DATE DOWNLOADED: Thu Nov 11 18:25:02 2021

SOURCE: Content Downloaded from HeinOnline

Citations:

Bluebook 21st ed.

Phoebe V. Moore, The Mirror for (Artificial) Intelligence: In Whose Reflection?, 41
COMP. LAB. L. & POL'y J. 47 (2019).

ALWD 6th ed.

Moore, P. V., The mirror for (artificial) intelligence: In whose reflection?, 41(1)
Comp. Lab. L. & Pol'y J. 47 (2019).

APA 7th ed.

Moore, P. V. (2019). The mirror for (artificial) intelligence: In whose reflection?.
Comparative Labor Law & Policy Journal, 41(1), 47-68.

Chicago 17th ed.

Phoebe V. Moore, "The Mirror for (Artificial) Intelligence: In Whose Reflection?,"
Comparative Labor Law & Policy Journal 41, no. 1 (Fall 2019): 47-68

McGill Guide 9th ed.

Phoebe V Moore, "The Mirror for (Artificial) Intelligence: In Whose Reflection?"
(2019) 41:1 Comp Lab L & Pol'y J 47.

AGLC 4th ed.

Phoebe V Moore, 'The Mirror for (Artificial) Intelligence: In Whose Reflection?'
(2019) 41(1) Comparative Labor Law & Policy Journal 47.

MLA 8th ed.

Moore, Phoebe V. "The Mirror for (Artificial) Intelligence: In Whose Reflection?."
Comparative Labor Law & Policy Journal, vol. 41, no. 1, Fall 2019, p. 47-68.
HeinOnline.

OSCOLA 4th ed.

Phoebe V Moore, 'The Mirror for (Artificial) Intelligence: In Whose Reflection?'
(2019) 41 Comp Lab L & Pol'y J 47

-- Your use of this HeinOnline PDF indicates your acceptance of HeinOnline's Terms and
Conditions of the license agreement available at
[Link]
-- The search text of this PDF is generated from uncorrected OCR text.
-- To obtain permission to use this article beyond the scope of your license, please use:
Copyright Information
 THE MIRROR FOR (ARTIFICIAL)
INTELLIGENCE: IN WHOSE REFLECTION?

Phoebe V. Mooret

While the original days of artificial intelligence (AI) research accepted

there would be some level of human intervention in experimentation, these
days, researchers hope-and even expect machines will behave
autonomously. Workplace professionals and work designers have joined the
current so-called AI "arms race" by introducing a variety of AI-augmented
tools and applications to work design experimentation, attributing specific
types of intelligences to machines. The integration of new technologies into
the workplace always has implications for liability and safety and health
risks, but the newest uses of AI tools introduces specific questions around
regulatory and ethical implications, which are addressed in this Special Issue.
The present articles deals with one aspect of these rapidly emerging
issues that has not, as yet, been fully debated: what is meant by "intelligence"
in discussions of the intelligence we hope can become increasingly human-
like, but still supposedly artificial? Who or what is mirroring who (or what)
in our search for human-like intelligence in machines? In 1972, Joseph
Weizenbaum observed that discoveries and inventions in specific historical
moments allow humans' self-image to develop accordingly, where, e.g.,
germ theory and studies on evolution challenged the hegemony of religious
ideas, or the invention of the microscope paved the way for other social
changes. 2 Likewise, the introduction of the computer has led to a paradigm
shift. The computer is not the same as the other machines which humans have
made. It has allowed us to seriously consider whether there can be a machine
that thinks and behaves almost identically to ourselves.

t Associate Professor Dr. Phoebe V Moore, Management & Organisation division, University of
Leicester, School of Business and WZB Social Research Centre Berlin Guest Researcher. Article prepared
for Special Issue, "Automation, Artificial Intelligence, and Labor Law," edited by Prof. Valerio de Stefano
for the Comparative Labor Law and Policy Journal and for presentation at the Regulating for Decent
Work conference, International Labour Organisation, Switzerland 8 - 10 July 2019.
1. Earlier versions of this paper and some of the research informing the work are found in Moore's
WZB Berlin Social Science Centre discussion paper under the same title (2019 fc) written as part of her
WZB Weizenbaum Research Fellowship Autumn 2018; and one commissioned special report Moore
wrote for the European Union Agency for Safety and Health (EU-OSHA), entitled OSH and the Future
of Work: Benefits & Risks of ArtificialIntelligence tools in workplaces.
2. Joseph Weizenbaum, On the Impact of the Computer on Society: How does one insult a
machine? 176 SCIENCE 609 - 614 (1972).

47
48 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47

After two "AI winters" in research lasting 1974 - 1980s and 1987

-
1993 were perceived to have flopped due to failed experiments in, e.g.,
machine translation of written documentation, the quite generous
government research funding during the Cold War was withdrawn. But now,
in 2019, we seem to have entered a new summer of AI innovations, where
machinic autonomous intelligence is held in high esteem and potentially even
prioritized over human decision-making, such as in the workplace. The
intelligence attributed to machinic AI-augmented tasks and aptitudes
includes AI-augmented chatbots, which are designed to assist workers by
taking basic questions from online banking customers; and wearable
technologies such as HoloLenses in factories used to replace in-person
training; all of which apply and rely on a type of intelligence that I call
"assistive." Human Resource (HR) tools called "people analytics" that apply
machine learning tools to predict worker performance and behavior are
expected to demonstrate a "predictive" intelligence. Algorithms to assign
customers to workers via platforms to drivers, riders, and clickworkers use a
"prescriptive" intelligence based on workers' profiles, locations, and
performance scores. Facial and mood recognition by way of videoing
subjects, and sentiment analysis via data mining technologies that aid with
identifying customer service workers' emotions in, e.g., call centers, are
practices called "affect recognition" or "emotion coding," which rely on
"affective" intelligence. Cobots in the factory, which work alongside people
in moving boxes across the console, are perceived to hold "collaborative"
intelligence.
Even as these practices also contribute to automation of entire jobs or
some tasks, both routine and more complex, the thinking behind the
introduction of AI directly into workplaces, reflects an intelligence that is
assumed to be that of humanity during the current period of Industry 4.0,
where discussions about AI rely on a productivity mindset, with relevant
surrounding features. Weizenbaum also noted in his classical piece "On the
Impact of the Computer on Society: How does one insult a machine?" that
"it is the autonomy of the computer we value," and also that, "time after time,
science has led us to insights that, at least when seen superficially, diminish
man." 3 The insights of this great thinker who was himself a forefather of AI,
are worth recalling, as we begin to increasingly expect our tools to start to
make decisions for us. Given the specific intelligences we now ascribe to AI-
augmented applications in the workplace, it is eminently reasonable to throw
caution on these processes and invite meaningful debate, which this article
hopes to both contribute to where it is occurring, but also to clearly indicate
caution in this arena.

3. Id at 610.
2019] THE MIRROR FOR (ARTIFICIAL) INTELLIGENCE 49

To legitimately discuss AI and its contemporary relevance and impact

on workplaces and working conditions, this article starts by outlining a short
history of discussions about intelligence that has been attributed to humans,
by humans. Cybernetics researchers sought for the best way to teach
machines to control themselves through communication; because it was
theorized that all of human interaction and attempts to influence one another
is linked to our ability for communication; adapting Pavlov's stimulus
experiments and focusing on complex mathematical models involving
functional and structural properties of "nerve nets," where
neurophysiological and neuroanatomical findings would help researchers
identify how neurons behave and thus inform machinic building in
humanity's image. These researchers, in fact, looked far more explicitly at
the interplay of machine/human intelligences than AI researchers. In fact, the
authors of Stanford University's One Hundred Year Study of Artificial
Intelligence claim that there is a "lack of a precise, universally accepted
definition of AI," which has actually "helped the field to grow, blossom and
advance at an ever-accelerating pace." 4 What these authors overlook is that
AI is not only about intelligence, but AI relies on intelligent behavior: as
cited below, the very person who is credited with inventing the term "AI,"
described AI to be the activity of "making a machine behave in ways that
would be called intelligent if a human were so behaving." 5 There is a wealth
of research in philosophy, sociology, and psychology, primarily by white
men, about "what is intelligence," and the thinkers who influenced early AI
researchers, even if not precisely referenced, lend purpose to "intelligent"
behavior. So, the first two sections ask, who is reflecting who (or what) in
the search for machinic intelligence?
The article then discusses how now intelligence is being attributed to
machines in the context of the integration of AI into specific practices in HR
practices, robotics, and algorithmic management in workplaces. Software
designers and engineers are creating calculation machines for workplace
decision making and design which are inextricably linked to the work design
frameworks into which they are integrated, putting a cadre of digital experts
in the driving seat for what might have once been the role of a management
guru with little or no technical expertise. What kinds of intelligence are
expected from AI-augmented machines in this context, and why does it
matter? This section furthermore explores what this may mean for human
workers, as technology integrators, who embrace new processes, imagine
specific machinic intelligences to have the potential to fulfill specific goals
which align with the types of human intelligences also needed to get work

4. JACOB TURNER, ROBOT RULES: REGULATING ARTIFICIAL INTELLIGENCE 1 (2019).

5. John McCarthy, Marvin Minsky, Nathaniel Rochester& Claude Shannon, A Proposalfor the
Dartmouth Summer Research Project on Artificial Intelligence, Dartmouth Proposal (Aug. 31, 1955),
available at [Link]
50 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47

done in the contemporary "era of agile" I have outlined elsewhere. 6 While

these intelligences appear innocuous at first glance, the seemingly inherent
value and reliability of these new dimensions of workplace practices is
assumed and of concern, if machinic is prioritized above human intelligence,
a definitive dimension of the employment relationship will potentially be
disrupted.
In conclusion, it is asked, what is this future of work, where machines
are increasingly ascribed superior, but still "artificial" intelligence, and how
will workers and our larger political and economic surroundings, ultimately,
be affected?

I. THE INTELLIGENCE OF HUMANS, BY HUMANS

Even before the term AI was conceived, theories of cognition,

consciousness and ideas were prevalent in thoughts about thinking. For lack
of space, it will be impossible to cover all ontological and epistemological
debates surrounding work design and management theories leading up the
present iterations of AI as it is influencing the workplace-which is the theme
of this article-of course. Scientific management is probably the best-known
era for linking machines with human behavior, where the machine is seen as
a superior calculator. But some important moments in the early developments
of thinking about human reason stand out, some of which were later
contemplated by the school of cybernetics, and then AI.
In British empirical philosophy, the mind is seen to be made up of ideas.
In the chapter "Of Reason and Science" in the Leviathan, Thomas Hobbes
mused: "When a man reasons, he does nothing else but conceive a sum total
from addition of parcels ... for reason is nothing but reckoning."7 So,
humans' capacity for reason, which animals were not expected to have, was
seen to be a process whereby humans simply carve the world into symbolic
units and use sums to make decisions, informing intention. We can consider
the consequences of our actions and make theories and aphorisms, reasoning
and reckoning: "not only in number, but in all other things whereof one may
be added unto or subtracted from one another." So, this form of human
intelligence is portrayed in terms of calculative processes. Another well-
known British empiricist, John Locke, held that the ideas that make up the
human mind exist in a wholly passive way, where, on the basis of sensory
contact with the outside world, pull themselves into bundles of similarity,

6. PHOEBE V. MOORE, THE QUANTIFIED SELF IN PRECARTY: WORK, TECHNOLOGY AND WHAT
COUNTS 1 (2017).
7. Thomas Hobbes, Chapter V. Of Reason and Science, in Leviathan, 29, 29-35 (Thomas Hobbs
ed., 1660), available at [Link]
murdvee/EconPsy/6/Hobbes_Thomas_1660_The_Leviathan.pdf.
2019] THE MIRROR FOR (ARTIFICIAL) INTELLIGENCE 51

borders, or cause and effect, similar to the ways in which neural networks
later were expected to behave.
Philosophers, social theorists, and cognitive and behavioral
psychologists worked on many theses relating to the human mind and what
defines consciousness, but only one significant body of research which
predates AI, explores the specific aspects of human thought with continuous
relation to machines: cybernetics. Although they were working sometimes in
direct parallel with AI researchers in the mid-1950s and beyond, cybernetics
researchers dedicated far more time and energy to considerations about
consciousness, thought, communication, and cognitive science than their AI
colleagues. There was cross-fertilization in their projects of course. This
section identifies some key points of thought during this period, to identify
how human and machinic intelligence were portrayed.
Norbert Wiener, one of the key forefathers of cybernetics work,
cogently noted in 1948 in Cybernetics, or, Control and Communication in
the Animal andthe Machine that "the thought of every age is its technique," 8
where there are continuous entanglements between human thought and
machinic invention. First published in 1950, in the second version of the text
The Human Use of Human Beings: Cybernetics and Society published in
1954, Wiener outlines "Cybernetics in History." Wiener was influenced by
Descartes; the British empirical philosophers; and Pavlov; and he highlights
their considerations of the workings of the human mind, to ultimately
compare the capacity of machines for mimicry. This mathematician was the
first to use the term "cybernetics," from the Greek word kubernetes, which
means "steersman," and the Greek word from which the word "governor"
later formed. The definition of cybernetics is tied to a theory of
communication, wherein Wiener theorizes that:
... society can only be understood through a study of the messages and
the communication facilities which belong to it; and that in the future
development of these messages and communication facilities, messages
between man and machines, between machines and man, and between
9
machines and machine, are destined to play an ever-increasing part.
Cybernetics researchers' purpose was to develop techniques and a
language to not only give humans the tools to "attack the problem of control
and communication in general," but also to "find the proper repertory of ideas
and techniques to classify their particular manifestations under concepts."10
The idea, ultimately, was to figure out how to give perfect orders to machine
and then to train machines to give orders to other machines (today's "Internet

8. NORBERT WIENER, CYBERNETICS, OR, CONTROL AND COMMUNICATION IN THE ANIMAL AND
THE MACHINE (2d. ed., 1961).
9. NORBERT WIENER, THE HUMAN USE OF HUMAN BEINGS: CYBERNETICS AND SOCIETY 1, 16 (2d.
ed., 1954).
10. Id. at 17.
52 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47

of Things"), on the basis of understanding how communication works as a

control mechanism. So, this way of identifying intelligence is through
communication itself, whereas we as humans are intelligent because we can
control each other via communication (see: basic management theory and lab
our process theories of control + consent).
In The Origins and History of Consciousness,another key cyberneticist
Erich Neumann explores mythological theories of foreknowledge and pre-
dualist consciousness. An ancient Egyptian symbol is described with the
following text: "Draco interfecit se ipsum, maritat se ipsum, impraegnat se
ipsum," which means "it slays, weds, and impregnates itself. It is man and
woman, begetting and conceiving, devouring and giving birth, active and
passive, above and below, at once." This symbolizes a one-ness, a
1
"wholeness, unity, nondifferentiation, and the absence of opposites.""
Children are treated with great respect in many primitive societies because
much of their behavior reflects a prenatal phase, where the 'feeble ego' is still
determined by an ancestral state.12 The myth of foreknowledge reflects the
human belief that all knowing is somehow linked to "memory." 1 3 Neumann's
beautifully written prose outlines several primitive, mythological, and
religious dispossessions to identify the history of consciousness (and beliefs
around unconsciousness), to show that "the act of becoming conscious
consists in the concentric grouping of symbols around the object, all
circumscribing and describing the unknown from many sides."1 4 It is difficult
to imagine how machines could mimic such fantastic delineations of the
human experience, but Neumann's work is central to the emerging
discussions around what separates humans from machines. In Understanding
Understanding:Essays on Cybernetics and Cognition, Heinz von Foerster
considers the feasibility of the existence of self-organizing systems and
focusses on memory, which is as important today for computational systems
as processing power. The book begins with the thesis that, in fact, self-
organizing systems cannot exist. Nonetheless, this cyberneticist deliberates
extensively on the thesis of the conditions and premise upon which a self-
organizing system could exist, proposing that:
By a self-organising system, I mean that part of a system that eats energy
and order from its environment;
There is a reality of the environment in a sense suggested by the
acceptance of the principle of relativity;
15
The environment has structure.

11. ERICH NEUMANN, THE ORIGINS AND HISTORY OF CONSCIOUSNESS 1, 599 (2015).
12. Id. at 583.
13. Id at 809.
14. Id. at 530.
15. HEINZ VON FOERSTER, UNDERSTANDING UNDERSTANDING: ESSAYS ON CYBERNETICS AND
COGNITION 1, 6 (2003).
2019] THE MIRROR FOR (ARTIFICIAL) INTELLIGENCE 53

Thus, there must be context for existence and consciousness, which

differentiates "semantics" from "syntax" in human/machinic consciousness.
Cyberneticists looked carefully at the depiction of human intelligence
and consciousness as tied to the tendency toward entropy. Wiener indicates
that "a machine" is "a device which locally and temporarily seems to resist
the general tendency for the increase of entropy (italics added by current
author). By its ability to make decisions it can produce around it a local zone
6
of organization in a world whose general tendency is to run down."'
Machines, then, were seen to have a distinct advantage over humans. These
ideas transferred into AI research, to which we now turn.

II. THE INTELLIGENCE OF MACHINES, BY HUMANS

The history of actually-termed "artificial intelligence" begins at an

academic conference in 1956, which were really a series of workshops led by
an Assistant Professor named John McCarthy, who worked with Marvin
Minsky of Harvard, Nathan Rochester of IBM and Claude Shannon (who was
also involved with the cybernetics community) of Bell Telephone
Laboratories to see whether they could make "a machine behave in ways that
would be called intelligent if humans were so behaving." McCarthy used the
term "artificial intelligence" to differentiate it from "cybernetics" and soon,
despite, or perhaps because of its overlaps in research questions, the field of
AI took precedence to cybernetics altogether.
The subsequent so-called Symbolic Approach to AI involved attempting
to mimic the logical processes of the brain, which was an era of investigation
that was later called GOFAI. In this first period of AI research, physical
symbol systems (PSS), also called formal systems, were envisaged by Allen
Newell and Herbert A. Simon, two of the "godfathers" of AI. These systems
are those which "have the necessary and sufficient means for general
intelligent action." 17 PSS relied on rationalist fundamentals and logical
assumptions in the tradition of Descartes, Hobbes, Leibniz, Kant and Hume,
where one must have abstracted a theory based on invariant features in
specific situations to deal with a domain. The claim that the PSS have the
means for general intelligent action is based on an understanding of how
humans think, i.e., that we only carry out symbol manipulation. Indeed, if that
is right, and humans only carry out symbolic manipulation and that equates
with intelligence, then machines can of course be intelligent.
The term AI itself flourished in popularity and in this early phase of AI
research researchers held that human-readable representations of problems,
in the form of symbols, should inform how all AI research should be

16. Wiener, supra note 8, at 34.

17. Allen Newell & Herbert Simon, 19 Computer Science as Empirical Inquiry: Symbols and
Search, COMMUNICATIONS OF THE ACM 113, 113-26 (Mar. 1976).
54 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47

conducted. This form of AI involved expert systems that reflected production

rules that are altered by human deduction in the context of emergent errors,
the processes relying on an "if this, then that" type of formula: basically, a
flow chart. Machine intelligence required "making the appropriate
inferences" from their seemingly internal representations, where a PSS was
seen to posit this, as quoted, "necessary and sufficient means for general
intelligent action." There was a lot of criticism of PSS and of GOFAI in
general, but nevertheless, it is the basis of AI in the earliest considerations.
Critics held that there are problems with expecting machines to supposedly
represent the world exactly as humans do or argued that humans don't
"represent" the world at all, but in fact, "'are" the world. This reasoning sits
alongside a more recent idea that machines should be fully autonomous.
So-called symbolic and connectionist AI researchers did not agree on
what constituted the more important features of intelligence, but Haugeland,
who coined the term "good old fashioned artificial intelligence" (GOFAI),
describes intelligent beings as demonstrating the following characteristics:
firstly, our ability to deal with things intelligently, where intelligence is
present due to our capacity to think about them reasonably (including
subconscious thinking); and secondly, the capacity to think about things
reasonably, which amounts to a faculty for internal/automatic symbol
manipulation. It also becomes clear that memory and ability to process
thoughts and ideas and turn those ideas into analysis; the capability to make
choices rather than simple decisions; and the ability for empathy and
sentience, are necessary for intelligence to be manifest which is particularly
important as machines are ascribed more forms of intelligence such as those
argued in the second section, namely, collaborative capacities, prediction-
making ability and prescriptive positioning. Marcus Hutter, who designed a
theory of universal AI, argued that "the human mind ... is connected to
consciousness and identity which define who we are. . . intelligence is the
most distinct characteristic of the human mind .. .enables us to understand,
explore and considerably shape our world, including ourselves." 18 AI
research, Hutter indicates, reflects this sentiment, since the "grand goal of AI
is to develop systems that exhibit general intelligence on a human-level or
beyond." 19
The next important invention in the history of AI and the first major
challenge to GOFAI and the hopes that machines could be trained to
"represent" the world in the ways that humans do, was the invention of the
first artificial neural network. Frank Rosenblatt, a psychologist, is said to
have invented the first one in 1958, just two years after this first conference.

18. Marcus Hutter One Decade of Universal Artificial Intelligence, 4 THEORETICAL FOUNDATIONS
OF ARTIFICIAL GENERAL INTELLIGENCE 67, 67-88 (2012).
19. Id. at 1.
2019] THE MIRROR FOR (ARTIFICIAL) INTELLIGENCE 55

This neural network was called Perceptron. It modeled the ways that human
brains process visual data and learn how to recognize objects at the same time
as picking similar cases: the first time this could be done in parallel (which
differentiates it from the PSS). Neural networks allow computers to make
decisions based on information within domains that is collected from various
silos which result in conclusions from either supervised or non-supervised
processes. That is what sets neural networks apart from GOFAI and brings
us into the "cognitive" period of AI research, where scientists moved away
from comparing machines directly to humans and expecting them to think in
the ways that humans do. Indeed, the invention of neural networks raised a
number of philosophical questions about how a theory of domain is
formulated.
About a decade later, in 1966, Joseph Weizenbaum, a German American
MIT computer scientist who is also considered one of the forefathers of AI,
invented the predecessor for today's Chatbots, naming this computer
program "ELIZA," after the ingenue in George Bernard Shaw's Pygmalion.
Pygmalion is a character in Greek mythology who develops a love interest in
his own sculpture ,which comes to life. This seemed an appropriate name for
this chatbot given "her" quickly observed capacity to induce emotion from
those speaking to her. Weizenbaum took this response from humans seriously
and was sincerely surprised when people responded so genuinely to Eliza.
Indeed, Weizenbaum was very skeptical about the integration of computers
into society and saw the dark sides that it introduced.20 Alan Turing would
have probably found this finding quite interesting, given the experiments he
pursued, where humans project their own assumptions onto machines.
Indeed, Alan Turing is one of the only theorists during the early discussions
of AI to thoroughly interrogate the question "can machines think?."2 1 His
illumination of the words "machine" and "think," in the first paragraph of
"Computing Machinery and Intelligence" 22 asks a similar question to mine,
i.e., "what is intelligence" when we talk about AI, and so on. In designing the
Imitation Game, Turing was also careful to state that "no engineer or chemist
claims to have been able to produce material which is indistinguishable from
the human skin.. . we do not wish to penalise the machine for its inability to
shine in beauty competitions, nor to penalise a man for losing in a race against
an aeroplane." 23 It is interesting that this particularly important researcher,
who was himself punished for his homosexuality, alludes that the difference
between the human and machine is very much corporeal, and by extension,
human intelligence cannot be separated ontologically from our physicality.
Philosophically, this could be linked to later new materialist research.

20. Weizenbaum, supra note 2.

21. Alan M. Turing, Computing Machinery and Intelligence, 49 MIND 433-460 (1950).
22. Id.
23. Id. at 434.
56 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47

The next section reveals the practical sides of the integration of AI into
our everyday lives in the workplace and the risks that emerge when
intelligence is assumed from the tools and applications implemented.

III. THE INTELLIGENCE OF MACHINES, BY MACHINES

AI has reached a stage now where it is expected to somehow transform

societies forever, with contrasting visions from the late Stephen Hawking
who declared that AI was the "worst event in the history of our civilization"
and Elon Musk who noted that "AI is a fundamental risk to the existence of
human civilization," to enthusiastic predictions about its ability to facilitate
superpower status for countries. Advanced countries are currently very active
in integrating AI across industries and as many spheres of society as possible,
allocating significant pots of funding in the order of billions to research and
development in AI, with the United States in the lead, closely followed by
China and Israel. 24 AI is predicted to provide a 26 % boost to gross domestic
product (GDP) by 2030 in China. North America is predicted to see a 14.5%
boost, 25 and some predictions indicate that AI will create as many jobs as it
eliminates. 26 In most cases, high-level governmental and organizational
reports are predicting that AI will improve productivity. However,
discussions seem to not link productivity with workers and workplaces,
despite this, it is precisely where AI is being introduced in order to improve
productivity-in the factories, in gig work, and in office contexts.
This section identifies explicitly the ways that AI is being ascribed with
forms of intelligence that are at points being prioritized over human
intelligence and thus used for workplace decision-making. HR decision
making, or people analytics and filmed interviews, is where humans expect
machines to show predictive intelligence. Where cobots are applied, the
implementers assume that these machines are both assistive and collaborative
in their intelligence capacities. Call center chatbots are given assistive
intelligence capacities. Wearable technologies in factories likewise are
assistive. Algorithms in platform work are predictive and proscriptive, where
clients and customers are matched better than humans could do. Within all of
these categories, machines are given calculative intelligence beyond and
superior to any level of human intelligence, which fits well with the capitalist
schema of abstracting concrete value of labor, where "assistance,"

24. LAURA DELPONTE, EUROPEAN ARTIFICIAL INTELLIGENCE LEADERSHIP, THE PATH FOR AN
INTEGRATED VISION 1 (European Parliament 2018).
25. PWC, ARTIFICIAL INTELLIGENCE IN HR: A NO-BRAINER (2018), available at
[Link]
26. PWC, AI WILL CREATE AS MANY JOBS AS IT DISPLACES BY BOOSTING ECONOMIC GROWTH
(July 17, 2018), available at [Link]
[Link].
2019] THE MIRROR FOR (ARTIFICIAL) INTELLIGENCE 57

"collaboration" and so on, are simultaneously enhanced with metrics-

producing activities against which human workplace performance are
judged.

A. PredictiveIntelligence

In the area of HR business execution, AI integration is called "people

analytics," defined broadly as the use of big data and digital tools to
"measure, report and understand employee performance, aspects of
workforce planning, talent management and operational management." 2 7
Computerization, data-gathering, and monitoring tools allow organizations
to conduct "real-time analytics at the point of need in the business process ..

.
[and allow] for a deeper understanding of issues and actionable insights for
the business." 28 Prediction machine algorithms applied for these processes
often reside in a "black box," 29 and people do not fully understand how they
work, but, even so, computer programs are given the authority to make
"prediction[s] by exception." 30 "Prediction by exception" refers to processes
whereby computers deal with large data sets and are able to make reliable
predictions based on routine and regular data and also to spot outliers and
even send notifications "telling" the user that checks should be made or that
human assistance or intervention should be provided.
Also called "human analytics," "talent analytics," and "human resource
analytics," in an era of "strategic HR," this application of AI-enabled tools is
defined broadly as the use of individualized data about people to help
management and HR professionals make decisions about recruitment (i.e.,
who to hire, for performance appraisals and promotion considerations) to
identify when people are likely to leave their jobs and to select future leaders.
People analytics are also used to look for patterns across workers' data, which
can help to spot trends in attendance, staff morale, and health issues at the
organizational level.
"People problems" are also called "people risks," 31 and are divided into
seven dimensions in a Chartered Institute for Personnel Development (CIPD)
report by Houghton and Green as:
talent management,

27. LAURENCE COLLINS, DAVID FINEMAN & AKIO TSHUCHICA, PEOPLE ANALYTICS:
RECALCULATING THE ROUTE (DELOITTE INSIGHTS, 2017), available at
[Link]
28. Id.
29. FRANK PASQUALE, THE BLACK BOx SOCIETY: THE SECRET ALGORITHMS THAT CONTROL
MONEY AND INFORMATION 1 (2015).
30. AJAY AGARWAL, JOSHUA GANS & AVI GOLDFARB, PREDICTION MACHINES: THE SIMPLE
ECONOMICS OF ARTIFICIAL INTELLIGENCE 1 (2018).
31. EDWARD HOUGHTON & MELANIE GREEN, PEOPLE ANALYTICS: DRIVING BUSINESS
PERFORMANCE WITH PEOPLE DATA (Chartered Institute for Personnel Development 2018), available at
[Link]
58 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47

health and safety,

employee ethics,
diversity and equality,
employee relations,
business continuity, and
reputational risk.
But perhaps people are not the only "problem." Based on the original
definition of AI, in which machinic intelligence will be demonstrated when
it behaves as a human would: if humans are discriminating and biased, then
we should not be surprised when AI provides biased answers. In other words,
machine learning operates only on the data that it is fed, and if that data
reveals past discriminatory hiring and firing practices, then the results of the
algorithmic process are likely to also be discriminatory. If the information
gathered about workers is not buffered with qualitative information about
individuals' life experiences and consultation with workers, unfair
judgements could be made. Cherry has shown that management's "search for
new pools of quantitative data are correlated with business and employment
success" and data is used to "make workplace decisions and to replace
subjective decision-making by managers." 32
Not all people analytics have to be, strictly speaking, AI. However,
programs' intelligent responses to algorithmic equations allow machine
learning, which generates predictions and asks associated questions that
emerge without human intervention except at the data input phase. Big data
has been seen as a lucrative growth area for some years, whereby the
collection of information about everything, all the time, has been an attractive
investment. Now, the big data era is paying off in HR circles, because the
extensive pools of data now available can be used to train algorithms to form
analyses and make predictions about workers' behavior via machine learning
and thereby assist management decision-making. On the basis of the patterns
identified, AI enables an algorithm to produce solutions and responses to
enquiries about patterns across data much more quickly than people could.
Machine learning responses are often unlike those that a human alone would,
or perhaps even could, generate. Data about workers can be gathered from
various sources both in and outside of the workplace, such as number of
keyboard clicks, information from social media, number of and content of
telephone calls, websites visited, physical presence, locations visited outside
the workplace through GPS (global positioning system) tracking, movement
around the office, content of emails, and even tone of voice and bodily
33
movements in sociometrics.

32. Miriam A. Cherry, People analytics and invisible labor, 61 ST. LoUIs U. L.J.1, 1-16 (2016).
33. Moore, supra note 6, at 146.
2019] THE MIRROR FOR (ARTIFICIAL) INTELLIGENCE 59

AI-enhanced HR practices can help managers obtain seemingly

objective wisdom about people even before they hire them, as long as
management has access to data about prospective workers, which has
significant implications for tailoring worker protection and preventing
occupational safety and health risks at the individual level. Indeed,
algorithmic decision-making in people analytics could be used to support
workforces by aligning employee performance feedback and performance
pay -and workforce costs -with business strategy and support for specific
workers. Workers should be personally empowered by having access to new
forms of data that help them to identify areas for improvement, that stimulate
personal development and that achieve higher engagement.
However, if processes of algorithmic decision-making in people
analytics do not involve human intervention and ethical consideration, this
machinic HR tool could expose workers to heightened structural, physical,
and psychosocial risks and stressors. How can workers be sure that decisions
are being made fairly, accurately, and honestly, if they do not have access to
the data that their employer holds and uses? The risks of stress and anxiety
arise if workers feel that decisions are being made based upon numbers and
data, where the machinic intelligence takes precedence over that of the
human-particularly in cases where they have no say in the processes
selected to gather data about themselves, nor access to the data itself.
This is particularly worrying if people analytics data leads to workplace
restructuring, job replacement, job description changes, and the like. People
analytics are likely to increase workers' stress if data are used in appraisals
and performance management without due diligence in process and
implementation, leading to questions about micro-management and workers
feeling "spied on." If workers know that their data are being read for talent
spotting or for deciding possible layoffs, they may feel pressurized into
improving their performance and begin to overwork, posing occupational
safety and health risks. Another risk arises with liability, in which companies'
claims about predictive capacities may later be queried for accuracy or
personnel departments be held accountable for discrimination.
Another form of people analytics involves filming job interviews. This
practice is carried out by organizations such as Nike, Unilever, and Atlantic
Public Schools. These companies are using products that allow employers to
interview candidates on camera, in which AI is used to judge both verbal and
nonverbal cues. One such product is made by a group called HireVue and is
used by over 600 companies. The aim is to reduce bias that can arise if, for
example, an interviewee's energy levels are low or if the hiring manager has
more affinity for an interviewee based on similar age, race, and related
demographics. However, there is evidence that preferences from previous
hiring managers are reflected in hiring, and heterosexual white men are, a
report by Business Insider reveals, the hiring preference, other things being
60 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47

equal. 34 If the data provided to an algorithm reflect the dominant bias over
time, then it may score someone with "in group" facial expressions higher
and give a lower rating to other cues tied to sexual orientation, age, and
gender that do not resemble a white male. Therefore, the expected
intelligence of the machine has in no way excelled against human
"intelligence," which for the activities of recruitment, hiring, firing, and
talent management in HR has been repeatedly demonstrated to be
discriminatory and biased.

B. CollaborativeandAssistive Intelligence: Cobots

Robots were at first built to carry out simple tasks, they are increasingly
3
enhanced with AI capabilities and are being "built to think, using A." s
Cobot intelligence is seen to be both assistive and collaborative, in the
following ways. Incredibly large orange robot arms have replaced workers
in factories in many parts of the world, where building car parts and
assembling cars have replaced conveyor belts lined with humans. AI should
not be confused with automation, however. Automation in its pure sense
involves, for example, the explicit replacement of a human's arm with a robot
arm. Lower skilled, manual work has historically been most at risk and is still
at high risk of automation. Now, automation can be augmented with
autonomous machine behavior or "thinking." So, the AI dimension of
automation reflects where workers' brains, as well as their limbs, may no
longer be needed.
Cobots are integrated into factories and warehouses where they work
alongside people in a collaborative way. They assist with an increasing range
of tasks, rather than necessarily automating entire jobs. Amazon has 100,000
AI-augmented cobots, which has shortened the time taken to train workers to
less than two days. Airbus and Nissan are using cobots to speed up production
and increase efficiency. AI-permitted pattern and voice recognition and
machine vision mean that not only are unskilled jobs at risk of replacement,
but now a range of nonroutine and nonrepetitive jobs can be carried out by
cobots and other applications and tools.

34. Richard Feloni I Tried the Software that Uses AI to Scan Job Applicants for Companies Like
Goldman Sachs and Unilever Before Meeting Them, and It's Not as Creepy as It Sounds,BUSINESS
INSIDER UK (Aug. 23, 2017), available at [Link]
AI-to-scan-job-applicants-for-companies-like-Goldman-Sachs-and-Unilever-before-meeting-them-and-
its-not-as-creepy-as-it-sounds/articleshow/[Link].
35. Jari Kaivo-oja, A Review on the Future of Work: Robotics (European Agency for Safety and
Health at work, Discussion Paper, Nov. 11, 2015), [Link]
publications/publications/future-work-robotics/view.
2019] THE MIRROR FOR (ARTIFICIAL) INTELLIGENCE 61

In that light, AI-enhanced automation enables many more aspects of

work to be done by computers and other machines. 36 As a recent Netherlands
Organisation for Applied Scientific Research (TNO) report states, there are
three types of risks in human-cobot-environment interactions:
(a) robot-human collision risks, in which machine learning can lead to
unpredictable robot behavior;
(b) security risks, in which robots' internet links can affect the integrity
of software programming, leading to vulnerabilities in security; and
(c) environmental risks, in which sensor degradation and unexpected
human action in unstructured environments can lead to risks to the
environment. 37
Cobots can reduce some physical risks, as they allow AI systems to
carry out other types of mundane and routine service tasks in factories, which
historically create stress, overwork, musculoskeletal disorders, and even
boredom as a result of repetitive work. However, AI-augmented robots in
factories and warehouses can create stress and a range of serious problems if
they are not implemented appropriately. Indeed, one UK-based trade unionist
indicated that digitalization, automation and algorithmic management when
"used in combination .. are toxic and are designed to strip millions of folks
of basic rights" (interview with Maggie Dewhurst of the Independent
Workers Union of Great Britain, 2017). Workplace issues include
psychosocial risk factors if people are driven to work at a cobot's pace (rather
than the cobot working at a person's pace) and the very real risk of robots'
unpredictable behavior, where collisions between cobots and people have no
obvious solution in terms of regulation and liability (i.e. who is responsible,
the manufacturer, designer, manager, or worker, if a collision occurs?).
Another cobot-related case of machine-human interaction creating new
working conditions and possible risks in assigning superior intelligence to
machines is when one person is assigned to "look after" one machine and is
sent notifications and status updates about machines on a personal device
such as a smartphone or a personal laptop. This can lead to risks of overwork,
whereby workers feel compelled to take note of notifications in out-of-work
hours and their work-life balance is disrupted. The requirement to constantly
respond to factory notifications is a clear demonstration of the prioritization
of machinic intelligence over humans, at least where a human is not fully
focused on productivity and efficiency or may desire work/life balance that
does not require notification responses over weekends!

36. Carl Frey & Michael Osborne, The Future of Employment: How Susceptible are Jobs to
Computerisation? (UNIVERSITY OF OXFORD, OXFORD MARTIN SCHOOL, Sept. 1, 2013),
[Link]
37. Anne Jansen, Dolf van der Beek, Anita Cremers, Mark Neerincx
& Johan van Middelaar, Emergent Risks to Workplace Safety; Working in the Same Space as a Cobot.
(Report for the Ministry of Social Affairs and Employment, 2018), 17-18.
62 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47

Sensors, software, and connectivity can be faulty and unstable, and all
vulnerabilities raise questions about who is legally responsible for any
damage that emerges. Is it a cobot's fault if it runs into a worker, the worker's
fault, the company who manufactured the cobot originally, or the company
that employs the worker and integrates the cobot? Again, there is no
immediate answer to questions around liability, ethics, nor intelligence as we
explicitly attribute intelligence to machines in the current world of work.
It is thought that cobots may someday have the competences to reason,
a final point in the development of machinic intelligence perhaps, and
therefore, it is reported by the TNO, must make humans feel safe. To achieve
this, cobots will have to demonstrate perception of objects versus humans
and the ability to predict collisions, adapt behavior appropriately, and
demonstrate sufficient memory to facilitate machine learning and decision-
making autonomy. 38

C. Assistive Intelligence: Chatbots

Chatbots are another AI-enhanced tool that can deal with a high
percentage of basic customer service queries, freeing up humans working in
call centers to deal with more complex questions. Chatbots work alongside
people, but in the back end of systems where they are used to deal with
customer queries over the phone using natural language processing. Dixons
Carphone uses a conversational chatbot now named Cami that can respond
to first-level consumer questions on the Currys website and through
Facebook Messenger. Insurance company Nuance launched a chatbot named
Nina to respond to questions and access documentation in 2017. Morgan
Stanley has provided 16,000 financial advisers with machine learning
algorithms to automate routine tasks.
Call center workers already face extensive risks because of the nature of
the work, which is repetitive, demanding, and subject to high rates of micro-
39
surveillance and extreme forms of measurement. An increasing number of
activities are already recorded and measured in call centers. Words used in
emails or stated vocally can be data-mined to determine workers' moods, a
process called 'sentiment analysis'. Facial expressions likewise can be
analyzed to spot signs of fatigue and moods that could be used to make
judgements and thus lower the risks of overwork. But chatbots, while
designed to be assistive machines, still pose psychosocial risks around fears
of job loss and replacement. Workers should be trained to understand the role
and function of workplace bots and to know what their collaborative and
assistive contributions are.

38. Id. at 16
39. JAMIE WOODCOCK, WORKING THE PHONES: CONTROL AND RESISTANCE IN CALL CENTERS 1
(2016).
2019] THE MIRROR FOR (ARTIFICIAL) INTELLIGENCE 63

D. Assistive Intelligence: Wearables

Wearable self-tracking devices are increasingly seen in workplaces. The

market for wearable devices in the industrial and healthcare sectors has been
predicted to grow from USD $21 million in 2013 to USD $9.2 billion by
2020.40 Between 2014 and 2019, 13 million fitness devices are predicted to
be incorporated into workplaces. This is already happening in warehouses
and factories where GPS, radio-frequency identification, and now haptic-
sensing armbands, such as those patented by Amazon in 2018, have replaced
the use of clipboards and pencils.
Wearable technologies with AI augmentation are seen to offer a level of
intelligence that again, excels over human intelligence in assistive
capabilities. One new feature of automation and Industry 4.0 processes in
which AI-enhanced automation is under way, is the practice of training
workers to perform productive tasks within lot size manufacturing. This
process involves workers being provided with glasses with screens and
virtual reality functionality, such as HoloLenses and Google Glasses, or
computer tablets on stands within the production line, which are used to carry
out on-the-spot tasks on said lines. The assembly line model, in which a
worker carries out one repeated, specific task for several hours at a time, has
not disappeared completely, but the lot size method has differed. Used in
agile manufacturing strategies, this method involves smaller orders made
within specific time parameters, rather than constant bulk production that
does not involve guaranteed customers.
Visual on-the-spot training enabled by a HoloLens screen or tablet
means that workers are required to carry out new tasks that are learned on-
the-spot, in augmented reality situations, where tasks are required
instantaneously and carried out only for the time required to manufacture the
specific order a factory receives. While at first glance these assistance
systems may appearto provide increased autonomy, personal responsibility,
and self-development, this is not necessarily the case.4 1
The use of on-the-spot training devices, worn or otherwise, means that
workers need less pre-existing knowledge or training, because they carry out
the work case by case. The risk of work intensification thus arises, as head-
mounted displays or tablet computers become akin to live instructors for
unskilled workers. Furthermore, workers do not learn long-term skills,
because they are required to perform on-the-spot, modular activities in
custom assembly processes needed to build tailor-made items at various

40. David Nield, In Corporate Wellness Programs, Wearables Take a Step Forward,FORTUNE (Apr.
15, 2014), available at [Link]
step-forward/.
41. Florian Butollo, Ulrich Jiirgens & Martin Krzywdzinski, From Lean Production to Industrie 4.0:
More Autonomy for Employees? (wissenshanftszentrum Berlin far Socialforschung (WZB) Discussion
Paper SP 111 2018-303, 2018), available at [Link]
64 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47

scales. While this is good for the company's efficiency in production, lot size
methods have led to significant risks of deskilling workers, because skilled
labor is needed only to design the on-the-spot training programs used by
those workers who no longer need to specialize. The wearable technologies
used are assistive, but also lead to explicit risks that can emerge because of
the lack of communication with e.g. specialists. If workers are not able to
comprehend the complexity of the new technology quickly enough,
particularly if they are also not trained to prepare for any hazards arising, they
are at risk of being replaced or disciplined, where the machine again is seen
to hold all intelligence in the workplace.
Next, we turn to another arena in which machines are being ascribed
intelligence: namely, in "gig work" environments.

E. Prescriptiveintelligence: Algorithms

"Gig work" is obtained by using online platforms made available by

companies such as Uber, Upwork, or Amazon Mechanical Turk (AMT). The
work can be performed online, obtained and carried out on computers in
homes, libraries, and cafes, for example, and includes translation and design
work; or offline, i.e. obtained online but carried out offline, such as taxi
driving or cleaning work.
Not all algorithms utilize AI, but the data produced by client-worker
matching services and customer assessment of platform workers provide data
that train profiles that then result in overall higher or lower scores that then
lead, for example, clients to select specific people for work over others. So,
algorithms are used to match clients with workers in online gig work (also
called microwork) and thus are expected to hold prescriptive intelligence.
One platform called BoonTech uses IBM Watson AI Personality Insights to
match clients and online gig workers, such as those gaining contracts using
AMT and Upwork. Issues of discrimination have emerged that are related to
women's domestic responsibilities when carrying out online gig work at
home, such as reproductive and caring activities in a traditional context.
In gig work, workers have been forced to register as self-employed
workers, losing out on the basic rights that formal workers enjoy, such as
guaranteed hours, sick and holiday pay, and the right to join a union. Gig
workers' online reputations are very important because a good reputation is
the way to gain more work. Digitalized customer and client ratings and
reviews are key to developing a good reputation, and these ratings determine
how much work gig workers obtain. Algorithms learn from customer
rankings and quantity of tasks accepted, which produce specific types of
profiles for workers that are usually publicly available. Customer rankings
are deaf and blind to the consideration of workers' physical health, care and
domestic work responsibilities, and circumstances outside workers' control
2019] THE MIRROR FOR (ARTIFICIAL) INTELLIGENCE 65

that might affect their performance, leading to further risks, where people
feel forced to accept more work than is healthy, or risk work exclusion.
Customer satisfaction rankings and number of jobs accepted can be used to
"deactivate" taxi drivers' use of the platform, as is done by Uber, despite the
paradox and fiction that algorithms are absent of "human bias." 42
Overall, while there are benefits to integrating AI into gig work,
including driver identity protection and allowing flexible work hours, good
for people's life and work choices, these same benefits can result in rising
risks, such as the case of the DiDi driver and the case of a double burden of
work for women online workers. Worker protections are generally already
scarce in these working environments and the risks are many and involve low
pay and long hours,43 endemic lack of training,44 and a high level of
insecurity. 45 Nonetheless, the almost blind faith in the capability of the
algorithm and its prescriptive intelligence witnessed in newly designated
platform business models and management methods 46 demonstrates the
endemic trend to look to machines for intelligence, perhaps forgetting the
word that directly precedes: "artificial."

IV. TOWARDS A CONCLUSION

The difference with AI and other forms of technological development

and invention for workplace usage is that because of the intelligence
projected onto autonomous machines, they are increasingly treated as
decision makers and management tools themselves, due to their seemingly
superior capacity to calculate and measure. Where many recent reports on AI
try to deal with the questions of "what can be done" or "how can AI be
implemented ethically," the issue is greater. A move to a reliance on machinic
calculation for workplace-intelligent decision-making actually introduces
extensive problems for any discussion of "ethics" in AI implementation and
use, at all.

42. Frey et. al., supra note 36.

43. Janine Berg, Income security in the on-demand economy: Findings and policy lessons from a
survey of crowdworkers, 74 International Labour Office Conditions of work and Employment Series
(2016), available at [Link]
travail/documents/publication/wcms_479693.pdf.
44. CHARTERED INSTITUTE FOR PERSONNEL DEVELOPMENT (CIPD), TO GIG OR NOT TO GIG?
STORIES FROM THE MODERN ECONOMY (Mar. 2017), available at [Link]
or-not-to-gig_2017-stories-from-the-modern-economy_tem [Link].
45. MATTHEW TAYLOR, GOOD WORK: THE TAYLOR REVIEW OF MODERN wORKING PRACTICES
(Department for Business, Energy and Industrial Strategy, July, 2017), available at
[Link]
1/[Link].
46. See Phoebe Moore & Simon Joyce, Black Box or Hidden Abode? The Expansion and Exposure
of Platform Work Managerialism, REVIEW OF INTERNATIONAL POLITICAL ECONOMY (Aug. 7, 2019);
JEREMIAS PRASSL. HUMANS AS SERVICE: THE PROMISE AND PERILS OF wORK IN THE GIG ECONOMY 1
(2018).
66 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47

In Locke's An Essay ConcerningHuman Understanding,the empiricist

philosopher wrote that ethics can be defined as "the seeking out [of] those
Rules, and Measures of humane Actions, which lead to Happiness, and the
Means to practice them" (Essay, [Link].3 1689).4 This is, of course, just one
quote, by one ethics philosopher, but it is worth noting that the seeking out
of and setting such rules, as are the parameters for ethics depiction, has only
been carried out and conducted, so far, by humans. When we introduce the
machine as an agent for rule setting, as AI does, the entire concept of ethics
falls under scrutiny. Rather than talking about how to implement AI without
the risk of death, business collapse, or legal battles-which are, effectively,
the underlying concerns that drive AI ethics discussions today-it would
make sense to rewind the discussions and focus on the question: why
implement AI at all? Will the introduction of AI into various institutions and
workplaces across society truly lead to prosperous, thriving societies, as is
being touted? Or, will it deplete material conditions for workers and promote
a kind of intelligence that is not oriented toward e.g. a thriving welfare state,
good working conditions, or qualitative experiences of work and life?
In China, the government has started to give people individual citizen
scores, or an economic and personal reputation scoring, which will look at
citizens' rent payments, credit rankings, phone use, and so on. It will be used
to determine conditions for obtaining loans, jobs and travel visas. Soon,
people analytics could be used to give people "worker scores," to be used for
decision-making in appraisals, which would introduce all sorts of questions
about privacy and surveillance as machinic reasoning is again prioritized over
human. While machines can perform some tasks faster than humans in
calculation and statistical generation, they cannot understand context. While
machines have more memory and processing power than ever before, which
is how they can participate in machine learning, they lack empathy and full
historical knowledge. The problem is that machines do not and cannot see
the qualitative aspects of life, nor the surrounding contexts. Cathy O'Neil,
author of Weapons of Math Destruction:How Big DataIncreasesInequality
and Threatens Democracy, made an insightful observation in a recent
interview with the author. While watching Deliveroo riders hurtle past her in
the rain, Dr. O'Neil considered the platforms directing the riders' work,
which operate on the basis of efficiency and speed, and thus instigate riders
to cycle in unsafe weather conditions. This clearly puts riders' very lives at
risk. Dr. O'Neil calls algorithms "toy models of the universe," because these
seemingly all-knowing entities actually only know what we tell them, and
thus have major blind spots.

47. John Locke. An Essay Concerning Human Understanding, in 1 THE wORKS OF JOHN LOCKE
1639 (12th ed., 1689), availableat [Link]
2019] THE MIRROR FOR (ARTIFICIAL) iNTELLIGENCE 67

If it is accepted that machines hold the same competences as humans, or

even better competences than us, what will be left for us? Are all machines
intelligent, or only some? How is intelligence defined? Driving the thinking
for this article lies the following questions: why do we want machines to
behave "intelligently"? what do we mean by "intelligence"? and what is at
risk for workers, when we ask machines to behave in our own image? Can
there be an ethical use for AI, given the complexity of rulemaking, when
something besides an intelligent human mind is expected to make rules?
Where will the final say in intelligence lie? Why do we want machines to
behave as we do, given that evidence already shows that machines can only
learn as much as already exists in the data that trains it, and if the data reflects
humans' discriminatory behavior, then the algorithms, almost necessarily,
will demonstrate or promote discrimination. The mythical invention of E. M.
48
Forster's all-encompassing machine in his classical science fiction story
was not, of course, subject to a range of ethical and moral review panels
before all of humanity began to live within it under the Earth's crust. As we
enter a new era of AI, it will remain important to recall the tension points in
positioning technologies in places of power in workplaces and maintain a
"human in command" approach 49 to rolling out any new technologies into
workplaces. As this Special Issue acknowledges, human responses to this
trend should involve careful regulation, where human intelligence takes
precedence, as the machine becomes increasingly evident in our working
lives.

48. E. M. FORSTER, THE MACHINE STOPS 1 (2011).

49. Valerio De Stefano, Negotiating the Algorithm: Automation, ArtificialIntelligence and Labour
Protection(International Labour Organization Working Paper No. 246, 2018).
68 COMP. LABOR LAW & POL'Y JOURNAL [Vol. 41:47