|Publisher:||Hoover Institution Press|
|Sold by:||Barnes & Noble|
|File size:||4 MB|
About the Author
Read an Excerpt
TECHNOLOGICAL CHANGE AND THE WORKPLACE
Part I. Advancing Technology and Its Impact
The computation and communication technologies that enabled the information revolution continue to advance rapidly, promising benefits and efficiencies but also raising questions concerning displacement of workers, inequality, and privacy. This chapter outlines further advances in automation and artificial intelligence (AI) and in the scale of collection and exploitation of data that can reasonably be foreseen over the next ten to fifteen years. These technologies are in widespread use today. But impressive new applications of artificial intelligence (such as self-driving vehicles) and dramatic increases in the amount of data collected by the "internet of things" (IoT) and stored in the cloud are on the near horizon. There is potential for further advances in health, safety, and productivity — and also for further disruption of working and personal lives.
Recent advances in software and hardware, combined with the availability of large sets of digital data, have enabled the development of machines with the ability to sense their environment, learn by trial and error, solve problems, and take action. It is no longer correct to say that machines only do what they are programmed to do. Machines can now be trained to learn from examination of large amounts of data. Machine learning is the aspect of artificial intelligence responsible for much of the disruption of the workplace.
The twenty-first century has seen unanticipated progress in machine learning that has enabled dramatic advances in practical applications such as speech recognition and language translation. In the near future, we will see autonomous vehicles, better diagnostics for the sick, and better prevention strategies for the healthy. The combination of artificial intelligence and advancements in other technologies such as robotics and 3D printing ensures that the rate of change in many industries will continue to be swift, with accompanying social and economic consequences. This chapter begins with a summary of what can now be expected in certain sectors.
Advances in sensors and machine learning have accelerated the development of self-driving vehicles, which are likely to become widely available in the next few years. Autonomous long-haul trucks are expected to be introduced soon; autonomous cars and trucks for city use will come later, as the technology is developed for safely navigating the more complex and less predictable urban environment. Self-driving cars promise to be substantially safer and could transform commuting into an opportunity for constructive activity. Self-driving trucks and autonomous robots will reduce transport and delivery costs. Autonomous vehicles and related transportation services could reduce incentives to own cars, especially in cities, and encourage new forms of public transportation based on smaller vehicles that transport people on demand from point to point. These safety, convenience, and efficiency benefits will come at a cost to those currently driving trucks, buses, taxis, and ride-sharing services.
Artificial intelligence technologies have the potential to improve health outcomes and quality of life through better clinical decisions, better monitoring and coaching of patients, and prevention of disease through early identification of possible health risks. Machines could learn which practices and procedures lead to the best outcomes by analyzing vast amounts of data collected in electronic medical records of millions of patients. They could also identify unintended negative effects of procedures and drugs. Machines trained by correlating electronic medical images with data on patient outcomes will enhance the accuracy of interpretation of medical images. The combination of human physicians and machine intelligence will enhance the accuracy of diagnoses of problems and recommendations for therapy and further tests. Additional sources of personal health information from personal fitness devices and social media, for example, and information on individual genomes will support more personalized diagnosis and treatment, along with an emphasis on prevention rather than cure.
We will continue to want a human physician to evaluate the output of machine intelligence in making clinical decisions and recommendations for treatment. The doctor would convey the outcome to the patient and help the patient understand and accept it. The role of nurses, who interact with patients, communicate with them, and make them comfortable, is less susceptible to disruption by automation in the near term.
Machines will not replace teachers. But over the next ten to fifteen years, the use of systems based on AI technologies in the classroom and in the home will expand substantially. Interactive machine tutors are being developed to help educate students and train workers in a variety of subjects, providing personalized coaching and support and monitoring progress. Online courses have promise for providing personalized interaction with students at all levels on a large scale, exposing students to courses that have proved successful and allowing them to work at their own pace using educational techniques that work for them. Current experimentation and online courses are producing feedback data that will allow the developers of educational systems to learn what works and to improve, including finding the best mix of machines and teaching assistants to provide support. An important application of online courses and intelligent tutoring systems is likely to be the retraining of workers and lifelong learning. Large-scale, personalized adult education and training can be part of the solution to the disruption of the workplace by changes brought about by advancing technologies.
For decades, manufacturing in many industries has moved offshore, driven by low labor costs, efficient freight systems, and trade agreements. Disruption of industries and loss of skilled, well-paying jobs have contributed substantially to the problems of governance. But globalization may have peaked; trade as a fraction of GDP is now declining. The combination of artificial intelligence, robotics, and 3D printing promises further fundamental changes in the way things are made, leading to production of goods, services, energy, and food close to the consumer. The falling costs and increasing capabilities of 3D printing, with its inherent ability to customize each item at no additional cost, will allow production of consumer goods and industrial products built to order for each individual customer. Hospital supplies and parts for cars, trucks, and aircraft can be produced when and where they are needed, rather than stockpiled. With AI, advanced robotics, and 3D printing technologies reducing labor costs and increasing quality and customization, the advantages of manufacturing in countries with low labor costs will be reduced, while the advantages of production of made-to-order products near the customer will grow. As one line of argument goes, "With the cost of labor no longer a significant advantage, it makes little sense to manufacture components in Southeast Asia, assemble them in China, and then ship them to the rest of the world when the same item can either be manufactured by robots or printed where it will be used." On shoring is the likely trend for the next ten to fifteen years, but the associated new jobs will be different from those that were lost to offshoring.
Employment and the Workplace
The list of industries where automation and artificial intelligence will change the workplace in fundamental ways is long and diverse, including:
Health care (automated diagnostics, image interpretation, robotic surgery, patient monitoring, risk assessment, and disease prevention)
Transportation (autonomous cars, trucks, and taxis; monitoring of aircraft engines)
Law (pretrial discovery)
Call centers (voice recognition and responses)
Education (interactive tutors, online courses)
Software (machines that write and debug software)
Logistics (automated warehouses, sensors for supply chain management)
Agriculture (autonomous vehicles, crop and animal monitoring, local indoor farms)
Elder care (automated transportation, monitoring, personalized health management, service robots)
Manufacturing (automated production lines of all kinds)
In all of these endeavors, large numbers of workers now onshore and offshore will be displaced by more efficient machines. In contrast to the nineteenth-century mechanization of manual tasks and the twentieth-century offshoring of routine tasks, twenty-first-century machines will be moving into a wide range of cognitive tasks that until now have been reserved for humans, including professional services. One result will be less expensive, better quality, and more customized goods and services, plus an improved standard of living. Another result will be loss of employment for workers in a broad range of skill and income levels.
A study by Frey and Osborne suggests that 47 percent of workers are in occupations with a high probability of displacement by automation. They conclude:
While computerization has been historically confined to routine tasks involving explicit rule-based activities, algorithms for big data are now rapidly entering domains reliant upon pattern recognition and can readily substitute for labor in a wide range of non-routine cognitive tasks. In addition, advanced robots are gaining enhanced senses and dexterity, allowing them to perform a broader scope of manual tasks. This is likely to change the nature of work across industries and occupations.
They find workers in service industries to be highly susceptible to automation, as well as workers in transportation and logistics, office and administrative support, and production. Machine learning is even assuming some of the tasks of software engineers.
As in the past, new industries and new jobs will be created. The number and nature of these new jobs are difficult to foresee. Certain tasks will become more important, creating opportunities for expansion, and new categories of employment could be created. The net effect on the total number of jobs is difficult to predict. What is clear is that a substantial fraction of the workforce may lose well-paying "cognitive" jobs to automation, perhaps more over time than the well-paying factory jobs lost to globalization.
The recent evolution of chess may provide a hint about the future. After a long period of development of hardware and software, a computer defeated the best human chess player, Gary Kasparov, in a well-known match in 1997. Today, a laptop computer with off-the-shelf software can play as well. Now there is a new game called free-style chess, in which a human player can draw upon machine support. For successful players, the human provides the strategy and uses a variety of machines to explore tactics and consequences of potential moves. The human-plus-machine combination is widely considered to play at a higher level than either humans or machines, and a human-plus-machine combination can defeat any human or any machine.
This lesson may be broadly applicable, suggesting that the best results will come from humans supported by intelligent machines — a combination of a doctor and a machine, a teacher and a machine, and so on. In the workplace of the near future, humans will do jobs (or portions of jobs) that machines do not do well, and work with machines in areas where machines have advantages. Davenport and Kirby describe future workplaces where machines and humans work together, the machines doing the computational work they do best, augmenting the humans who see the big picture and have interpersonal skills. One example would be insurance underwriting, where machines make detailed risk assessments and premium calculations for each application for insurance, and humans address exceptional cases, manage the company's overall risk profile, and communicate with individuals whose applications were denied.
It is not clear whether the number of jobs created and jobs retained in modified form will match the number of jobs lost to artificial intelligence, automation, and robotics. Those who have studied this question are closely divided. Historically, over the two hundred years since the Luddite rebellion, gains in productivity have, over time, led to new jobs in new industries. That could continue. Or this time could be different.
What is clear is that in the near term, large-scale disruption of the workplace will continue and probably accelerate. In contrast to the twentieth century, where job loss was concentrated on middle-income, mid-level skill occupations, the current advances in artificial intelligence and automation in the twenty-first century will affect workers at all skill and income levels (high and low as well as middle), including some well-paying cognitive jobs.
For the country as a whole, the advance of technologies that exploit artificial intelligence and automation will likely continue to increase national wealth and income, but these benefits will be distributed unevenly. Some workers performing routine tasks at all skill and income levels will be adversely affected. But low- and middle-income workers without a college education will feel the most pressure on wages and employment. This is a continuation of a well-established trend. Median household income has not risen significantly since 1999, even as GDP has grown 38 percent. All of the gains in income have gone to the upper end (table 1.1).
Other metrics — including median income per capita, total wealth, and life expectancy — also demonstrate growing inequality. The spread of automation contributes to this growing inequality in wealth and income. It is easy to see how this happens. Brynjolfsson and McAfee give the example of TurboTax, a provider of tax preparation software. Many customers find it cheaper, quicker, and more accurate than having tax preparers produce their returns. TurboTax has therefore created a great deal of value for its users. The small cadre who created TurboTax has benefited handsomely. But a much larger number who earned their living as tax preparers now find their jobs and income threatened. Replicating this example throughout the economy, new technology adds to the nation's GDP and standard of living, concentrates new wealth in a small number of entrepreneurs and their skilled employees, and threatens the livelihoods of a larger number of displaced workers.
The failure of median incomes to rise as GDP grows contributes to the widespread perception among low- and middle-income citizens that the present economy does not work for them.
The recent and surprising advances in artificial intelligence for practical applications are based on the application of innovative processing power and software to very large sets of digital data. For example, machines have recently become quite proficient at translation by comparing vast amounts of digital text collected from many sources. These translation systems are not programmed on grammar, syntax, or spelling; they are trained by examining very large amounts of text in various languages, enabling them to learn how a phrase in one language correlates with the corresponding phrase in another. Similarly, machines with no knowledge of biology are trained to interpret medical images by finding correlations between digital images and medical outcomes, based on lots of data from very large numbers of cases.
"Big data" is quantitatively very big. Google processes over 3.5 billion search queries each day and saves information on each one. Facebook uploads more than 300 million photographs each day. This enormous growth in scale results in a qualitative change as well. Collection of so much information — all or nearly all of the information on a subject — has facilitated a transformation away from drawing statistical inferences from a sample of data to drawing deeper, more detailed, and more reliable conclusions by examining all the data, not just a sample. Given a very large data set, machines can learn to find patterns and correlations and make reliable predictions without considering the physical or biological processes involved. As the data set continues to grow, the predictions get better. The role of vast amounts of data in the success of artificial intelligence is so central that the entire field is coming to be called "data science," two key components of which are data collection and machine learning to draw conclusions from the data and make predictions.(Continues…)
Excerpted from "Beyond Disruption"
Copyright © 2018 Board of Trustees of the Leland Stanford Junior University.
Excerpted by permission of Hoover Institution Press.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.
Table of Contents
Preface George P. Shultz,
Introduction Jim Hoagland,
1 Technological Change and the Workplace James Timbie,
2 Technological Change and the Fourth Industrial Revolution T. X. Hammes,
3 Governance and Security through Stability Raymond Jeanloz and Christopher Stubbs,
4 Governance in Defense of the Global Operating System James O. Ellis, Jr.,
5 Technological Change and Global Biological Disequilibrium Lucy Shapiro and Harley McAdams,
Reflections On Disruption John B. Taylor,
6 Governance and Order in a Networked World Niall Ferguson,
7 Governance from a Contemporary European Perspective William Drozdiak,
Reflections On Disruption Nicole Perlroth,
8 Governance and the American Presidency David M. Kennedy,
9 Technological Change and Language Charles Hill,
Afterword James Timbie,