinks to customers. The same tools can conducted in Las Vegas, this approach resulted in
be used to automatically detect fraud and theft more demerits and citations, which ultimately
orders of magnitude more efficiently than a tradi- could lead to 9,000 fewer cases of food poisoning
tional store detective could. Online firms like Net- and over 500 fewer hospitalizations per year.
flix, Spotify, and Pandora already use AI to provide In the same vein, human-inspired AI is apparently
personalized recommendations for music and mov- used by the U.S. Army in the recruitment of future
ies. In the future, an analysis of your past choices soldiers through an advanced SGT Star AI system
combined with facial recognition through your that is rumored to be able to recognize emotions.
phone’s camera (think iPhone X) allows those firms SGT Star is an interactive virtual agent that applies
to also detect your current mood and propose AI to respond to questions, review qualifications,
matching entertainment content. Alternatively, and assign selected candidates to actual human
standalone applications like Replika, your AI friend recruiters. SGT Star does the workload of more than
developed by the San Francisco-based Luka Inc., 50 recruiters with a 94% accuracy rate and boosted
allows you to build a diary and, in a way, acts like engagement time for applicants from 4 minutes to
an AI-enabled therapist. This will likely be a major over 10 minutes. Of course, another way to lever-
threat to online therapy providers like BetterHelp age the power of AI in a military context is to rely on
or Talkspace. AI-enabled robotic soldiers. This is, unfortunately,
The combination of human-inspired AI and robot- not a science fiction idea but becoming a reality. In
ics is also where we can get a first glimpse into the 2013, over 100 researchers, security experts, and
world of humanized AI. In 1964, Joseph Weizen- company leaders wrote an open letter to the UN
baum from MIT created the first natural language asking to ban AI-enabled robots in war. Automatic
processing computer program called ELIZA. The systems, including drones, missiles, and machine
BUSHOR-1521; No. of Pages 11

8 EXECUTIVE DIGEST

guns, can lead to a level of escalation that older definition is better in the hope that we know it
readers may remember from the 1983 movie War when we see it, following the approach of Supreme
Games. Court Justice Potter Stewart when describing his
A natural question arises when combining AI and threshold test for obscenity (Jacobellis v. Ohio,
governments: Where does improvement end and an 1964).
Orwellian surveillance state begin? China has pro-
posed a social credit system which combines mass
surveillance, big data analytics, and AI to reward 4. Implications of AI: Are you afraid of
the trustworthy and punish the disobedient. In the poison?
proposed initiative, punishment for undesirable be-
havior can include flight bans and restriction relat- The above examples illustrate that AI will have
ed to private schools access, real estate purchases, implications for any kind of organization, both in-
or even taking a holiday. In Shenzhen, authorities ternally and externally. Internally, AI will allow a
already use facial recognition systems to crackdown multitude of tasks to be conducted faster, better,
on crimes like jaywalking; in Xiamen, users receive and at lower cost. In the medium term, this will not
mobile phone messages when they are calling citi- only affect simple tasks but also more complex
zens with low social credit scores (Xu & Xiao, 2018). ones; even knowledge-heavy industries like consult-
The examples bring up the question of regulation ing, financial services, and law will see major
and the need for government intervention in the changes. Externally, it will impact the relationship
domain of AI, especially when reaching humanized between firms and their customers, other firms, and
AI. While some voices argue for immediate and with society at large.
proactive regulation on a national and international To help organizations prepare for this future,
level given the quick progress of AI–—though it may we look more closely at three common traits that
otherwise be too late–—others are concerned that are of relevance both internally and externally:
regulation could slow down AI development and confidence, change, and control–—the three Cs of
limit innovation. The middle ground is to develop the organizational implications of AI (see Table 2).
common norms instead of trying to regulate tech- Like when Snow White decided to trust the evil
nology itself, similar to the consumer and safety queen by biting into the red apple, managers will
testing done for physical products. Such norms have to be trusted in their ability to manage and
could include requirements for the testing and consumers will have to put confidence in the
transparency of algorithms, possibly in combination company to not misuse their data in any way. In
with some form of warranty. This would also the fairytale, Snow White would admittedly have
allow for regulations to remain stable and eliminate done better not to put confidence in the queen. In
the need for constant updates in response to real life, there will have to be certain control
technological advances. This proposal is complicat- mechanisms to protect from damage, internally
ed by the idea of what AI is and what it can do. AI is by controlling the machines and externally
itself a moving target and more an issue of inter- through the State controlling the corporations
pretation than definition. Should AI be vaguely and institutions implementing AI. Finally, change
defined for legal purposes with the risk that every- will be ever present, be it change in employees’
thing could count as AI, or defined narrowly, job descriptions or the rapid change of copying
focusing only on certain aspects? Or perhaps no external competitors.

Table 2. The three Cs of organizational implications of AI

Internal External
Confidence Managers need to exude confidence with Consumers need to put confidence in the
respect to their employees in a fast- abilities and recommendations of an
evolving work environment organization’s AI systems
Change Employees need to constantly change and Competitors need to be monitored and
adapt their functions and skills through outperformed permanently by use of better
lifelong learning hardware or data
Control Machines need to be controlled to avoid States need to control the ecosystem of
autonomous decisions and implicit biases managers, employees, machines,
consumers, and competitors
BUSHOR-1521; No. of Pages 11

EXECUTIVE DIGEST 9

4.1. Internal the fact that AI systems can be deliberately hacked,

similar to the way a computer virus or random ware
4.1.1. Managers can take ownership of a PC. Even when leaving all
Managers need to adopt a leadership style that those concerns aside, AI systems are only as smart
engenders confidence from employees at a time as the data used to train them since they are, in
when AI will fundamentally transform the work- their essence, nothing more than fancy curve-fitting
place in unprecedented ways. Several leadership machines. Using AI to support a court ruling can be
skills will be essential, including enabling an open highly problematic if past rulings show bias toward
dialogue; proficiency in healthy conflict resolution; certain groups since those biases get formalized and
and human, ethical, open, and transparent man- engrained, which makes them even more difficult to
agement. Actions taken by such managers start with spot and fight against.
the smallest of changes. Instead of AI, IBM decided
to use the terms cognitive computing and augment- 4.2. External
ed intelligence to signal that systems are designed
to make employees more efficient, not to replace 4.2.1. Consumers
them. But they do have a much further reach. Any AI system implemented by firms is only useful if
Managers in the future need to be experts in assess- consumers accept it in one way or another. This
ing skills to identify the best position of each em- specifically means that consumers need to put con-
ployee in a hybrid system of people and AI. And they fidence into the recommendations provided by AI
need to become creative innovators in order to and the use of their personal data. While AI systems
emphasize the types of intelligence AI will not be are already capable of diagnosing X-ray images as
able to replace. Managers will need to act as em- proficiently as, or even better than, most physi-
pathetic mentors and data-driven decision makers. cians, most consumers would have a difficult time
trusting the verdict of a machine. A new stream of
AI research called explainable AI focuses on extract-
4.1.2. Employees
ing rules ex-post so that humans can at least get a
AI will lead to a constant change in the type of work
rudimentary understanding of the working of AI
conducted by employees. While it is unlikely that AI
systems. A necessary precondition to this confi-
will be able to replace entire jobs completely, there
dence, besides better explanation, is that firms
will be more and more tasks that will be outsourced
do not overpromise on the potential of AI. It is well
to AI and employees will have to adapt. Combining
known from service research that consumers judge
the rapid speed of AI development with the increas-
the quality of a service based on the gap between
ing lifespan of humans (and employees) gives the
their expectations and reality (Parasuraman,
concepts of lifelong learning and career flexibility a
Zeithaml, & Berry, 1985). All of us who have tried
whole new meaning (Pucciarelli & Kaplan, 2016). In
to speak casually with their Amazon Alexa, Google
such a world, employees will have to constantly
Home, or Apple Siri will know that the frictionless
develop new skills to complement advances in AI
expectations created by the advertising for those
technology. Entrepreneurs, innovators, creators,
products rarely match reality.
and generally people who are keen on taking on
new challenges and opportunities will be of increas-
4.2.2. Competitors
ing importance. The necessary training can be fi-
In the world of AI, firms will continually be chal-
nanced by firms themselves (e.g., telecom giant
lenged to establish a long-lasting competitive ad-
AT&T spends $30 million annually to reimburse em-
vantage. Since the basic mathematical concepts
ployees for the cost of training in digital skills) or
underlying AI are widely known–—and often avail-
provided by outside entities such as trade unions,
able to everyone–—there are only two ways firms can
especially for smaller firms where in-house training
outperform their competition: through faster hard-
may not be efficient.
ware or more data. In terms of hardware, chip
designers are already working on developing new
4.1.3. Machines types of CPUs that embed basic AI tools such as
Machines and AI systems, first and foremost, require neural nets on the hardware (chip) level instead of
control by humans. Even the smartest AI systems programming them into a general purpose system.
can make very stupid mistakes. Look at the systems Looking at data, there will likely be a consolidation
powering self-driving cars. Tesla’s autopilot con- toward few large firms that acquire more and more
fused a white truck with a cloud in the sky and data, which leads to better AI systems and even
Uber’s self-driving car did not recognize a pedestri- more data in a self-fueling spiral of growth. In such a
an in Arizona. And these mistakes do not account for world, firms need to constantly change to adapt to
BUSHOR-1521; No. of Pages 11

10 EXECUTIVE DIGEST

evolutions in performance among existing compet- lifetime. So it seems clear that a system with true
itors as well as the emergence of new firms. ASI would easily be able to outperform humans. Yet,
what is often forgotten is that humans are used to
4.2.3. States thinking on a human level while an ASI system would
Similar to the idea that AI systems need to be think on an ASI level. Just as humans can never truly
controlled to avoid making stupid mistakes, the understand how chimpanzees think, despite the
whole ecosystem of managers, employees, ma- fact that they share 99% of our DNA, we will not
chines, consumers, and competitors requires close be able to understand how an ASI system thinks.
monitoring from the side of the State, as already This limits our ability to control such systems, which
mentioned. There is a need for rules, legislation, again makes them appear more dangerous than
and control to avoid AI getting out of hand. This can useful.
include the requirement to spend a certain percent- Another issue deals with the displacement of the
age of revenue in training to prepare employees for human workforce by machines. Most jobs consist of
upcoming challenges. It also may consist of artifi- a series of tasks and not all of them can easily be
cially constraining the competition between ma- conducted by AI. While a neural network can easily
chines and humans. In France, a law exists that beat the best players in Go, it has a much harder
limits access to the IT self-service systems of public time assembling an IKEA chair–—a fact labeled as the
administration bodies after working hours. Kamprad test in analogy to the well-known afore-
A central issue in this context is the topic of privacy mentioned Turing test. The vast abilities of AI com-
and data protection. In the near future, we will face a bined with the increasing availability of data makes
wide range of dilemmas that require balancing social it likely that the shift to AI has a more fundamental
advances in the name of AI and fundamental privacy impact on workforce in general than the Industrial
rights. The European Union recently made a signifi- Revolution from 1820—1840. This leads to a series of
cant step in legislating this question by introducing questions, starting from the idea of a universal basic
the General Data Protection Regulation (GDPR). income to fundamental issues that philosophy and
While such rules clearly protect consumers, they also religion will need to deal with, namely how humans
mean that the EU will unlikely be able to challenge can find purpose in life when all their work is
the U.S. or China in AI dominance anytime soon. This conducted by machines.
puts States in the complex position to decide how Given all those challenges and open questions, it
much privacy can and should be sacrificed on the altar is not surprising that the views on AI range from
of economic growth. Different countries are already outright alarmist as expressed by Elon Musk to
making different decisions in this respect and those euphoric like the vision of futurist Ray Kurzweil.
decisions will likely shape AI trends in years and even Recently deceased theoretical physicist Stephen
decades to come. Hawking called AI “either the best, or the worst
thing, ever happen to humanity” (Herm, 2016). This
brings us back to Snow White who, at the end, had a
5. And they lived happily ever after happy ending with Prince Charming, even after
being temporarily poisoned by the apple. We just
Years ago, Holloway and Hand (1988, p. 70) pub- have to hope that humanity will not end up stuck in
lished an article in Business Horizons on artificial a glass coffin made out of the illusion of AI conve-
intelligence that started as follows: “Artificial in- nience and that one day, even if it’s only at the last
telligence is no longer an academic term, but a moment, some prince–—or princess!–—will come.
reality. And, in some companies, it seems that
the AI system has replaced the human as the busi-
ness and ethical decision maker.” Today, we know
References
that this statement was probably a bit premature.
But the revolution Holloway and Hand predicted did
Boyatzis, R. E. (2008). Competencies in the 21st century. Journal
happen eventually and it comes with a series of of Management Development, 27(1), 5—12.
questions that will need to be answered. Gardner, H. (1999). Intelligence reframed: Multiple intelli-
One question deals with the third generation of gences for the 21st century. New York, NY: Basic Books.
AI–—artificial super intelligence (ASI)–—and whether Grimm, J., Grimm, W., & Kliros, T. (1994). Snow White and other
fairy tales. North Chelmsford, MA: Courier Corporation.
this is something to aim for or to avoid. Technically, Herm, A. (2016, October 16). Stephen Hawking: AI will be ‘either
even a standard microprocessor available for $200 best or worst thing’ for humanity. The Guardian. Available at
today runs at 10 million times the speed as a human https://www.theguardian.com/science/2016/oct/19/
neuron and computers can memorize more pieces of stephen-hawking-ai-best-or-worst-thing-for-humanity-
information in 1 second than a human could in a cambridge
BUSHOR-1521; No. of Pages 11

EXECUTIVE DIGEST 11

Holloway, C., & Hand, H. H. (1988). Who’s running the store, McCarthy, J., Minsky, M. L., Rochester, N., & Shannon, C. E.
anyway? Artificial intelligence!!! Business Horizons, 31(2), (1955). A proposal for the Dartmouth summer research
70—76. project on artificial intelligence. Available at http://
Hopkins, M. M., & Bilimoria, D. (2008). Social and emotional www-formal.stanford.edu/jmc/history/dartmouth/
competencies predicting success for male and female execu- dartmouth.html
tives. Journal of Management Development, 27(1), 13—35. McClelland, D. C., & Boyatzis, R. E. (1982). Leadership motive
Huang, M.-H., & Rust, R. T. (2018). Artificial intelligence in pattern and long-term success in management. Journal of
service. Journal of Service Research, 21(2), 155—172. Applied Psychology, 67(6), 737—743.
Jacobellis v. Ohio, 378 U.S. 184 (1964) Minsky, M. L. (1968). Semantic information processing. Cam-
Kaplan, A. M. (2012). If you love something, let it go mobile: bridge, MA: MIT Press.
Mobile marketing and mobile social media 4  4. Business Parasuraman, A., Zeithaml, V. A., & Berry, L. L. (1985). A
Horizons, 55(2), 129—139. conceptual model of service quality and its implications for
Kaplan, A. M. (2018). A school is “a building that has 4 walls . . . future research. Journal of Marketing, 49(4), 41—50.
with tomorrow inside”: Toward the reinvention of the business Pucciarelli, F., & Kaplan, A. (2016). Competition and strategy in
school. Business Horizons, 61(4), 599—608. higher education: Managing complexity and uncertainty.
Kaplan, A. M., & Haenlein, M. (2010). Users of the world, unite! Business Horizons, 59(3), 311—320.
The challenges and opportunities of social media. Business Saarikko, T., Westergren, U. H., & Blomquist, T. (2017). The
Horizons, 53(1), 59—68. Internet of Things: Are you ready for what’s coming? Business
Kaplan, A. M., & Haenlein, M. (2016). Higher education and the Horizons, 60(5), 667—676.
digital revolution: About MOOCs, SPOCs, social media, and Stubbs Koman, E., & Wolff, S. B. (2008). Emotional intelligence
the Cookie Monster. Business Horizons, 59(4), 441—450. competencies in the team and team leader: A multi-level
Keating, J., & Nourbakhsh, I. (2018). Teaching artificial intelligence examination of the impact of emotional intelligence on team
and humanity. Communications of the ACM, 61(2), 29—32. performance. Journal of Management Development, 27(1),
Kosinski, M., Stillwell, D., & Graepel, T. (2013). Private traits 55—75.
and attributes are predictable from digital records of human Suwajanakorn, S., Seitz, S. M., & Kemelmacher-Shlizerman, I.
behavior. Proceedings of the National Academy of Sciences (2017). Synthesizing Obama: Learning lip sync from audio.
of the United States of America (PNAS), 110(15), 5802— ACM Transactions on Graphics, 36(4), 95:1—95:13.
5805. Turing, A. (1950). Computing machinery and intelligence. Mind,
Krotov, V. (2017). The Internet of Things and new business 59(236), 433—460.
opportunities. Business Horizons, 60(6), 831—841. Xu, V. X., & Xiao, B. (2018). China’s social credit system seeks to
Lee, I. (2017). Big data: Dimensions, evolution, impacts, and assign citizens scores, engineer social behaviour. Australia
challenges. Business Horizons, 60(3), 293—303. Broadcasting Corporation. Available at http://www.abc.net.
Luthans, F., Welsh, D. H. B., & Taylor, L. A., III. (1988). A au/news/2018-03-31/chinas-social-credit-system-punishes-
descriptive model of managerial effectiveness. Group and untrustworthy-citizens/9596204
Organization Studies, 13(2), 148—162.