|
A peer-reviewed electronic journal published by
the Institute for Ethics and ISSN 1541-0099 25(1) – June 2015 |
Technological Unemployment, AI, and Workplace
Standardization: The Convergence Argument
Marc Saner
Department of Geography &
Institute for Science, Society and
Policy
University of Ottawa
and
Wendell Wallach
Interdisciplinary Center for Bioethics
Yale University
Journal of Evolution and Technology - Vol. 25 Issue 1
– June 2015 - pgs 74-80
Abstract
The current debate over technological unemployment sacrifices
significant analytic value because it is one-sided, limited in scope, and
sequential. We show that analyzing technological
innovations in parallel with apparently independent socio-economic innovations and trends offers important analytical
benefits. Our focus is on socio-economic innovations and trends that
standardize education, workplace requirements, and culture. A highly
standardized workplace is not only more suitable for international outsourcing;
it is also more suitable for machine labor. In this context, we identify five
specific research questions that would benefit from parallel analysis and
scenarios. We also introduce the concepts functional equivalency and functional
singularity (in juxtaposition to technological singularity) to provide
semantic tools that emphasize the importance of an integrated approach, capable
of tracking and analyzing two interacting and potentially converging trends.
Introduction
The current
debate over technological unemployment1 sacrifices significant
analytic value because it is one-sided, limited in scope, and sequential. The
focus of the analysis is almost exclusively upon modeling the speed, potential
limitations, and social implications of technological
innovations in the form of artificial intelligence and robotics. Both
enthusiasts and critics, including commentators such as Bill Gates, Stephen
Hawking, Elon Musk, Nouriel Roubini, and Larry
Summers, have selected this perspective. The longer-term prospects identified
by this technological focus, such as superintelligence (Bostrom 2014) or a technological
singularity (Vinge 1993, Kurzweil 2005), are tantalizing and receive ample attention. They are,
however, also highly speculative and controversial.
Our concern
is that the focus on these particular technological trajectories naturally draws
attention toward the extreme endpoints, at the cost of distracting from the most
accessible policy options available here and now. We will show that analyzing technological innovations in parallel
with apparently independent socio-economic
innovations and trends offers a more complete perspective for addressing key
policy questions.
The
interactions between technological innovations and socio-economic innovations
are complex. Technological innovations can drive socio-economic innovations and
vice versa. Furthermore, there are a
number of different socio-technological trends that place a downward pressure
on job creation and wage growth. For example, people are living longer and many
retire later, thus freeing up fewer jobs for those entering the workforce. Our
focus here is on the socio-economic innovation of processes, standards, and
regulations that render workers and workplaces more vulnerable to displacement
by technology.
Standardized educational, workplace, and cultural
norms
We
acknowledge that software and robotic technology will create novel workplace
functions and that some of these will be more suitable for humans than for
machines. Our aim, thus, is not to forecast technological unemployment rates. Instead,
we want to emphasize the benefits of a more inclusive methodological research
lens.
This lens provides
a focus on the functional equivalence
between human and machine at the workplace. The equivalence, we argue, is not
achieved only by the greater technological capacity of software and robotic machines;
it is also achieved by rendering workers, jobs, and the
workplace more machine-like (more standardized and predictable). A high
degree of functional equivalence, no matter whether it is the result of
technological advances, workplace design, or both, will result in a greater
ease by which technological unemployment can progress. Once machine capacity
and workplace requirements become a perfect match, the equivalence has been
achieved and the displacement of human labor becomes possible.
Of course, there
are other issues such as price points, maintenance costs, adaptability, and so
forth that will determine whether the functional equivalence will indeed lead
to displacement in the labor market. If the rate of new equivalency events is
matched by the rate of the creation of novel workplace functions for humans,
then the issue is not job availability; it is, rather, training, retooling, and
adaptation. Arguably this has been the situation over the past two centuries as
steam engines, automobiles, and robots on assembly lines have replaced human
workers. Nevertheless, we ought to pay attention to how education and workplace
design interact with the likelihood of functional equivalency events. By doing
so we gain an important analytical tool for evaluating whether the adoption of
new technologies is actually contributing to any slowness in recovery or growth
in the unemployment rate.
Three
examples will illustrate socio-economic innovations and trends that facilitate
functional equivalence. The examples address education, the workplace, and
cultural norms, respectively:
1. Education
is becoming more standardized internationally and locally. An example is the Programme for International Student Assessment (PISA) carried out annually by the Organization for Economic Co-operation and
Development (OECD). This type of performance indicator has the effect that
countries adjust their curricula to improve their rankings which, in turn,
leads to greater uniformity in the key topics that are taught internationally. Standardized
university testing and the education standards enforced by professional
associations further this effect. The obvious benefit for employers
lies in the improved ability to accept students and workers trained elsewhere, while
students and employees gain greater international mobility. A research focus on
functional equivalency, however, would question whether the greater emphasis in
education toward measurable qualities will diminish the training of human
qualities such as creativity, care, communication, and touch that remain
harder to achieve by machines.
2. An
increasing number of workplaces are also becoming standardized internationally.
An example is the success of the ISO 9000 series of quality management
standards issued by the International
Organization for Standardization. These standards facilitate international
outsourcing of labor and greater consistency of production. Noteworthy is that information
technology facilitates the implementation of these standards through Computer
Business Systems (Head 2014) and that human competition as well as machine
competition is driving workers to deliver ever-greater consistencies (they become,
in a manner of speaking, robotized). It is easy to appreciate that the
creation of a workplace quality system, once it exists, will facilitate the
development of machine labor fitted exactly into what the performance
measurement system requires. These socio-economic innovations, thus, facilitate
engineering for functional equivalence.
3. Cultural
norms are also trending in the direction of greater standardization. In the
public sector, New Public Management is characterized by a greater emphasis on
performance measurements, often justified by the demand for greater accountability.
As in the case of education, many performance indicators are simplistic and
provide an incentive to optimize behavior toward a narrow standard. Another
cultural trend is the increasing acceptance of surveillance technology, in
concert with the ubiquity of smart phones, both of which can easily monitor workplace
behavior that is not by the book. Finally, cultural norms are trending toward
speech codes, which have paved the way to the regulation of acceptable speech
in some workplaces. Soon
we may even have functionally equivalent robotic politicians whose answers to
all issues are scripted to respond to the attitudes of their partys base
constituents. Arguably we already have human robots running for office. We believe that the strong self-censorship of speech to
conform to company or cultural norms is relatively recent and, like workplace
standardization, facilitates engineering for functional equivalence.
These
examples show that education, workplace behavior, and even speech are in some
contexts becoming more automated and scripted. We do not deny the strong
justifications for these socio-economic innovations, such as efficiency gains
or accountability standards. We are, however, concerned that they reduce or
even erode creativity and personal judgment while facilitating technological
displacement at a grand scale. If creativity and opinion are becoming such a
threat to job security, and if strict self-censorship becomes the norm, then we
not only lose our humanistic ideals, but will likely also underuse the full
potential and breadth of human abilities in the workplace. However, it is
possible, that the new work functions created by novel technologies will make
up, or even over-compensate, for this potential void. Our argument is not about
numbers, but about methodological perspective.
Convergence and research questions
The
integrated consideration of, on the one hand, the capacity of machines to do
human labor of all stripes (their humanization) and, on the other hand, the
socio-economic drivers of increasingly standardized workers and workplaces (their
robotization) will lead to more inclusive debates, more credible scenarios, and
better justified policy responses. Specifically, we argue that this parallel focus
improves the understanding of the drivers of technological unemployment and also
opens the door to analyze co-evolutionary effects such as feedback loops. Instances
of the crossover of the functions of robotized humans with those of
humanized robots are already evident in your local coffee shop. The most
productive establishments feature employees who behave like mobile animatronics
that speak pre-recorded sentences, while their increasingly competent and
compatible machines produce the goods that are then delivered with a mandatory
human smile. Creativity and personal judgment are not forbidden, but are also
not encouraged.
Note the
precedent for a parallel analysis in the methodological development of the
scenarios for climate (Moss
et al. 2010) and biodiversity (Pereira et al. 2010). In these
contexts, too, the focus was initially directed at physical attributes but, as
the fields matured, the approach changed from sequential approaches to the integrated
assessment of both physical and socio-economic components. We should learn from
these analogous insights and better integrate The Two Cultures (a term coined
by Snow 1961). At the least we need to integrate a scientific analysis with a
sociotechnical one.
The adoption
of an integrative perspective has practical consequences for research and
policy agendas. Five key research questions will help illustrate this point.
They each would benefit from an integrated assessment model that includes drivers
of and responses to both technological and socio-economic innovation:
1. Which jobs are most vulnerable to displacement
in the near future? Technological unemployment will vary from work context
to work context as noted in the much-cited Oxford Martin School paper by Frey
and Osborne (2013). The authors
estimate that 47 per cent of human jobs in the U.S. are at high risk of computerization.
In their analysis, telemarketers and accountants are extremely vulnerable while
athletic trainers and clergy remain resilient to the risk of technological
unemployment. We should also note that markets rapidly adapt, that human abilities
and demographics (including life expectancies) are changing, and that it is
next to impossible to reliably forecast future job functions and numbers.
However, forecasting will and should continue, and the analysis of which job
functions will become vulnerable to technological unemployment is more complete
and robust if both predicted technological capacities and workplace
standardizations are included. With such an analysis the percentage of jobs
subject to computerization may prove to be significantly higher or lower than
Frey and Osbornes estimate, and actual computerization might proceed very
rapidly or quite slowly.
2. What are the interactions (such as positive
feedback loops) between technological and socio-economic innovation processes?
Ongoing automation and computer support often render associated human jobs more
standardized and thus displaceable. For example, employee interaction with
software often reduces the elbowroom for judgment and creativity because the
interfaces only permit a narrow set of actions. The software that makes it
possible to outsource a call-center to a country with lower wages also renders
the behavior of the workers more robotized because they spend a lot of time
reading English sentences from their computer screens. The technological and
socio-economic innovations that enable the replacement of domestic workers with
teleworkers (and, subsequently, the replacement of those teleworkers with
machines) represents a complex technological and
social feedback loop.
3. Are technological
acceleration, the great decoupling, and
jobless growth inevitable? Brynjolfsson and
McAfee (2011, 2014) have suggested that
we are facing the great decoupling – a period when labor productivity
gains are no longer coupled with gains in employment or family income. They
recently authored an Open Letter on the
Digital Economy in the MIT Technology
Review (Brynjolfsson et al. 2015) that included a
set of policy recommendations. We agree with the policy recommendations, but we
argue for the inclusion of a parallel analysis of the convergence of
technological and socio-economic innovation.
4. Where should the attention for policy debate
be directed and where is governance action most urgent, important and feasible?
The tantalizing technological possibilities of artificial intelligence and
robotics can easily become a distraction from the important analysis and policy
debate over how we should design education and workplace practices and over how
we should evaluate our cultural standards. In the broader field of technology
governance, ideologies clash between those who want to strictly control emerging
technologies and those who point to the inevitability of technological progress
in a globalized world. In the context at hand, the latter position is bolstered
by the belief that all complex problems will become computable – the
position of Technological Solutionism (Morozov 2013). Assuming that education, workplace quality, and
culture should matter to everyone, parallel analysis of socio-economic
innovation will lessen debate and improve the discussion between ideological
camps, lessen techno-fatalistic views, and broaden the toolkit of possible
mitigating and adaptive policy options.
5. How should education and workplace practices
adapt to the forces of technological innovation and the trend toward
standardization of human abilities? Lets not neglect what is of direct
interest to academics: science policies and university curriculum. Taylorism (Taylor 1911) and performance measurement are now
an integral part of funding allocations and teaching evaluation. The
universities are also very much affected by the uncertain potential of massive
open online courses (MOOCs) and new expectations from students who have not
experienced the world before the Internet. Both technological and social innovations
influence university practices including education, funding, hiring, and
promotion, and the convergence of the two trends should be researched.
From functional equivalence to functional singularity
The
introduction of meaningful labels can provide a useful service to the research
community. The issue we highlighted above is the convergence of two separate
trends, both fostering the functional equivalence of the current requirements
at the workplace and the increasing capacities of software and robotic tools.
Thinking ahead, one can take this point further. We really cannot predict if
the techno-optimists or the neo-Luddites are more credible forecasters, and we remain
skeptical of both. Nevertheless, we have argued that the dual focus on the
technological and the cultural is better than a singular focus on either the technological
or the cultural in isolation.
Conversely,
we see Kurzweils technological singularity
and Bostroms superintelligence
as the result of a relatively singular focus on the technologically possible
(in the context of exponential growth). A dual focus would suggest an
additional label for an event that resembles an event horizon – a future time
characterized by exponential growth in the displacement of current workplace
functions by machines coupled with the increasing inability to forecast and
agree on policy directions. Following the focus on functional equivalence, this
term would be a functional singularity.
A focus on functional equivalence and a potential functional singularity are of
greater policy relevance than research focused on the
technological singularity or superintelligence. Indeed, societal disruption from
a functional singularity, necessitating policy responses, will occur long
before the advent of superintelligence and may well dictate future developments
in artificial intelligence. While it provides a useful label, however, the idea
of a functional singularity lacks analytical specificity and its occurrence may
be recognized only in retrospect.
Our goal is
to direct attention to factors leading to the functional equivalencies of labor
performed by humans and machines, which will pay off in forestalling a functional
singularity and mitigating its socially destabilizing impact.
Note
1. A term coined by John Maynard Keynes in 1930 in
the wake of the Great Depression (Keynes 1963).
Acknowledgments
We
acknowledge the important contributions of the discussion group at the Yale Interdisciplinary Center for Bioethics, during
the December 3, 2014 meeting. We are also grateful for the helpful commentaries
by Alin Charrire, two anonymous reviewers, and the editor-in-chief
of JET. All errors or misconceptions remain our own.
References
Bostrom, N. 2014. Superintelligence: Paths, dangers, strategies.
Oxford: Oxford University Press.
Brynjolfsson, E., and A. McAfee. 2011. Race against the machine. Lexington, MA:
Digital Frontier Press.
Brynjolfsson, E., and A. McAfee. 2014. The second machine age. New York: W.W.
Norton & Co.
Brynjolfsson et al.
2015. Open letter on the digital economy. MIT Technology Review, June 4.
Available online at http://www.technologyreview.com/view/538091/open-letter-on-the-digital-economy/.
Frey, C. and M. Osborne. 2013. The future
of employment: How susceptible are jobs to computerisation?
Oxford: Oxford Martin School Working Paper.
Head, S. 2014. Mindless:
Why smarter machines are making dumber humans. New York: Basic Books.
Keynes, J.M. 1963. Economic
possibilities for our grandchildren. In J.M. Keynes, Essays in persuasion, 358–73. New York: W.W.
Norton & Co.
Kurzweil, R. 2005. The
Singularity is near: When humans transcend biology. New York: Viking.
Morozov, E. 2013. To save
everything, click here: The folly of technological solutionism. New York: Public Affairs.
Moss,
R.H., et al. 2010. The
next generation of scenarios for climate change research and assessment. Nature 463: 747–56.
Pereira, H.M. et al. 2010. Scenarios for global biodiversity in
the 21st century. Science 330: 1496–1501.
Snow, C.P. 1961. The
two cultures and the scientific revolution: The Rede
Lecture 1959. New York: Cambridge
University Press.
Taylor, F.W.
1911. The principles of
scientific management. New York and London: Harper & Brothers.
Vinge, V. 1993. Presentation at the VISION-21 Symposium March 30–31, 1993. Symposium organized by
NASA Lewis Research Center and the Ohio Aerospace Institute. Available online
at